RESUME DAN RINGKASAN AMNIOSINTESIS DAN FETOSKOPI

PADA MATRIKULASI FETO MATERNAL

Oleh:

MARISASANTI PUTRI

NIM: PO 71242210024

Dosen Pembimbing : Imelda, S.SiT, M.Bmd

POLTEKKES KEMENKES JAMBI JURUSAN KEBIDANAN

PRODI PENDIDIKAN PROFESI BIDAN
2021
 RESUME JURNAL AMNIOSINTESIS

JUDUL : Penatalaksanaan dugaan primer infeksi Toxoplasma gondii pada wanita hamil
di Norwegia: dua puluh tahun pengalaman amniosentesis pada populasi
dengan prevalensi rendah

VOLUME : DOI 10.1186/s12884-017-1300-1

TAHUN : 2017

PENULIS : Findal dkk. BMC Kehamilan dan Melahirkan

A. Latar Belakang
Jurnal ini memiliki latar belakang berdasarkan Infeksi primer Toxoplasma gondii selama
kehamilan dapat mengancam janin. Wanita yang terinfeksi sebelum pembuahan tidak
mungkin menularkan parasit ke janin,. Jika serologi ibu menunjukkan kemungkinan infeksi
primer, kemudian dilakukan amniosentesis untuk analisis PCR toksoplasma dan diberikan
pengobatan antiparasit.
Namun, amniosintesis dapat membedakan antara infeksi primer dan laten yang terjadi dan
Kehilangan janin terkait prosedur setelah amniosentesis menjadi perhatian yang harus di
pertimbangkan

B. Tujuan penulisan Jurnal

Secara garis besar, jurnal ini dibuat dengan tujuan untuk menentukan apakah amniosentesis
dilakukan pada pasien yang benar dan apakah prosedur tersebut aman untuk indikasi ini.

C. Metode penelitian
Metode penelitian yang digunakan dengan menggunakan Studi retrospektif menganalisis
data dari semua kehamilan tunggal (n = 346) di Oslo University Hospital yang menjalani
amniosentesis karena dugaan infeksi toksoplasma primer ibu selama 1993-2013
D. Hasil penelitian
Dalam jurnal ini disebebutkaan bahwa hasil penelitiannya adalah 50% (173) wanita
terinfeksi sebelum kehamilan, 23% (80) kemungkinan pada kehamilan dan 27% (93) pasti
terinfeksi selama kehamilan. Wanita dengan IgM positif dan aviditas IgG rendah/titer uji
pewarna tinggi maka lima belas anak terinfeksi toksoplasma, salah satunya dengan PCR
negatif dalam cairan ketuban.
E. Kesimpulan
Kesimpulan dari jurnal yang dapat di Tarik adalah Setengah dari populasi penelitian wanita
terinfeksi sebelum kehamilan. Untuk mengurangi amniosentesis yang tidak perlu, kami
menyarankan serologi konfirmasi 3 minggu setelah hasil yang dicurigai dan menyarankan
agar serologi ditafsirkan oleh staf multidisiplin yang berdedikasi. Amniosentesis aman dan
berguna sebagai prosedur diagnostik dalam mendiagnosis infeksi toksoplasma kongenital
bila dilakukan setelah 15 GW.
 TINJAUAN TEORI AMNIOSINTESIS

A. DEFENISI
Amniocentesis adalah prosedur yang dilakukan saat kehamilan untuk memeriksa sampel
air ketuban. Prosedur ini berguna untuk mengetahui ada atau tidaknya kelainan pada janin.
Bila diperlukan, amniocentesis akan direkomendasikan kepada ibu hamil saat usia
kehamilan mencapai 15-20 minggu.
Dalam prosedur amniocentesis, dokter menggunakan jarum khusus untuk mengambil
sampel cairan amnion (air ketuban), dengan cara menusukkannya ke perut ibu sampai ke
rahim. Dokter akan memeriksa cairan yang mengandung sel-sel untuk memberi petunjuk
mengenai kondisi janin.
Sel-sel diperiksa berdasarkan ukuran dan nomor kromosom-kromosom janin, yang
menunjukkan apabila ada risiko atau gangguan yang membahayakan janin, salah satunya
untuk mengetahui sindrom Down.

B. Indikasi Amniocentesis
Dokter merekomendasikan prosedur amniocentesis ketika usia kehamilan menginjak
15-20 minggu. Hal ini dilakukan dengan tujuan:

 Mengetahui ketidaknormalan kromosom janin sebelum kelahiran. Pemeriksaan

amniocentesis dilakukan bila setelah pemeriksaan USG kehamilan dicurigai adanya
kelainan pada janin, misalnya sindrom Patau.
 Mengetahui perkembangan paru-paru janin.
 Memastikan terjadinya chorioamnionitis, yaitu infeksi bakteri pada kantung ketuban
(amnion) dan lapisan pembentuk ari-ari (chorion).
  Mengevaluasi kelainan pada janin akibat alloimmunization, yaitu kelainan akibat respons
sistem imun atau kekebalan tubuh ibu yang ikut dipindahkan kepada janin, serta
menimbulkan masalah bagi janin. Kelainan akibat alloimmunization ini adalah kelainan
akibat ketidakcocokan rhesus (inkompabilitas rhesus) atau hidrops fetalis. Bila
inkompabilitas rhesus tidak terdeteksi sejak awal bisa membahayakan kondisi janin.
 Pengobatan polihidramnion, yaitu dengan memberikan obat secara langsung ke dalam
ketuban, untuk mengurangi tekanan dalam rahim. Amniocentesis juga dapat digunakan
untuk memberikan obat secara langsung ke janin.

Kelainan pada janin lebih rentan terjadi pada ibu hamil dengan kondisi sebagai berikut:

 Berusia di atas 35 tahun.

 Memiliki riwayat keluarga atau anak yang lahir sebelumnya dengan gangguan
metabolisme atau kelainan gen, seperti sindrom Down, penyakit Tay-Sachs, atau cystic
fibrosis.

C. Peringatan Amniocentesis
Amniocentesis merupakan prosedur yang aman dilakukan. Walau begitu, tetap ada
beberapa kondisi yang menyebabkan ibu hamil perlu berhati-hati sebelum melakukan
tindakan amniocentesis. Keadaan tersebut antara lain:

 Kurangnya jumlah air ketuban (oligohidramnion).

 Kelainan posisi ari-ari.
 Memiliki alergi terhadap obat bius, bahan lateks, atau perekat.
 Sedang menggunakan obat-obatan lain, misalnya obat pengencer darah.
 Memiliki riwayat gangguan pembekuan darah.
 Perbedaan golongan darah rhesus dengan janin yang dikandung.
 Menderita hepatitis atau HIV.

D. Sebelum Amniocentesis
Tidak ada persiapan khusus sebelum menjalani amniocentesis. Ibu hamil juga tidak
perlu berpuasa sebelum dilakukan tindakan. Dalam beberapa kasus, ibu hamil dianjurkan
untuk menahan buang air kecil, dikarenakan prosedur ini lebih mudah dilakukan apabila
urine memenuhi saluran kemih. Mintalah kepada suami atau keluarga untuk mengantar dan
menemani selama prosedur.
E. Prosedur Amniocentesis
Dokter akan meminta Anda berbaring dengan nyaman di ranjang ruang periksa.
Dokter akan membantu untuk memposisikan diri Anda ke posisi litotomi yaitu posisi
berbaring terlentang, lutut dan pinggul ditekuk, serta kedua kaki akan ditopang. Saat Anda
sudah berbaring dengan nyaman, dokter akan menggunakan USG untuk memeriksa kondisi
janin, denyut jantung janin, lokasi plasenta, dan lokasi air ketuban.Dokter akan
menggunakan anestesi yang disuntikkan di sekitar perut untuk mengurangi rasa sakit.
Namun, anestesi tidak selalu digunakan dalam amniocentesis karena pengaruhnya dirasa
tidak begitu penting.
 USG juga digunakan sebagai panduan untuk menusukkan jarum ke dinding perut
hingga ujung jarum berada di pusat kantung ketuban. Dokter akan mengambil kira-kira 30
ml (sekitar 2 sendok makan) cairan. Prosedur ini berlangsung singkat, yakni sekitar 30
detik hingga beberapa menit. Apabila pengambilan jumlah cairan dirasakan cukup, dokter
akan secara hati-hati menarik jarum keluar dari perut. Setelah itu, dokter akan
mengoleskan cairan antiseptik dan menutup area tusukan di perut dengan perban.
F. Setelah Amniocentesis
Setelah amniocentesis, dokter akan memeriksa denyut jantung janin dengan alat
khusus, memastikan bahwa janin tidak mengalami stres. Apabila Anda memiliki golongan
darah rhesus negatif, dan janin dicurigai memiliki golongan darah rhesus positif, dokter
akan memberikan suntikan Rho setelah prosedur. Suntikan Rho bertujuan untuk mencegah
reaksi alloimunization terhadap janin.
Dokter akan memperbolehkan Anda untuk pulang dan menyarankan beristirahat di
rumah, serta menghindari aktivitas dengan gerakan yang berulang-ulang dan hubungan
seksual selama 1-2 hari. Sampel air ketuban akan diperiksa lebih lanjut di laboratorium dan
hasilnya bisa diperoleh dalam beberapa hari hingga satu bulan. Diskusikan bersama dokter
hasil amniocentesis yang telah dijalani.
G. Komplikasi Amniocentesis
Konsultasikan kepada dokter apabila Anda mengalami komplikasi setelah menjalani
amniocentesis. Amniocentesis dapat menyebabkan sejumlah komplikasi berupa:

 Penularan infeksi. Ibu hamil yang memiliki infeksi, seperti hepatitis atau HIV, berisiko
menginfeksi janin melalui amniocentesis.
 Kebocoran air ketuban. Walau jarang, kebocoran air ketuban bisa saja terjadi. Apabila
terjadi, kondisi ibu dan janin akan terus dipantau oleh dokter, terutama bila adanya
infeksi. Dalam kasus ini, risiko komplikasi kelahiran prematur akan meningkat, yang
umumnya disebabkan oleh sedikitnya jumlah cairan ketuban yang tersisa.
 Penelitian justru menunjukkan peluang amniocentesis untuk menyebabkan risiko
keguguran sangatlah kecil. Terjadinya keguguran akibat amniocentesis hanya 0,2-0,3
persen dari seluruh kehamilan.
 Cedera pada janin, seperti gangguan paru-paru, dislokasi pinggul, atau kaki pekuk
(clubfoot).
 RESUME JURNAL FETOSCOPI

JUDUL : Mosaik fetoskopik berbasis pembelajaran mendalam untuk perluasan bidang

pandang

VOLUME :-

TAHUN : 2020

PENULIS : Sophia Bano, dkk. Jurnal Internasional Radiologi dan Bedah Berbantuan
Komputer (2020) 15:1807–1816 https://doi.org/10.1007/s11548-020-02242-8

A. Latar Belakang
Jurnal ini memiliki latar belakang dimana Fotokoagulasi laser fetoskopik adalah prosedur
bedah invasif minimal yang digunakan untuk mengobati sindrom transfusi kembar-ke-
kembar (TTTS), yang melibatkan lokalisasi dan ablasi koneksi vaskular abnormal pada
plasenta untuk mengatur aliran darah pada kedua janin
B. Tujuan penulisan Jurnal
Secara garis besar, jurnal ini dibuat dengan tujuan untuk Mosaik fetoskopik agar dapat
membantu dalam menciptakan gambar dengan bidang pandang yang diperluas yang dapat
memudahkan dokter selama prosedur TTTS.
C. Metode penelitian
Metode penelitian yang digunakan mengusulkan kerangka kerja mosaik berbasis
pembelajaran mendalam untuk beragam video fetoskopik yang diambil dari berbagai
pengaturan seperti simulasi, hantu, ex vivo, dan lingkungan in vivo
D. Hasil penelitian
Dalam jurnal ini disebebutkaan bahwa hasil penelitiannya adalah Kami melakukan evaluasi
kuantitatif dan kualitatif pada 5 video fetoskopik beragam (2400 frame) yang menangkap
lingkungan yang berbeda. Untuk mendemonstrasikan kekokohan kerangka yang diusulkan,
perbandingan dilakukan dengan metode homografi citra dalam dan berbasis fitur yang ada.
E. Kesimpulan
Kesimpulan dari jurnal yang dapat di Tarik adalah Kerangka kerja mosaik yang diusulkan
mengungguli metode yang ada dan menghasilkan mosaik yang bermakna, sekaligus
mengurangi penyimpangan yang terakumulasi, bahkan dengan adanya tantangan visual
seperti sorotan spekular, refleksi, kurangnya tekstur, dan resolusi video rendah
 TINJAUAN TEORI FETOSKOPI
A. DEFENISI

Fetoscopy adalah tindakan memasukkan instrumen melalui abdomen ke rongga uterus

untuk inspeksi janin secara visual. Fetoscopy adalah prosedur menggunakan alat
endoskopi seperti halnya laparoscopy dan hysteroscopy, namun tindakan fetoscopy ini
dilakukan pada wanita hamil dengan tujuan untuk mengevaluasi secara langsung kondisi
janin, air ketuban, placenta dan sekaligus bisa melakukan tindakan tertentu seperti biopsi
atau laser oklusi seperti pada twin-twin tranfusion sindrome.

Gambar Fetoskopi

B. JENIS-JENIS FETOSKOPI

Ada dua jenis fetoskopi: eksternal dan endoskopi.

1. FETOSKOPI EKSTERNAL

Sebuah fetoskop eksternal menyerupai stetoskop, tetapi dengan headpiece. Hal ini
digunakan secara eksternal pada perut ibu untuk auskultasi bunyi jantung janin setelah
18 minggu kehamilan. Hal ini juga memungkinkan untuk memantau janin dan
memastikan bayi mentoleransi tenaga kerja tanpa harus dilampirkan ke monitor terus
menerus.
 2. FETOSKOPI ENDOSKOPI

Tipe kedua fetoskop adalah endoskopi serat optik. Hal ini dimasukkan ke dalam rahim
baik transabdominal (melalui perut) atau transcervically (melalui leher rahim) untuk
memvisualisasikan janin, untuk mendapatkan sampel jaringan janin, atau untuk
melakukan operasi janin.

C. Beberapa kelainan janin yang dapat diobati oleh fetoskopi adalah :

1. Hernia diafragma kongenital (CDH). Pada bayi dengan CDH, diafragma (otot tipis
yang memisahkan dada dari perut) tidak berkembang dengan baik, organ-organ perut
dapat masuk rongga dada melalui lubang (hernia) dan menyebabkan hiperplasia paru.
HDK terjadi sekitar 1/4000 kelahiran hidup. Defek diafragmatika menyebabkan
herniasi pada organ abdominal ke rongga thorax, sehingga terjadi hipoplasia pulmonal.
Sekitar 80% dari defek ini terjadi pada sisi kiri, 15% terjadi pada sisi kanan dan 5%
bilateral. Secara keseluruhan risiko mortalitas sekitar 50%.
Beberapa tahun terakhir, berbagai cara dilakukan untuk mencegah perkembangan
abnormal paru janin, termasuk dengan operasi bedah terbuka yang melibatkan
laparotomi dan histerotomi, dilanjutkan thorakotomi dan repair defek diafragma.
Tetapi hal tersebut telah ditinggalkan karena berhubungan dengan tingginya
morbiditas maternal dan tidak meningkatkan survival rate janin. Sekarang, tindakan
invasif minimal telah dikembangkan dan menggantikan operasi bedah terbuka.
Fetoskop dimasukkan ke dalam uterus, kemudian masuk kedalam mulut janin,
orofaring, dan trakhea. Sebuah balon digunakan untuk menutup trakhea dan mencegah
keluarnya sekret paru. Sehingga terjadi peningkatan tekanan dan peningkatan luas
penampang paru menghasilkan stimulasi pertumbuhan paru. Balon biasanya diinsersi
pada umur kehamilan 26 minggu dan dikeluarkan umur kehamilan 34 minggu.
2. Obstruksi saluran kemih. Uretra (tabung yang membawa urin dari kandung kemih ke
luar tubuh) dapat terjadi intra uterine growth restriction atau gagal untuk berkembang
secara normal. Ketika ini terjadi, urin dapat membuat cadangan kedalam ginjal dan
merusak jaringan atau menyebabkan kandung kemih menjadi membesar. Jumlah
 cairan ketuban juga berkurang karena urin janin komponen utama. Hipoplasia paru
biasanya menghasilkan karena paru-paru mengandalkan cairan ketuban dalam
perkembangan mereka.
Pada kondisi ini janin tidak dapat mengosongkan kandung kemih sehingga kandung
kemihnya menjadi semakin besar. Selain itu, karena cairan amnion dibentuk dari urin
janin pada pertengahan trimester kedua, kantung amnion menjadi kering. Efeknya
terjadi peningkatan dari morbiditas dan mortalitas janin. Termasuk juga terjadi
kelainan seperti hidronefrosis, displasia ginjal, dan hipoplasia pulmo.
Insidensi berdasarkan data dari Northern Region Congenital Anomaly Register
England memperlihatkan bahwa LUTO terjadi 2,2 per 10.000 kelahiran. Katup urethra
posterior terjadi 1,4 per 10.000 kelahiran, atresia urethra terjadi 0,7 per 10.000
kelahiran, dan sisanya tidak teridentifikasi.
Penyebab obstruksi bermacam-macam, paling sering karena adanya katup urethra
posterior pada janin laki-laki. Pada janin wanita, tersering adalah karena atresia
urethra. Penyebab lain obstruksi antara lain ureterocele, striktur urethra atau agenesis,
kloaka persisten, dan megalourethra. Hasil pemeriksaan USG pada kelainan diatas
mungkin serupa dan biasanya sulit dibedakan hingga janin lahir.
Terapi dapat dilakukan dengan bedah terbuka atau dengan fetoskopi dilakukan
Vesicoamniotic Shunt
3. Twin/kembar sindrom transfusi (TTTS). Dalam beberapa kehamilan kembar, dua janin
akan berbagi plasenta (disebut kehamilan monokorionik). TTTS terjadi pada sekitar
15% dari kembar ini ketika volume darah antara janin adalah tidak seimbang,
menyebabkan volume darah rendah yang tidak normal dalam kembar donor dan
volume darah abnormal tinggi dikembar penerima. Sering ada perbedaan besar dalam
ukuran antara kembar. Sekitar 70-80% dari janin menderita TTTS akan mati tanpa
intervensi. Mortalitas mencapai 90% dan sekitar 30% yang bertahan memperlihatkan
kelainan perkembangan saraf. Terapi TTTS dilakukan dengan amnioreduksi atau laser
ablasi fetoskopik.
4. Acardiac kembar. Kondisi ini juga terjadi pada kehamilan monokorionik, tapi satu
kembar mengembang normal sementara yang lain berkembang tanpa jantung. The
acardiac kembar menerima suplai darah dari kembar normal, yang jantungnya
 sekarang harus memopa lebih keras melalui kedua janin. Sekitar 50-75% dari kembar
acardiac akan mati sebagai hasilnya. Kembar acardiac terjadi pada 1% kehamilan
monokorionik dan satu dari 35.000 kehamilan secara keseluruhan.
Kondisi ini terjadi 1 % dari kehamilan kembar monokorion. Darah mengalir atau
dipompakan kepada kembar lainnya (kembar akardiak) dengan aliran retrograde
sehingga menyebabkan kembar resipien memperoleh darah rendah oksigen.
Prosedur tersebut salah satunya dengan fetoskopi. Terapi TRAP dengan fetoskopi
dapat berupa ligasi tali pusat (umbilical cord ligation), terapi laser pada pembuluh
darah plasenta (laser therapy of the placental vessels), oklusi tali pusat dengan laser
(laser umbilical cord occlusion).
5. Amnion adalah membran yang mengelilingi janin di dalam rahim, jika terjadi ruptur
maka helaian selaput dapat mengambang didalam kantung amnion sehingga dapat
menimbulkan ikatan pada bagian badan janin dan menyebabkan trauma pada janin, hal
tersebut disebut sebagai Amniotic band syndome.
Kelainan ini pertama kali didefinisikan oleh Montgomery tahun 1832. Terjadi 1 dari
1.200 - 15.000 kelahiran hidup. Jika tidak diterapi, jeratan helaian ini semakin erat
pada bagian badan janin, menyebabkan amputasi, deformitas berat pada ekstremitas,
jari kaki dan tangan berselaput, atau defek berat pada kraniofasial dan tulang belakang
Jika tidak diterapi, jeratan helaian ini semakin erat pada bagian badan janin,
menyebabkan amputasi, deformitas berat pada ekstremitas, jari kaki dan tangan
berselaput, atau defek berat pada kraniofasial dan tulang belakang. Harap diingat
bahwa fetoskopi adalah prosedur jarang digunakan dan untuk setiap pasien didiagnosis
dengan salah satu kondisi di atas, hanya beberapa prosedur yang akan membutuhkan
intervensi janin.

D. Risiko Pada Fetoskopi

Risiko utama dari fetoskopi yang melukai dan kehilangan janin selama prosedur. Risiko
dan manfaat dari prosedur akan dijelaskan dengan hati-hati. Jika semua berjalan dengan
baik dengan prosedur, kehamilan Anda akan dipantau dengan cermat untuk persalinan
prematur dan kelahiran prematur
Findal et al. BMC Pregnancy and Childbirth (2017) 17:127
DOI 10.1186/s12884-017-1300-1

RESEARCH ARTICLE Open Access

Management of suspected primary

Toxoplasma gondii infection in pregnant
women in Norway: twenty years of
experience of amniocentesis in a
low-prevalence population
Gry Findal1,2*, Anne Helbig2, Guttorm Haugen1,2, Pål A. Jenum1,3 and Babill Stray-Pedersen1,2

Abstract
Background: Primary infection with Toxoplasma gondii during pregnancy may pose a threat to the fetus. Women
infected prior to conception are unlikely to transmit the parasite to the fetus. If maternal serology indicates a
possible primary infection, amniocentesis for toxoplasma PCR analysis is performed and antiparasitic treatment
given. However, discriminating between primary and latent infection is challenging and unnecessary amniocenteses
may occur. Procedure-related fetal loss after amniocentesis is of concern. The aim of the present study was to determine
whether amniocentesis is performed on the correct patients and whether the procedure is safe for this indication.
Methods: Retrospective study analysing data from all singleton pregnancies (n = 346) at Oslo University Hospital
undergoing amniocentesis due to suspected maternal primary toxoplasma infection during 1993–2013. Maternal,
neonatal and infant data were obtained from clinical hospital records, laboratory records and pregnancy charts. All serum
samples were analysed at the Norwegian Institute of Public Health or at the Toxoplasma Reference Laboratory at Oslo
University Hospital. The amniocenteses were performed at Oslo University Hospital by experienced personnel. Time of
maternal infection was evaluated retrospectively based on serology results.
Results: 50% (173) of the women were infected before pregnancy, 23% (80) possibly in pregnancy and 27% (93) were
certainly infected during pregnancy. Forty-nine (14%) women seroconverted, 42 (12%) had IgG antibody increase and
255 (74%) women had IgM positivity and low IgG avidity/high dye test titre. Fifteen offspring were infected with
toxoplasma, one of them with negative PCR in the amniotic fluid. Median gestational age at amniocentesis was 16.7
gestational weeks (GWs) (Q1 = 15, Q3 = 22), with median sample volume 4 ml (Q1 = 3, Q3 = 7). Two miscarriages occurred
4 weeks after the procedure, both performed in GW 13. One of these had severe fetal toxoplasma infection.
Conclusions: Half of our study population were infected before pregnancy. In order to reduce the unnecessary
amniocenteses we advise confirmatory serology 3 weeks after a suspect result and suggest that the serology is
interpreted by dedicated multidisciplinary staff. Amniocentesis is safe and useful as a diagnostic procedure in diagnosing
congenital toxoplasma infection when performed after 15 GW.
Keywords: Toxoplasma gondii, Toxoplasma infection, Pregnancy, Antenatal, Amniocentesis, Congenital, Prenatal
diagnosis, PCR, Toxoplasma antibodies

* Correspondence: gryfi@medisin.uio.no
1
University of Oslo, Institute of Clinical Medicine, Oslo, Norway
2
Division of Gynaecology and Obstetrics, Oslo University Hospital, Oslo,
Norway
Full list of author information is available at the end of the article

© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Findal et al. BMC Pregnancy and Childbirth (2017) 17:127
Background
Infection caused by the parasite Toxoplasma gondii is
usually asymptomatic in immune-competent humans,
but primary infection during pregnancy may pose a
threat to the fetus. The risk and severity of a fetal infec-
tion depend on the gestational age (GA) at the time of
infection [1–3]. A fetal infection in the first trimester,
which is rare due to the low mother-to-child transmission
rate (<10%), often leads to serious clinical manifestations
such as fetal death, hydrocephaly and intracerebral calcifi-
cations [3, 4]. The transmission rate is higher in the third
trimester (50–70%), but the fetuses are affected less se-
verely and are likely to have a subclinical infection or
be asymptomatic at birth [1, 5, 6]. Nevertheless, up to
one-third of infected children will develop complica-
tions—most often ocular lesions (chorioretinitis)—dur-
ing the first years of life [3, 7]. It is therefore important
to detect primary T. gondii infection during pregnancy.
The immune response elicited by a primary infection
results in the parasite rapidly forming semi-dormant
intracellular pseudocysts, inflicting a latent stage [4].
Primary maternal infection is suspected when positiv-
ity for immunoglobulin M (IgM) antibodies and low im-
munoglobulin G (IgG) avidity are present [8]. Latent
maternal T. gondii infection can be detected indirectly
through the demonstration of toxoplasma-specific IgG
antibodies. However, discriminating between latent and
primary infection is a challenge unless seroconversion is
observed, since toxoplasma IgM positivity and low IgG
avidity may persist for months or even years after pri-
mary infection [5, 9–12]. When a primary toxoplasma
infection is suspected during pregnancy, the pregnant
woman should start antiparasitic treatment (spiramycin
or azithromycin). Further, amniocentesis for toxoplasma
polymerase chain reaction (PCR) analysis of the amni-
otic fluid is recommended to confirm fetal infection and
assess the need for prolonged treatment and follow-up
[4]. A positive toxoplasma PCR will lead to the treat-
ment being changed to pyrimethamine-sulphadiazine
with folic acid, which is continued until birth [13, 14].
Procedure-related complications after amniocente-
sis—the most serious of which is fetal loss—are of con-
cern [15–17]. The risk might be lower than for genetic
indications, due to the smaller volume of fluid obtained,
and the higher average GA at which the procedure is
performed.
No systematic serological screening is performed in
Norway due to the low prevalence of T. gondii in the
pregnant population. Few studies have addressed the im-
pact of diagnostic amniocentesis in similar settings. We
do not know the complication rate in this context, and
have never evaluated our diagnostic guidelines.
The present study therefore evaluated the use of am-
niocentesis in our setting with the aim of determining
Page 2 of 9
whether the procedure is performed on the correct pa-
tients and if it is safe for this indication. To address
these issues we performed a retrospective study analys-
ing data from all women with singleton pregnancy at
Oslo University Hospital undergoing amniocentesis due
to suspected maternal primary toxoplasma infection.
Methods
Study population
This retrospective study included all singleton pregnan-
cies in which amniocentesis was performed at Oslo
University Hospital as part of the diagnostic workup for
serologically suspected primary T. gondii infection (n =
346). The inclusion period was from 1 September 1992
to 31 December 2013. The included women were
mainly referred from primary health-care centres in the
South-East Health Region, which constitutes about 50%
of the pregnant population in Norway [18]. The pa-
tients had been serologically tested because of risk fac-
tors, symptoms or (more commonly) at the initiative of
the woman or her health-care provider.
Maternal, neonatal and infant data were obtained from
clinical hospital records, laboratory records and preg-
nancy charts. Additional information was collected using
a questionnaire sent to 40 women whose information on
the date of delivery and birthweight was incomplete.
The GA was based on an ultrasound assessment per-
formed early in second trimester or, if not available, on
the first day of the last menstrual period. Maternal toxo-
plasma serology was mapped as the first and second
serological tests performed during pregnancy, and a
third serology was performed at birth. The second sero-
logical test was usually performed 3 weeks after the first
to confirm the test result and to detect immunological
changes [19]. The time of seroconversion, unknown in
most patients, was set as the GA at the first positive
toxoplasma antibody sample.
All amniocenteses were performed under ultrasound
guidance; from 1992 to 1996 by experienced obstetri-
cians, after 1996 by trained sub-specialists at the Section
of Fetal Medicine at Oslo University Hospital.
Laboratory methods
All samples were analysed at the Norwegian Institute of
Public Health (before 2002) or at the Toxoplasma Refer-
ence Laboratory at Oslo University Hospital (established
in 2002). The serum samples were analysed by indirect en-
zyme immunoassay (EIA) (Platelia Toxo IgG and Toxo
IgM, Diagnostic Pasteur/Bio-Rad, Marnes-la-Coquette,
France), microparticle enzyme immunoassay (MEIA)
(Axsym, Abbott, Wiesbaden, Germany) or chemilumin-
escent microparticle immunoassay (CMIA) (Architect,
Abbott, Wiesbaden, Germany) together with direct ag-
glutination (DA IgG) and immunosorbent agglutination
Findal et al. BMC Pregnancy and Childbirth (2017) 17:127
assays (ISAGA IgM and IgA) (Toxo-Screen DA and
Toxo ISAGA, bioMérieux, Marcy l´Etoile, France) and
dye test [20]. An antibody increase was defined as sig-
nificant if there was at least a doubling in the IgG level
expressed as IU/ml or as two-step titre increase.
Prior to 2005 the avidity method was performed as de-
scribed previously using an in-house method based on
the Platelia Toxo IgG assay [9], while a commercially
available avidity test (Platelia Toxo IgG avidity, Bio-Rad)
was used from 2005. The results obtained prior to 2005
were expressed as the percentage of antibodies resistant
to elution by urea and from 2005 as an avidity index,
with both of these measures being interpreted as low,
borderline or high, according to a previous publication
[9] or the manufacturer’s recommendation respectively.
In 1992 the B1-PCR method (PCR detecting the B1
gene of T.gondii) was introduced as a diagnostic indica-
tor of antenatal toxoplasma infection [21]. Mouse inocu-
lation was performed until 2002 as previously described
[8, 21].
The presence of maternal toxoplasma infection was
confirmed by either seroconversion of the toxoplasma
IgG during pregnancy, or by a significant increase in IgG
between the first and second toxoplasma samples [8].
The women were divided into three groups according to
their toxoplasma serological profile: (i) IgG seroconver-
sion, (ii) IgG increase or (iii) IgM positivity and low IgG
avidity (dye test titre of >300 IU/ml was used before the
IgG avidity test was established in 1996). In addition, the
women were categorized retrospectively into three
groups according to the suspected time of infection
based on maternal and neonatal serology [8]: (i) infected
before pregnancy, (ii) possibly infected during pregnancy
and (iii) certainly infected during pregnancy. The cat-
egorisation was performed independently by two experi-
enced investigators: a professor in microbiology (P.A.J.)
and a professor in obstetrics (B.S.P.), both with expertise
in toxoplasma diagnostics and infections during preg-
nancy. Consensus was reached for interpretation of the
results.
The presence of congenital infection was confirmed by
at least one of the following criteria being fulfilled: (i)
positive toxoplasma PCR in amniotic fluid or neonatal
cord blood, (ii) positive mouse inoculation of amniotic
fluid or cord blood, (iii) positivity for toxoplasma IgM or
immunoglobulin A (IgA) in postnatal serum, or (iv)
toxoplasma IgG persisting in the infant at 12 months
after birth [8].
Statistical analysis
Group comparisons were done using different bivariate
analyses depending on the type of variable and whether
the variables conformed to a normal distribution. For all
Page 3 of 9
tests, a p value <0.05 was considered to indicate a statis-
tically significant difference.
For GA at birth, miscarriages and induced abortions
before 22 GWs were excluded.
Due to variation in missing information in the differ-
ent variables, the denominators in the text may diverge
from 346.
A database was constructed in EpiInfo (version 3.5.4,
CDA, Atlanta, GA, USA) and the data were analysed
and figures created in IBM SPSS Statistics (version
20.01, IBM Corporation, New York, NY, USA).
Results
Our study included 346 women with singleton pregnan-
cies whose characteristics are given in Table 1.
During the study period we performed on average 15
amniocenteses per year due to possible T. gondii infec-
tion during pregnancy with a decreasing tendency over
time. The retrospective assessment concluded that 173
women (50.0%) were infected before pregnancy, 80
(23.1%) were possibly infected during pregnancy and 93
(26.9%) were certainly infected during pregnancy. The
serological profiles indicated that 49 (14.2%) women ser-
oconverted, 42 (12.1%) had IgG antibody increase and
255 (73.7%) women had IgM positivity and low IgG
avidity/high dye test titre (Table 2).
In total, 15 (4.3%) fetuses were infected during preg-
nancy, of which 14 had a positive toxoplasma PCR in
amniotic fluid (Table 3). In the single amniotic PCR-
negative pregnancy, maternal infection was suspected at
gestational week (GW) 17 (Table 3). Intrauterine infec-
tion was diagnosed retrospectively based on serology
and PCR results at birth and during infancy.
In the 15 pregnancies with confirmed fetal infection,
11 women had seroconverted, 2 exhibited an IgG-
antibody increase and 2 showed IgM positivity and low
IgG avidity. A range of clinical manifestations was ob-
served at prenatal ultrasound or after birth in 10 of the
15 (67.0%) infected offspring (Tables 3 and 4).
Maternal antiparasitic treatment had been given in all
cases with manifestations.
The offspring were infected in 11 of the 49 (22.4%)
mothers with seroconversion compared to 2 of the 42
(4.8%) mothers with antibody increase and 2 of the 255
(0.8%) mothers with IgM positivity and low IgG avidity.
All of the infected fetuses were in the group of “cer-
tainly infected during pregnancy”.
The median GA at the first toxoplasma antibody test
during pregnancy was 10.7 GWs (Q1 = 8.6, Q3 = 13.4),
and did not differ between the three serological groups
(p = 0.07). At the second serologic test and at amniocen-
tesis, the GA was significantly higher for those with
seroconversion (p = 0.001) (Fig. 1). Seroconversion was
detected at a median of 27.7 GWs (Q1 = 22, Q3 = 33.9).
Findal et al. BMC Pregnancy and Childbirth (2017) 17:127 Page 4 of 9

Table 1 Characteristics of 346 women in South-East Norway undergoing amniocentesis for suspected primary Toxoplasma gondii infection
All cases Infected before Possibly infected Certainly infected
n = 346 pregnancy during pregnancy during pregnancy
n = 173 n = 80 n = 93
Maternal age in years; mean (SD) 29.6 (5.3) 29.8 (5.3) 29.02 (5.1) 29.6 (5.4)
Parity (n & %)
P0 166 49.7 82 48.8 42 52.5 42 48.8
P≥1 168 50.3 86 51.2 38 47.5 44 51.2
Missing information 12 5 7
Nationality (n & %)
Norwegian 281 81.2 143 82.7 63 78.8 75 80.6
Northern Europe/Northern America 12 3.5 6 3.5 2 2.5 4 4.3
Other 53 15.3 24 13.8 15 18.8 14 15.1

Amniocenteses was performed at a median of 16.7
GWs (Q1 = 15, Q3 = 22), with a median sample volume
of 4 ml (Q1 = 3, Q3 = 7). All patients had a detailed fetal
ultrasound examination at the time of amniocentesis,
and 77.6% (242/311) received a follow-up ultrasound
examination in third trimester. Ultrasound abnormalities
were identified in 25 fetuses (7.2%), and 6 of these
fetuses were infected. The ultrasound findings and preg-
nancy outcomes are presented in Tables 3 and 4.
Mean GA at birth in the uninfected fetuses was 40.0
GWs (SD 2.0) compared to 38.3 GWs (SD 3.6) in the in-
fected fetuses, p = 0.003 (95% CI 0.6, 2.8).
Almost all (98.4%, 302/307) of the women were
treated with antiparasitic drugs, for a median of 21 days
(Q1 = 21, Q3 = 28), most commonly with azithromycin
(86%) or spiramycin. In 29 women the treatment was
changed to the pyrimethamine-sulphadiazine regime
post-amniocentesis.
Two miscarriages occurred: one with a severely in-
fected fetus and the other with a non-infected fetus,
both 4 weeks after the procedure (Table 4). Amniocen-
tesis (obtaining 13 ml of normal-coloured fluid) was per-
formed at GW 13.7 in the first case; the woman had had
one previous miscarriage. In the other case the amnio-
centesis was performed at GW 13.3, obtaining 5 ml of
normal-coloured fluid in a woman with no previous mis-
carriages. In addition, one intrauterine fetal death was
recorded at GW 28. This fetal death was detected at the
scheduled amniocentesis. The infection most likely oc-
curred prior to conception.
There were three pregnancy terminations with the fol-
lowing characteristics (Table 4): (i) positive PCR in the
amniotic fluid, pathological ultrasound findings and in-
fection confirmed by autopsy;
(ii) terminated pregnancy due to chromosomal aberra-
tion (and negative toxoplasma PCR); and (iii) pregnancy
terminated at GW 15 upon patient request because of
psychological distress due to the possibility of toxo-
plasma infection (negative PCR).
Serology results were recorded in 82% of the children
at birth and in 42% during the first year of life.
Discussion
We aimed to determine retrospectively whether we had
performed amniocenteses in the correct patients. We
observed that maternal infection occurred prior to con-
ception in 50% of the cases. Primary maternal infection
during pregnancy was identified in 26.9% of the women,
and infection during pregnancy could not be ruled out
in 23.1%. For most of the infected fetuses, the maternal
infection was detected in late second trimester, whereas
most cases of infection prior to conception were tested
late in the first trimester.
One possible explanation for the high proportion of
preconceptionally infected women undergoing amnio-
centesis is misinterpretation of the serological tests. A

Table 2 Retrospective assessed time of maternal toxoplasma infection according to maternal serologic group. Number of infected
fetuses in [ ]
Seroconversion IgG increase IgM positivity and n (%)
low IgG avidity
Infected before pregnancy - - 173 173 (50.0)
Possibly infected during pregnancy - - 80 80 (23.1)
Certainly infected during pregnancy 49 [11] 42 [2] 2 [2] 93 (26.9) [15]
Total n (%) 49 (14.2) 42 (12.1) 255 (73.7) 346 (100)
Findal et al. BMC Pregnancy and Childbirth (2017) 17:127 Page 5 of 9

Table 3 Neonatal outcome of 15 fetuses infected with Toxoplasma gondii, in relation to maternal serologic group
Year of Serologic Time of Gestational PCR in Antiparasitic Ultrasound Gestational Birth Findings after birth
AC group detected age at AC amniotic fluid treatment findings age at birth weight
infection
Trimester Weeks Drugs Weeks Gram
(weeks)
1994 Sc 3 (29.1) 31.0 + Sp, PS – 39.0 3830 I.c. calcifications
1995 Sc 2 (19.4) 29.4 + PS Small BPD 30.3 1500 I.c. calcifications, hydrocephaly,
chorioretinitis
1995 Sc 2 (26.6) 35.6 + Az, PS – 40.9 3700 Chorioretinitis
1995 Sc 2 (27.0) 29.3 + ? ? 31.4 ?
1995 Sc 3 (33.9) 36.3 + PS – 31.4 No findings
1995 Sc 3 (36.3) 37.7 + Az – 38.9 3880 I.c. calcifications, parasites
in placenta and CSF
1999 Sc 3 (28.0) 30.4 + Sp, PS I.c. calcifications, 41.9 3450 Chorioretinitis,
splenomegaly leptomeningeal changes on CT
2000 Sc 3 (34.0) 35.1 + Sp, PS Splenomegaly, 37.4 3268 I.c calcifications, operated
i.c. calcifications, because of atrial septal defect
renal pyelectasy
2000 Sc 3 (35.3) 37.0 + PS I.c. calcifications, 42.1 3510 No findings
splenomegaly
2002 Sc 2 (25.9) 28.0 + Az, Sp, PS – 39.9 3650 No findings
2005 Sc 3 (35.3) 37.7 + Az, Sp, PS I.c. calcifications, 38.3 3335 Jaundice, reduced vision one
splenomegaly eye
1997 Ti 2 (14.7) 17.0 + Sp – 17.7 204 Termination. Parasites at autopsy;
granulomas, paraventricular
micronecrosis and T.gondii cysts
in CNS. Lymphoid cells and
macrocyts in lungs, necrosis
and cysts in placenta
1999 Ti 1 (8.6) 13.7 + Az Enlarged nuchal 15.4 44 Planned termination but
translucency spontaneous fetal death.
Parasites at autopsy found
in CNS, lungs, liver and placenta.
T.gondii cysts in lungs
and placenta
1993 IgM+/ 1 (11.7) 16.4 + Sp – 40.6 3695 No findings
IgGav
2001 IgM+/ 2 (17.6) 20.4 – Az – 40.0 IgM, PCR of placenta and
IgGav umbilical cord blood positive
at birth, IgA increase after
6 months and IgG increase
first year of life
Ac amniocentesis, Az azitromycin, BPD biparietal diametre, CNS central nervous systeme, CSF cerebro spinal fluid, I.c. Intra cerebral, PCR polymerase chain reaction,
PS pyrimethamine-sulpha, Sc seroconversion, Sp spiramycin, Ti Titre/antibody increase, IgM+/IgGav IgM positie with low IgG avidity, ? missing information

possible solution to avoid amniocentesis in the group
with preconceptional infection is to perform an add-
itional serological test 3–4 weeks after the second test if
the test result is indeterminate. In other words, amnio-
centesis should only be performed in those with increas-
ing IgG antibody or IgG avidity levels, since this
indicates an ongoing immunological process [12].
The cut-off values for low IgG avidity in our study
were in accordance with the manufacturer’s recommen-
dations and similar to values used in previous studies
[19, 22, 23]. However, the cut-off values may have been
too high, resulting in unnecessary amniocenteses in
women with latent infection. Lowering of the IgG avidity
cut-off value could contribute to fewer unnecessary am-
niocenteses but increase the risk of underdiagnosing in-
fected women.
Interpreting toxoplasma serology may be difficult, as
illustrated by 23.1% of the women in our study retro-
spectively being categorised as possibly infected during
pregnancy. Amniocentesis in women with latent infec-
tion is probably not always avoidable and occurs even in
countries with antenatal screening programs [24]. The
toxoplasma serology should be interpreted by dedicated
staff, preferably a cooperation between obstetricians and
Findal et al. BMC Pregnancy and Childbirth (2017) 17:127 Page 6 of 9

Table 4 Ultrasound findings and outcome in 346 offspring with Congenital infection was confirmed in 4.3% of the 346
suspected maternal Toxoplasma gondii infection offspring, which is a lower rate than found in larger
Non-infected Infected studies performed in France and Austria [3, 24]. In the
offspring offspring French study, 24.7% of the fetuses were infected with the
n = 331 n = 15
parasite, but maternal infection was confirmed as an in-
Ultrasound abnormalities 19 6
crease in IgG or seroconversion prior to amniocentesis.
Intracerebral calcifications 3 In the Austrian study based on the Austrian Toxoplas-
Small biparietal diameter 3 1 mosis register, 11.8% of the fetuses were infected. How-
Increased nuchal translucency 1 ever, 45.4% of the women were considered to be infected
Ventriculomegaly, holoprosencephaly, 1 prior to pregnancy and were excluded from the analyses.
cleft lip/palate It is notable that the results of these two studies differed
Renal pyelectasy, splenomegaly, 1 markedly despite the application of screening programs.
intracerebral calcifications If we apply similar criteria as the two studies, i.e. includ-
Chromosomal markers: 6 ing only seroconversion and antibody increase, the infec-
Absent nasal bone, choroid tion rate was 14.3% (13 out of 91).
plexus cysts, single umbilical artery
Despite only 14 having positive PCR in the amniotic
Asymmetric growth, growth restriction 5 fluid antenataly, 29 women received pyrimethamine-
Renal pyelectasy 2 sulphadiazine, most of them exhibiting seroconversion
Oligohydramnios 1 or IgG antibody increase. Starting the treatment regime
Abnormalities, but details not given 1 when PCR was negative, was occasionally done during
the nineties due to an old protocol, but is as a rule not
Missing information 26
done any more.
Positive PCR 14
A cross-sectional study performed in Norway during
Miscarriage 1 1 1992–1994 included 60% of the pregnant population,
Termination of pregnancy 2 1 and investigated the diagnostics and epidemiology of T.
Stillbirth 1 gondii [1, 26]. Eight of the 15 infected children in our
study were identified during that period and the subse-
microbiologists, in order to reduce unnecessary amnio- quent 3 years. No fetal infection in women undergoing
centeses, antiparasitic treatment, postnatal follow-up antenatal amniocentesis have been detected the last
and parental worries [25]. In addition, in at-risk groups 8 years despite an increase in the number of maternal
the first toxoplasma serology should be obtained as early samples sent to the Toxoplasma Reference Laboratory
in pregnancy as possible. [27]. This finding could be explained by several factors,

Fig. 1 Gestational age at amniocentesis for suspected Toxoplasma gondii infection, according to maternal serology
Findal et al. BMC Pregnancy and Childbirth (2017) 17:127
including reduced incidence, missed diagnoses due to
misinterpretation of serological tests, scepticism in the
population or the health-care providers towards amnio-
centesis, or patients at risk not being detected [1].
The prevalence of toxoplasma IgG and incidence of
congenital toxoplasma infections varies greatly between
countries due to variations in several factors including
climate, hygiene, diet, parasite type and virulence [4, 28].
The results from studies performed outside Scandinavia
can therefore not be generalised to our population. The
prevalence of toxoplasma IgG among pregnant women
in Norway has decreased only slightly over the past
40 years: 12.6% in 1974, 10.9% in 1994 and 9.3% in 2010
[26, 29, 30]. A Norwegian study from 1994 estimated
that the incidence of maternal toxoplasma infection dur-
ing pregnancy was 1,4 per 1000 pregnancies in Norway
and 4,6 per 1000 in the capital Oslo [1]. Based on these
findings and the relatively stable toxoplasma IgG preva-
lence over 40 years, we expect the incidence of maternal
infection in Norway during the last 20 years to be close
to that in 1994, indicating that a substantial proportion
of infected children are not being detected [26]. Unlike
Austria, Norway does not have a registry for new-borns
infected with T. gondii, and the current incidence of
children with diagnosed congenital infection in our
country is unknown [24].
The amniotic fluid PCR was negative in 1 of the 15 in-
fected children. In a Norwegian study performed in
1998, Jenum et al. found that the sensitivity and specifi-
city of the toxoplasma PCR method were 59 and 94%,
respectively [21]. A recently published systematic review
and meta-analysis of the performance of PCR in amni-
otic fluid found a sensitivity of 87% and specificity of
99% when performed for up to 5 weeks after a maternal
diagnosis of seroconversion or an IgG increase [31].
Another study found that the sensitivity of PCR was sig-
nificantly higher if the infection occurred during 17–21
GWs compared to an earlier or later GA [32]. The
infected child with a negative PCR in our study was
infected before 17 GWs. The negative PCR might be ex-
plained by several factors, such as test failure, eradica-
tion of parasites by treatment before the amniocentesis
or delayed placental transmission resulting in parasite
transmission after the amniocentesis [31, 33]. This case
highlights the need to perform testing on mother and child
at birth when congenital toxoplasma infection is strongly
suspected, in addition to follow-up serology of the children
during the first year of life. In our study most of the infants
were tested at birth, but follow-up serology during the first
year of life was only performed in 42%, and decreasing dur-
ing the last decade. We therefore might have missed chil-
dren with congenital infection and a negative PCR in the
amniotic fluid. However, in most cases maternal and neo-
natal serology and PCR data were obtained at birth.
Page 7 of 9
The proportion of clinical findings in the infected off-
spring was relatively high (10 out of 15) [2]. A study
from Austria reported clinical manifestations in 10.6% of
the infected offspring (15 out of 141), all of which had
chorioretinitis and 12 had cerebral manifestations [24].
Among the group of women infected during pregnancy
in our study, the first toxoplasma serology was per-
formed in the first trimester and the second test (detect-
ing seroconversion or an antibody increase) was
performed at a median of 8 weeks later. This treatment
delay may have resulted in mother-to-child transmission
or more severe manifestations in the fetuses [2]. How-
ever, the transmission rate of 14.3% (13 out of 91) in this
study is similar or lower than other published rates [1, 3,
24]. A European multicentre study found that treatment
did not reduce mother-to-child transmission, but the in-
fected children had less severe manifestations [34].
There were few infected infants in our study and it
is relevant to ask whether patients at risk are tested
and referred. Although the prevalence of toxoplasma
IgG during pregnancy has been found to be higher
among immigrants in Norway [26, 35], the proportion
of non-Norwegians in our study (15%) was surpris-
ingly small given the relatively large immigrant popu-
lation in the South-East Health Region (13% in 1998
and 22% in 2013 of the female population of fertile
age). (https://www.ssb.no/innvbef ). This might indi-
cate that this group requests testing to a lesser extent
or that health-care workers are unaware of this risk
profile.
Probably only one of the two miscarriages may be re-
lated to the amniocentesis (0.3%, 1 out of 343); the other
is most likely due to severe fetal infection. If the 15 preg-
nancies with infected fetuses are removed from the
equation, the figures remain the same (0.3%, 1 out of
328). Amniocentesis was performed at 13 GWs in both
cases of miscarriage. At our department we have
followed the international trend away from early amnio-
centesis, and is now only perform after 14.9 GWs due to
the higher risk of fetal loss at an earlier GA [15].
Procedure-related pregnancy loss is rare, and studies
have shown that second-trimester amniocentesis is safe
with no significant risk of miscarriage [15, 36–39]. Am-
niocenteses were performed in the present study under
ultrasound guidance by trained specialists, aspirating less
amniotic fluid than for a genetic amniocentesis. Oper-
ator experience is associated with the prevalence of
procedure-related complications [15, 40]. The smaller
amount of fluid removed in our study (4 ml versus
15 ml in genetic amniocentesis) might be one reason for
the low rate of fetal loss, though the statistical power is
too low to draw a definitive conclusion. To our know-
ledge only two studies have investigated the relationship
between the amount of amniotic fluid removed and the
Findal et al. BMC Pregnancy and Childbirth (2017) 17:127
rate of fetal loss. In both studies a trend towards a lower
fetal-loss rate was found. However, in the study of Cebe-
soy et al. the finding was not significant and in the study
of Tharmaratnam and colleagues no control group was
used [41, 42]. A comparison of our results with other
studies on complications after amniocentesis is challen-
ging because we performed the procedure at a wide
range of GAs.
While the present study had a retrospective design, we
obtained complete data on laboratory results during
pregnancy as well as information on the GA at sero-
logical testing and at amniocentesis. The material was
collected over a period of 20 years, which leads to a cer-
tain degree of heterogeneity within the study population.
The diagnostic techniques evolved during the study
period, in particular with the introduction of IgG avidity,
possibly resulting in the number of amniocenteses de-
creasing over time. The management of women and
children did not change substantially during the follow-
up period, other than the introduction of the IgG avidity
test and a lower rate of toxoplasma testing during the
first year of life. Our laboratory has a national reference
function and all the serological analyses were performed
according to international recommendations. The De-
partment of Microbiology and Section of Fetal Medicine
have worked with this issue for more than 20 years. We
therefore consider our results to be relevant for other
centres managing obstetric patients and performing pre-
natal diagnostic procedures, especially in a setting where
toxoplasma screening is not part of the routine antenatal
program.
Conclusions
In our low-prevalence setting about 50% of the amnio-
centeses were performed on women with latent infection
in which antiparasitic treatment, amniocentesis and fur-
ther follow-up now seem unnecessary. To decrease the
number of unnecessary amniocenteses, toxoplasma ser-
ology should be interpreted by dedicated staff. We advise
that a second maternal serological test should be per-
formed 3 weeks after a suspect test result is obtained,
and a further sample 3–4 weeks thereafter in cases with
inconclusive serology, before performing amniocentesis.
We found that amniocentesis as a diagnostic procedure
after 15 GWs is safe and useful in diagnosing congenital
toxoplasma infection when infection is suspected by ser-
ology. Because of the possibility of a false-negative PCR
result, the necessity of serology and PCR at birth and
during first year of life needs to be emphasized.
Abbreviations
GA: Gestational age; GW: Gestational week; IgA: Immunoglobulin A;
IgG: Immunoglobulin G; IgM: Immunoglobulin M; PCR: Polymerase chain reaction
Page 8 of 9
Acknowledgements
We thank Tone Berge at the Department of Medical Microbiology, Oslo
University Hospital for excellent technical assistance.
Funding
The project received no funding.
Availability of data and materials
The dataset generated and analysed during the current study are not
publicly available due to us not having consent from the patients and
because the cases are special, rare and may be possible to identify through
the dataset. De-identified data’s are available from the corresponding author
on reasonable request.
Authors’ contributions
Conceived and designed the study: GF, BSP, PAJ, GH. Performed collection of
data: GF. Performed the experiments: BSP, PAJ, GH, AH. Analysed the data:
GF, BSP, PAJ, GH, AH. Wrote the paper: GF, AH, GH, PAJ, BSP. All authors read
and approved the final manuscript.
Competing interests
The authors declare that they have no competing interests.
Consent for publication
The paper is approved for publication by the Board of Patient Security at
Oslo University Hospital and the leaders of the Microbiology department and
Women’s and Children’s Division at Oslo university hospital (20.05.2016).
Ethics approval and consent to participate
Our project was classified as a “quality control project” by the Regional
Committee for Medicine and Health Research Ethics (project no. 2011/1310/
REK.14.09.11). In addition, the study was evaluated and approved by the
Board of Patient Safety at Oslo University Hospital (approval no. 2012/
9519.13.06.12). Because of the project classification and long inclusion period,
written consent was not needed. The patient information was de-identified
prior to analysis, anonymised after analysis and will eventually be destroyed.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
Author details
1
University of Oslo, Institute of Clinical Medicine, Oslo, Norway. 2Division of
Gynaecology and Obstetrics, Oslo University Hospital, Oslo, Norway.
3
Department of Laboratory Medicine, Section of Medical Microbiology, Vestre
Viken Hospital Trust, Drammen, Norway.
Received: 25 August 2016 Accepted: 6 April 2017
References
1. Jenum PA, Stray-Pedersen B, Melby KK, Kapperud G, Whitelaw A, Eskild A,
Eng J. Incidence of Toxoplasma gondii infection in 35,940 pregnant women
in Norway and pregnancy outcome for infected women. J Clin Microbiol.
1998;36(10):2900–6.
2. Thiebaut R, Leproust S, Chene G, Gilbert R. Effectiveness of prenatal
treatment for congenital toxoplasmosis: a meta-analysis of individual
patients’ data. Lancet. 2007;369(9556):115–22.
3. Wallon M, Peyron F, Cornu C, Vinault S, Abrahamowicz M, Kopp CB, Binquet
C. Congenital toxoplasma infection: monthly prenatal screening decreases
transmission rate and improves clinical outcome at age 3 years. Clin Infect
Dis. 2013;56(9):1223–31.
4. Remington JS MR, Wilson CB, Desmonts G. Toxoplasmosis. In: Remington JS
KJ, editor. Infectious diseases of the fetus and newborn infant. 7th ed.
Philadelphia: Saunders/Elsevier; 2011. p. 918–1041.
5. Wong SY, Remington JS. Toxoplasmosis in pregnancy. Clin Infect Dis. 1994;
18(6):853–61. quiz 862.
6. Dunn D, Wallon M, Peyron F, Petersen E, Peckham C, Gilbert R. Mother-to-
child transmission of toxoplasmosis: risk estimates for clinical counselling.
Lancet. 1999;353(9167):1829–33.
Findal et al. BMC Pregnancy and Childbirth (2017) 17:127
7. Berrebi A, Assouline C, Bessieres MH, Lathiere M, Cassaing S, Minville V,
Ayoubi JM. Long-term outcome of children with congenital toxoplasmosis.
Am J Obstet Gynecol. 2010;203(6):552.e551–556.
8. Lebech M, Joynson DH, Seitz HM, Thulliez P, Gilbert RE, Dutton GN, Ovlisen
B, Petersen E. Classification system and case definitions of Toxoplasma
gondii infection in immunocompetent pregnant women and their
congenitally infected offspring. European Research Network on Congenital
Toxoplasmosis. Eur J Clin Microbiol Infect Dis. 1996;15(10):799–805.
9. Jenum PA, Stray-Pedersen B, Gundersen AG. Improved diagnosis of primary
Toxoplasma gondii infection in early pregnancy by determination of
antitoxoplasma immunoglobulin G avidity. J Clin Microbiol. 1997;35(8):1972–7.
10. Murat JB, L’Ollivier C, Fricker Hidalgo H, Franck J, Pelloux H, Piarroux R.
Evaluation of the new Elecsys Toxo IgG avidity assay for toxoplasmosis and
new insights into the interpretation of avidity results. Clin Vaccine Immunol.
2012;19(11):1838–43.
11. Lappalainen M, Hedman K. Serodiagnosis of toxoplasmosis. The impact of
measurement of IgG avidity. Ann Ist Super Sanita. 2004;40(1):81–8.
12. Findal G, Stray-Pedersen B, Holter EK, Berge T, Jenum PA. Persistent low
Toxoplasma IgG avidity is common in pregnancy: experience from
antenatal testing in Norway. PLoS One. 2015;10(12):e0145519.
13. Montoya JG, Remington JS. Management of Toxoplasma gondii infection
during pregnancy. Clin Infect Dis. 2008;47(4):554–66.
14. Daffos F, Forestier F, Capella-Pavlovsky M, Thulliez P, Aufrant C, Valenti D,
Cox WL. Prenatal management of 746 pregnancies at risk for congenital
toxoplasmosis. N Engl J Med. 1988;318(5):271–5.
15. Tabor A, Alfirevic Z. Update on procedure-related risks for prenatal diagnosis
techniques. Fetal Diagn Ther. 2010;27(1):1–7.
16. Cederholm M, Haglund B, Axelsson O. Infant morbidity following
amniocentesis and chorionic villus sampling for prenatal karyotyping. BJOG.
2005;112(4):394–402.
17. Cederholm M, Haglund B, Axelsson O. Maternal complications following
amniocentesis and chorionic villus sampling for prenatal karyotyping. BJOG.
2003;110(4):392–9.
18. Norwegian Institute of Public Health. Årstabeller for medisinsk
fødselsregister 2010. Oslo: Norwegian Institute of Public Health; 2012.
19. Liesenfeld O, Montoya JG, Kinney S, Press C, Remington JS. Effect of testing for
IgG avidity in the diagnosis of Toxoplasma gondii infection in pregnant women:
experience in a US reference laboratory. J Infect Dis. 2001;183(8):1248–53.
20. Sabin AB, Feldman HA. Dyes as microchemical indicators of a new
immunity phenomenon affecting a protozoon parasite (Toxoplasma).
Science. 1948;108(2815):660–3.
21. Jenum PA, Holberg-Petersen M, Melby KK, Stray-Pedersen B. Diagnosis of
congenital Toxoplasma gondii infection by polymerase chain reaction (PCR) on
amniotic fluid samples. The Norwegian experience. APMIS. 1998;106(7):680–6.
22. Lachaud L, Calas O, Picot MC, Albaba S, Bourgeois N, Pratlong F. Value of 2
IgG avidity commercial tests used alone or in association to date
toxoplasmosis contamination. Diagn Microbiol Infect Dis. 2009;64(3):267–74.
23. Villard O, Breit L, Cimon B, Franck J, Fricker-Hidalgo H, Godineau N, Houze S,
Paris L, Pelloux H, Villena I, et al. Comparison of four commercially available
avidity tests for Toxoplasma gondii-specific IgG antibodies. Clin Vaccine
Immunol. 2013;20(2):197–204.
24. Prusa AR, Kasper DC, Pollak A, Olischar M, Gleiss A, Hayde M. Amniocentesis for
the detection of congenital toxoplasmosis: results from the nationwide Austrian
prenatal screening program. Clin Microbiol Infect. 2015;21(2):191.e191–198.
25. Khoshnood B, De Vigan C, Goffinet F, Leroy V. Prenatal screening and
diagnosis of congenital toxoplasmosis: a review of safety issues and
psychological consequences for women who undergo screening. Prenat
Diagn. 2007;27(5):395–403.
26. Jenum PA, Kapperud G, Stray-Pedersen B, Melby KK, Eskild A, Eng J.
Prevalence of Toxoplasma gondii specific immunoglobulin G antibodies
among pregnant women in Norway. Epidemiol Infect. 1998;120(1):87–92.
27. KK M, PA J. Årsrapport 2015. Nasjonalt referanselaboratorium for Toxoplasma-
diagnostikk. In.: Avd. for mikrobilologi, Oslo universitetssykehus Rikshospitalet;
2015. https://oslo-universitetssykehus.no/fag-og-forskning/laboratorietjenester/
mikrobiologi/nasjonalt-referanselaboratorium-for-toxoplasmose.
28. Pappas G, Roussos N, Falagas ME. Toxoplasmosis snapshots: global status of
Toxoplasma gondii seroprevalence and implications for pregnancy and
congenital toxoplasmosis. Int J Parasitol. 2009;39(12):1385–94.
29. Stray-Pedersen B, Lorentzen-Styr AM. The prevalence of toxoplasma
antibodies among 11,736 pregnant women in Norway. Scand J Infect Dis.
1979;11(2):159–65.
Page 9 of 9
30. Findal G, Barlinn R, Sandven I, Stray-Pedersen B, Nordbo SA, Samdal HH,
Vainio K, Dudman SG, Jenum PA. Toxoplasma prevalence among pregnant
women in Norway: a cross-sectional study. APMIS. 2015;123(4):321–5.
31. de Oliveira Azevedo CT, do Brasil PE, Guida L, Lopes Moreira ME.
Performance of polymerase chain reaction analysis of the amniotic fluid of
pregnant women for diagnosis of congenital toxoplasmosis: a systematic
review and meta-analysis. PLoS One. 2016;11(4):e0149938.
32. Romand S, Wallon M, Franck J, Thulliez P, Peyron F, Dumon H. Prenatal
diagnosis using polymerase chain reaction on amniotic fluid for congenital
toxoplasmosis. Obstet Gynecol. 2001;97(2):296–300.
33. Thalhammer O. Prenatal incubation period (author’s transl). Padiatr Padol.
1972;7(1):14–9.
34. Cortina-Borja M, Tan HK, Wallon M, Paul M, Prusa A, Buffolano W, Malm G,
Salt A, Freeman K, Petersen E, et al. Prenatal treatment for serious
neurological sequelae of congenital toxoplasmosis: an observational
prospective cohort study. PLoS Med. 2010;7(10): e1000351. https://doi.org/
10.1371/journal.pmed.1000351.
35. Bjerke SE, Vangen S, Holter E, Stray-Pedersen B. Infectious immune status in
an obstetric population of Pakistani immigrants in Norway. Scand J Public
Health. 2011;39(5):464–70.
36. Wulff CB, Gerds TA, Rode L, Ekelund CK, Petersen OB, Tabor A. Risk of fetal
loss associated with invasive testing following combined first-trimester
screening for Down syndrome: a national cohort of 147 987 singleton
pregnancies. Ultrasound Obstet Gynecol. 2016;47(1):38–44.
37. Mungen E, Tutuncu L, Muhcu M, Yergok YZ. Pregnancy outcome following
second-trimester amniocentesis: a case–control study. Am J Perinatol. 2006;
23(1):25–30.
38. Tabor A, Philip J, Madsen M, Bang J, Obel EB, Norgaard-Pedersen B.
Randomised controlled trial of genetic amniocentesis in 4606 low-risk
women. Lancet. 1986;1(8493):1287–93.
39. Odibo AO, Gray DL, Dicke JM, Stamilio DM, Macones GA, Crane JP.
Revisiting the fetal loss rate after second-trimester genetic amniocentesis: a
single center’s 16-year experience. Obstet Gynecol. 2008;111(3):589–95.
40. Margioula-Siarkou C, Karkanaki A, Kalogiannidis I, Petousis S, Dagklis T,
Mavromatidis G, Prapas Y, Prapas N, Rousso D. Operator experience reduces
the risk of second trimester amniocentesis-related adverse outcomes. Eur J
Obstet Gynecol Reprod Biol. 2013;169(2):230–3.
41. Tharmaratnam S, Sadek S, Steele EK, Harper MA, Stewart FJ, Nevin J, Nevin
NC, Dornan JC. Early amniocentesis: effect of removing a reduced volume
of amniotic fluid on pregnancy outcome. Prenat Diagn. 1998;18(8):773–8.
42. Cebesoy FB, Balat O, Pehlivan S, Kutlar I, Dikensoy E, Ugur MG. Is pregnancy
loss after amniocentesis related to the volume of amniotic fluid obtained?
Arch Gynecol Obstet. 2009;279(3):357–60.
Submit your next manuscript to BioMed Central
and we will help you at every step:
• We accept pre-submission inquiries
• Our selector tool helps you to find the most relevant journal
• We provide round the clock customer support
• Convenient online submission
• Thorough peer review
• Inclusion in PubMed and all major indexing services
• Maximum visibility for your research
Submit your manuscript at
www.biomedcentral.com/submit
International Journal of Computer Assisted Radiology and Surgery (2020) 15:1807–1816
https://doi.org/10.1007/s11548-020-02242-8

ORIGINAL ARTICLE

Deep learning-based fetoscopic mosaicking for field-of-view

expansion
Sophia Bano1 · Francisco Vasconcelos1 · Marcel Tella-Amo1 · George Dwyer1 · Caspar Gruijthuijsen2 ·
Emmanuel Vander Poorten2 · Tom Vercauteren3 · Sebastien Ourselin3 · Jan Deprest4 · Danail Stoyanov1

Received: 13 April 2020 / Accepted: 30 July 2020 / Published online: 17 August 2020
© The Author(s) 2020

Abstract
Purpose Fetoscopic laser photocoagulation is a minimally invasive surgical procedure used to treat twin-to-twin transfusion
syndrome (TTTS), which involves localization and ablation of abnormal vascular connections on the placenta to regulate
the blood flow in both fetuses. This procedure is particularly challenging due to the limited field of view, poor visibility,
occasional bleeding, and poor image quality. Fetoscopic mosaicking can help in creating an image with the expanded field of
view which could facilitate the clinicians during the TTTS procedure.
Methods We propose a deep learning-based mosaicking framework for diverse fetoscopic videos captured from different
settings such as simulation, phantoms, ex vivo, and in vivo environments. The proposed mosaicking framework extends an
existing deep image homography model to handle video data by introducing the controlled data generation and consistent
homography estimation modules. Training is performed on a small subset of fetoscopic images which are independent of the
testing videos.
Results We perform both quantitative and qualitative evaluations on 5 diverse fetoscopic videos (2400 frames) that captured
different environments. To demonstrate the robustness of the proposed framework, a comparison is performed with the existing
feature-based and deep image homography methods.
Conclusion The proposed mosaicking framework outperformed existing methods and generated meaningful mosaic, while
reducing the accumulated drift, even in the presence of visual challenges such as specular highlights, reflection, texture
paucity, and low video resolution.

Keywords Deep learning · Surgical vision · Twin-to-twin transfusion syndrome (TTTS) · Fetoscopy · Sequential mosaicking

Introduction

This paper is based on the work: “Bano, S., Vasconcelos, F., Amo, M.T.,
Dwyer, G., Gruijthuijsen, C., Deprest, J., Ourselin, S., Vander Poorten,
E., Vercauteren, T. and Stoyanov, D., 2019, October. Deep sequential
mosaicking of fetoscopic videos. In: Shen D. et al. (eds) Medical Image
Computing and Computer Assisted Intervention-MICCAI 2019. MIC-
CAI 2019. Lecture Notes in Computer Science, vol 11764. Springer,
Cham.”.
Electronic supplementary material The online version of this article
(https://doi.org/10.1007/s11548-020-02242-8) contains
supplementary material, which is available to authorized users.
B Sophia Bano
sophia.bano@ucl.ac.uk
Francisco Vasconcelos
f.vasconcelos@ucl.ac.uk
Danail Stoyanov
danail.stoyanov@ucl.ac.uk
Twin-to-twin transfusion syndrome (TTTS) is a rare con-
dition during pregnancy that affects 10–15% of genetically
identical twins sharing a monochorionic placenta [5]. It is
caused by abnormal placental vascular anastomoses on the
chorionic plate of the placenta resulting in uneven flow of
1 Wellcome/EPSRC Centre for Interventional and Surgical
Sciences (WEISS) and Department of Computer Science,
University College London, London, UK
2 Department of Mechanical Engineering, KU Leuven
University, Leuven, Belgium
3 School of Biomedical Engineering and Imaging Sciences,
King’s College London, London, UK
4 Department of Development and Regeneration, University
Hospital Leuven, Leuven, Belgium

123
1808 International Journal of Computer Assisted Radiology and Surgery (2020) 15:1807–1816
blood between the fetuses. This condition puts both twins
at risk and requires treatment before birth to increase their
survival rate. Fetoscopic laser photocoagulation (Fig. 1), a
minimally invasive procedure, is the most effective treatment
for TTTS in which the surgeon uses a fetoscopic camera to
inspect and identify abnormal vascular anastomoses on the
placental chorionic plate, and uses a retractable laser ablation
tool in the working channel of the scope to photocoagulate
the vascular anastomoses to separate the blood circulation of
each twin. Limited field of view (FoV) and maneuverability
of the fetoscope, poor visibility [18] due to variability in light
source, and unusual placenta position [11] may impede the
procedure leading to increased procedural time and incom-
plete ablation of anastomoses resulting to persistent TTTS.
Fetoscopic mosaicking can create an expanded FoV image
of the placental surface, which may facilitate the surgeons in
localizing vascular anastomoses during the procedure.
Mosaicking for the FoV expansion in fetoscopy has
been explored using feature-based, intensity-based, and
deep learning-based methods [2,3,10,21,22,26,27]. Reeff et
al. [22] and Daga et al. [10] used the classical image feature-
based matching method for creating mosaics from planar
placenta images. Reeff et al. [22] experimented with the feto-
scopic images of an ex vivo placenta submerged in water,
while Daga et al. [10] used images of an ex vivo phantom
placenta. A mosaic is generated by aligning the relative trans-
formations between the pair of consecutive fetoscopic images
with respect to a reference frame. A small error in the rela-
tive transformations can introduce large drift in the mosaic,
where global consistency alignment techniques and use of
electromagnetic tracker can help to minimize the drifting
error [21,26,27]. Tella-Amo et al. [26,27] assumed placenta
to be planar and static and integrated the electromagnetic
tracker with the fetoscopic in a synthetic and ex vivo setup
to propose a mosaicking approach capable of handling the
drifting error. However, current clinical regulations and lim-
ited form factor of the fetoscope hinder the use of such a
tracker in intraoperative settings. Peter et al. [21] proposed
a direct pixel-wise alignment of gradient orientations for
creating a mosaic from a single in vivo fetoscopic video.
Gaisser et al. [15] detected stable regions on veins of the
placenta using a region-based convolutional neural network
and then used these detected regions as features for placen-
tal image registration in an underwater phantom setting [15].
Bano et al. [3] proposed a deep learning approach for pla-
cental vessel segmentation and registration for mosaicking
and showed that vessels can act as unique landmarks for
creating mosaics with minimum drifting error. Mosaicking
from fetoscopic videos particularly remains challenging due
to fetoscopic device limitations (monocular low-resolution
fetoscopic camera with FoV limitation), occlusion by the
fetus, ablation tool presence and occasional bleeding, non-
planar views, turbid amniotic fluid, specular highlights and
123
reflection due to variation in the light source, distortion due to
light refraction [9], and texture paucity. Automatic selection
of occlusion-free fetoscopic video segments [4] can help in
identifying relevant segments for mosaicking. We showed
in [2] that deep learning-based image alignment helps in
reducing the accumulated drift, even in the presence of visual
challenges such as specular highlights, reflection, texture
paucity, and low video resolution.
Supervised deep learning-based techniques estimate the
correspondences between pair of disparate natural scene
images [13,23,25] by using benchmark datasets of disparate
natural scene images with known ground-truth correspon-
dence for training. However, ground-truth correspondences
are unknown in fetoscopic videos. Moreover, [13] and [25]
used pair of high-resolution natural scene images which are
sharp and rich in both texture and color contrast, contrary to
fetoscopic videos which are of low resolution, lack both tex-
ture and color contrast since the in vivo scene is monotonic
in color, and are unsharp because of the introduced aver-
aging to compensate for the honeycomb effect of the fiber
bundle scope. As a result, hand-crafted feature-based meth-
ods perform poorly on the fetoscopic videos. Shen et al. [23]
and Srivastava et al. [25] used pretrained deep learning fea-
tures as backbone for learning the correspondences between
natural images. However, in the case of fetoscopic videos,
due to poor texture and contrast, feature maps computed
from pretrained networks may not capture distinct features
for robust correspondence estimation since these models are
pretrained on natural images (like ImageNet) which does not
capture the fetoscopic data distribution. Moreover, none of
the approaches [13,23,25] extended beyond pair of images
matching to expand the field of view from videos.
Self-supervised deep learning-based solutions can over-
come some of the challenges associated with fetoscopic
mosaicking. Image homography estimation methods have
been proposed [12,20] that use pairs of image patches
extracted from a single image to estimate the homography
between them. In practice, a full mosaic is generated by com-
puting sequential homographies between adjacent frames in
an image sequence, where fetoscopic video poses additional
challenges due to artifacts and occlusions, thus affecting the
stitching problem. However, such challenges can be tackled
by estimating the homographies between a pair of con-
secutive frames by extracting random pair of patches each
time and estimating the most consistent homography. In this
paper, we adopt this approach and propose a framework for
mosaicking from fetoscopic videos captured from various
fetoscopes and in various experimental settings. We adapt
the deep image homography (DIH) [12] estimation method
for training by assuming that the transformation between two
adjacent frames contains a rotation and translation compo-
nent only. We propose a controlled data generation approach
that uses a small set of fetoscopic images of varying qual-
International Journal of Computer Assisted Radiology and Surgery (2020) 15:1807–1816 1809

Fig. 1 Pictorial illustration of the fetoscopic laser photocoagulation for the treatment of TTTS and 1-to-1 mapping of 4-pt and 3 × 3 homography
parameterizations

ity and appearance, for training. We then perform outlier
rejection to find the consistent homography estimate between
multiple pair of patches extracted at random from two consec-
utive frames. Controlled data generation and outlier rejection
combine to minimize the drift without the use of any external
sensors. We perform comparison of the proposed fetoscopic
video mosaicking (FVM) framework with existing methods
using 5 diverse datasets. This paper is an extended version
of the work presented at the MICCAI 2019 conference [2]
and provides a broader insight into the fetoscopic mosaick-
ing research, comprehensive analysis of both qualitative and
quantitative results and comparison with the existing meth-
ods.
Problem formulation
A homography (rigid) or projective transformation, repre-
sented by a 3 × 3 matrix H, is a nonlinear transformation that
maps image points x → x between two camera views under
the planar scene assumption:
⎡ ⎤ ⎡ ⎤⎡ ⎤
u h 11 h 12 h 13 u homography parameterization 4 p H [1], instead of the 3×3
⎣v  ⎦ ∝ ⎣h 21 h 22 h 23 ⎦ ⎣v ⎦ ,
1 h 31 h 32 h 33 1 where i = 1, 2, 3, 4, denote the four corners of an image
where (u, v) is a 2D point that is mapped to (u  , v  ) with the
homography H and H is defined up to a scale; hence, it has
eight degrees of freedom.
The problem of generating a mosaic from an image
sequence is to find the pairwise homographies between
frames Fk and Fk+l (where k and l are not necessarily
consecutive frames) followed by computing the relative
homographies with respect to a fixed reference frame, also
termed as the mosaic plane. The relative homography is
represented by left-hand matrix multiplication of pairwise
homographies:

k+l−1
k
Hk+l = i
Hi+1 .
i=k
This assumes a piecewise planarity of the placental surface
observed by the fetoscope, and while not generally true, it
is sufficient for local patches of the placenta. Note that a
rigid transformation model is considered since the placental
surface does not deform over time. Moreover, there is no
perceptible placental vessel expansion/contraction due to the
breathing of the patient.
Deep image homography
Deep image homography (DIH) model [12] uses a convo-
lutional neural network to estimate the relative homography
between pairs of image patches extracted from a single image
by learning to estimate the four-point homography.
Four-point homography estimation
The rotation and shear components in the 3 × 3 parameteriza-
tion H have smaller magnitude compared to the translation;
as a result, their effect on the loss function during train-
ing is small. Therefore, DIH model [12] uses the 4-point
(1) parameterization H (Eq. 1) for the estimation. Let (u i , vi ),
PA and (u i , vi ) denote the four corners in an overlapping
image PB . Then, u i = u i − u i and vi = vi − vi give the
displacement of the ith corner point, and the 4-point homog-
raphy parameterization 4 p H is given by:
 T
u 1 u 2 u 3 u 4
4p
H= . (3)
v1 v2 v3 v4
A one-to-one mapping exists between the 4-point 4 p H and
3 × 3 H homography parameterizations. With the (u i , vi )
and (u i , vi ) known in Eq. 1, H can be computed by applying
direct linear transform [16] (Fig. 1).
DIH is a VGG-like [24] network (Fig. 2) which is used
(2) for regressing the displacement between the four corner
points. The network consists of 8 convolutional layers and

123
1810 International Journal of Computer Assisted Radiology and Surgery (2020) 15:1807–1816
Fig. 2 DIH regression network with controlled data generation for training
2 fully connected layers. The input of the network is PA
and PB extracted at random from a single image, and out-
put is their relative homography 4 p H A . For the gradient
B
back-propagation during the training process (represented
by dotted line in Fig. 2), the network uses the Euclidean loss
1
4 p
2
L2 =
H −4 p H
, (4)
2
where 4 p H and 4 p H are the ground-truth (GT) and pre-
dicted 4-point homographies. Note that [12] used the MS-
COCO dataset for training, where pair of patches were
extracted from a single static real-world image, free of arti-
facts (e.g., specular highlights, amniotic fluid particles) that
appear in sequential data.
Limitation of deep image homography
For the DIH model, the training data are generated by ran-
domly selecting an image patch PA of size 128 × 128 from a
grayscale image and randomly perturbing all its four corners
by a maximum of 32 pixels to obtain PB and the relative
GT 4 p H BA . It is observed through experimentation that data
generation by performing random perturbation (as proposed
in [12]) results in scenarios that are difficult for the net-
work to learn, hence resulting in higher error. In the case
of mosaicking, where homography between frames Fk and
Fk+l is computed by accumulating the intermediate pairwise
homographies (Eq. 2), even a small error in pairwise homog-
raphy will accumulate over time resulting in increasing drift.
Therefore, this data generation model cannot be used as it is
for creating mosaics from sequential data.
Fetoscopic video mosaicking
An overview of the proposed fetoscopic video mosaick-
ing (FVM) is shown in Fig. 3, which can be divided into
two stages, (1) data generation for regression network train-
ing (Sect. 4.1) and (2) consistent homography estimation
(Sect. 4.2). To overcome the limitation of DIH, we propose
123
controlled data generation in which DIH is trained on pairs
of patches generated by introducing translation and rotation
transformations only on a single image. During testing, we
apply the median filter to decomposed homography estima-
tions from different patches of the same pair of consecutive
frames to get a robust estimate of the homography.
Data generation for regression network training
In sequential data, pairwise homography between two adja-
cent frames Fk and Fk+1 is related by affine transformations
including rotation, translation, scale, and shear, where the
scale can be considered as constant since fetoscopy is
performed at a fixed distance from the placental surface.
Moreover, the motion of the fetoscope is physically con-
strained by the incision point (remote center of motion),
making shear component extremely small compared to the
translation and rotation components. Therefore, scale and
shear components can be neglected. We assume that Fk and
Fk+1 are related by the translation and rotation components
only. This assumption helps in minimizing the error in rela-
tive homography between frames.
For controlled data generation, given a grayscale image I
of size 256 × 256, an image patch PA of size 128 × 128 is
extracted at a random location with corner points given by
(u i , vi ), where i = 1, 2, 3, 4. The four corners for PB are
then computed by applying a rotation by β and translation
by (dx , d y ) to (u i , vi ):
      
ui cosβ sinβ ui d
= + x , (5)
vi −sinβ cosβ vi dy
where β, dx and d y are empirically selected. 4 p H BA is then
obtained using Eq. 3. PA , PB , and 4 p H BA form the input of
the regression network. During training (Fig. 2), the relative
homography is learned between patches that are extracted
from a single image following the controlled data generation
process. During testing where the patches are extracted from
the consecutive frames, the homography estimate may not
International Journal of Computer Assisted Radiology and Surgery (2020) 15:1807–1816 1811

Fig. 3 Overview of the proposed FVM framework

always be accurate due to texture paucity and poor contrast useful for outlier rejection. Hence, we compute the median
in fetoscopy. over all the iterations for (
θn )n=1
N to get its argument:

Consistent homography estimation

θi = arg median((
θi )n=1
N
), (7)
n
Outlier rejection is performed during testing by applying
median filtering to the homographies estimates obtained from which gives the most consistent value for θ . The values of
random patches of the same pair of frames (Fig. 3). To
θi ,
γi , s yi ,
sxi , txi and
t yi are then plugged into Eq. 6 to get
obtain a robust estimate of the homography, we first need to the consistent i H k .
k+1
decompose H. We apply singular value decomposition [19]
which decomposes H into a rotation, a non-uniform scale,
and another rotation, given by: Experimental details
     
h 11
h 12 cos
θ sin
θ sg 0 γ
cos γ
sin
= , Dataset
h 21 h 22 −sin
θ cos
θ 0
sh −sin
γ γ
cos
(6)
For the experimental analysis, we use five fetoscopic videos
that captured phantom and real human placenta in ex vivo
where θ, sg and
γ , sh are the unknowns. Since the left-hand
and in vivo environments. Our video data include 2 videos
side in Eq. 6 is known, we can solve the simultaneous equa-
from the existing literature, namely synthetic (SYN)—a dis-
tions for θ,
γ , sg and sh (for derivation refer to [7]). The
as continuous version of this sequence was used in [26], and an
translation components can be extracted directly from H
ex vivo in water (EX) data reported from [14]. We also cap-
tx = h 13 and t y = h 23 . For affine transformation, the homog-
tured two videos using placenta phantom in in-house settings.
raphy parameters h 31  0 and h 32  0. In Eq. 6, the rotation
The first phantom video, referred as PHN1, was captured
matrices are orthogonal and the scale matrix is diagonal.
using a rigid placenta model in air placed in a surgical trainer
For a pair of consecutive frames Fk and Fk+1 , we compute
k for N = 99 iterations such that at box [17]. The second phantom video, referred as PHN2, was
the homography n H k+1
captured using a TTTS phantom.1 PHN1 and PHN2 were
each iteration, a new random pair of patches n Pk and n Pk+1 is
captured with Storz rigid 30◦ and 0◦ scopes, respectively,
used. This results in slightly varying estimations at some iter-
having light source in one of their working channels. The fifth
ations due to varying visual quality across the sequence. For
video sequence is from an in vivo TTTS procedure (INVI)
each iteration, we obtain the decomposed parameters where
that intraoperatively captured the human placenta. PHN1,
the rotation components can be represented as ( θn )n=1
N and
PHN2, and INVI differ significantly from SYN and EX as
γn )n=1 . The variations in scale components are very small
( N

due to fixed scale assumption during the training process. But

the variations in the rotation components are significant and 1 Surgical Touch Simulator https://www.surgicaltouch.com/.

123
1812 International Journal of Computer Assisted Radiology and Surgery (2020) 15:1807–1816

Table 1 Main characteristics of the videos used for the experimental analysis
Video name Imaging # Frames Frame resolu- Cropped frame Camera view Motion type
source tion (pixels) resolution (pixels)

Synthetic – 811 385 × 385 260 × 260 Planar Circular

(SYN) [26]
Ex vivo in water Stereo 404 250 × 250 250 × 250 Planar Spiral
(EX) [14]
Phantom without Rigid 30◦ scope 681 1280 × 960 834 × 834 Non-planar Circular (freehand)
fetus (PHN1)
TTTS Phantom Rigid scope 350 720 × 720 442 × 442 Non-planar with Exploratory (freehand)
in water (PHN2) heavy occlusions
In vivo TTTS Rigid scope 150 470 × 470 312 × 312 Non-planar with Exploratory (freehand)
procedure (INVI) heavy occlusions

they captured non-planar views with freehand motion, thus
creating challenging scenarios for mosaicking methods.
Table 1 summarizes the main characteristics of the five
test videos, and representative images from these videos are
shown in Fig. 4. The visual quality, appearance, frame reso-
lution, and imaging source vary across all five videos. These
variations pose challenging scenarios for mosaicking meth-
ods. SYN and EX captured a planar view and followed a
circular loop closure and spiral motion, respectively. EX was
captured using a KUKA articulated arm robot and followed a
pre-programmed fixed spiral trajectory [14]. PHN1 captured
non-planar views that followed a freehand circular trajectory
depicting a scenario with loop closure. PHN2 and INVI cap-
tured highly non-planar views containing heavy occlusions.
Experimental setup
The recorded videos are converted to frames using the FFm-
peg software.2 To extract a square frame from the circular
(Fig. 4), in order to use it as the input for the proposed
network, we detect the circular mask of the scope through
pixel-based image threshold and morphology. A square
cropped frame is then extracted such that it is an inscribed
square within the circular mask (Table 1). Note that the
resolution of frames varies as they were captured from dif-
ferent imaging sources. For training (Sect. 4.1), we extracted
2
600 frames at random from SYN, PHN1, PHN2, INVI, and
k −1
and train the regression network for 15 hours on a Tesla
V100 (32GB) using a learning rate of 10−4 and ADAM
optimizer. Pairs of patches with controlled data augmenta-
tion (Sect. 4.1) are generated at run time in each epoch by
randomly selecting β between (−5◦ , +5◦ ), and dx and d y
between (−16, 16). The regression network is trained for
60,000 epochs with a batch size of 32. Same training settings
are used for training the regression model without controlled
data augmentation where each corner point of PA is perturbed
at random between (−16, 16).
Evaluation protocol
We perform comparison of the proposed FVM with a feature-
based (FEAT) [8] and DIH [12] methods. FEAT extract
speeded up robust features [6] from a pair of images and
performs an exhaustive search for feature matching to esti-
mate the homography. In fetoscopic videos, the GT pairwise
homographies between consecutive frames are unknown.
Hence, the accumulated error over time can mainly be
observe through qualitative analysis. The GT homographies
are only available for the SYN data. Therefore, we compute
the residual error for the evaluation on this data as:
S I
W SI H
1

eH =
(Hk+1 ) xi − (Hk+1
k
)−1 xi
, (8)
another in-house ex vivo still images dataset. Note that the SI W SI H
i=1
still image ex vivo dataset is not a video sequence; hence,
it was only used during training as it does not satisfy the
continuous video assumption. EX dataset (Table 1) was not where xi is the coordinate of the ith position in the image,
used during training; hence, it is a completely unseen data k+1 and Hk+1 are the estimated and GT homographies from
H k k
used for testing the generalization of the proposed method Fk to Fk+1 , respectively, and S I W is the width and S I H is the
to an unseen dataset. We converted the training images to height of a patch.
grayscale and resized them to 256 × 256 resolution. We For the quantitative evaluation of the remaining videos,
use Keras with Tensorflow backend for the implementation we report the root mean square error between the GT 4 p H BA
A 4-point homographies obtained when
and estimated 4 p H B
2 FFmpeg https://ffmpeg.org. the two patches are extracted from a single image (Sect. 4.1).

123
International Journal of Computer Assisted Radiology and Surgery (2020) 15:1807–1816 1813

(a) Synthetic (SYN) (d) TTTS Phantom (PHN2)

(b) Ex-vivo (EX)

(e) Invivo (INVI)

(c) Phantom Placenta (PHN1)

Fig. 4 Representative frames from the five videos under analysis

(a) GT (b) FVM (c) DIH (d) FEAT

(e) Residual error (f) Root mean square error (g) Photometric error
Fig. 5 a–d Visualization of mosaics for one circular loop (360 frames) of the SYN sequence. e–g Quantitative comparison of FEAT, DIH and FVM

S I
W SI H
This is given by: 1


eP =
P − Pk+1
. (10)
k
SI W SI H
i=1
 4   1/2
i=1 (u i − 
u i )2 + (vi − 
vi )2
eR = , (9) We report the box plots for the three error metrics in the next
4
section.

where u i and vi are the GT displacements of the four

corners and  u i and  vi are the estimated displacements. Results and discussion
Finally, we report the photometric error (PE) between patch
Pk+1 and reprojected patch Pk obtained by warping Pk using Figure 5a–d shows the qualitative comparison results on one
the estimated homography H k . The photometric error is circular loop (360 frames) of the SYN sequence. Figure 5e–g
k+1
computed as: shows the quantitative comparison in terms of the residual,

123
1814 International Journal of Computer Assisted Radiology and Surgery (2020) 15:1807–1816
Fig. 6 Quantitative comparison of FVM, DIH, and FEAT on the test videos
root mean square and photometric errors for the complete
length of the SYN sequence. We can observe from these
visualizations that the drift is minimal in the case of FVM
compared to DIH and FEAT. In the case of DIH, tracking
is lost in just 30 frames mainly because of random per-
turbation of the four corners to generate the training data
(Sect. 3.2). This results in extremely high mean residual error
for DIH (Fig. 5e). Compared to DIH and FEAT, the three
error metrics (Fig. 5e–g) are low for FVM which correlates
with the observations from the visualizations. For FVM, the
median values for residual, root mean square, and photomet-
ric errors are 3.88, 0.29, and 2.42, respectively, which are
significantly better than FEAT median values (15.67, 7.6,
2.94).
Quantitative comparison of the proposed FVM with FEAT
and DIH reporting the root mean square error is presented
in Fig. 6a for the EX, PHN1, PHN2, and INVI videos.
For the proposed FVM, the median value of the root mean
square error for EX is 0.30, PHN1 is 0.27, PHN2 is 0.30,
and INVI is 0.29. These values are significantly lower than
DIH and FEAT. Error for DIH is also low but not as low
as FVM. The root mean square errors for FVM and DIH
are particularly low because the optimization of these meth-
ods is done on the 4-point homographies, and root mean
square error is also calculated using this representation. How-
ever, the introduced error in DIH is higher, compared to
FVM, mainly because of the random perturbation of the
corner points during training. A similar performance trend
is observed from the photometric error (Fig. 6b) for which
FVM returned the median value of 0.90 in EX, 1.72 in
PHN1, 1.46 in PHN2, and 2.41 in INVI videos. Note that
the median in the case of FVM for all four test videos is
lower than the first quartile (25th percentile) of DIH and
123
FEAT. Moreover, the interquartile range is very small in the
case of FVM, depicting that the error at each frame is con-
centrated in a small range. Compared to FVM, the mean and
interquartile range of DIH and FEAT are high because of
inaccurate homography estimates resulting in higher repro-
jection error. These results and observations are in line with
the qualitative analysis that is presented in the subsequent
paragraphs.
Figure 7 shows the visualization result of the proposed
FVM for the EX, PHN1, PHN2, and INVI videos. The visu-
alization results for DIH and FEAT are presented in the
supplementary material. EX (unseen data) is of low reso-
lution with blurred frames and captures a spiral scanning
motion. FVM created a meaningful mosaic for EX video
with minimum drift accumulation over time. PHN1 is the
longest video under analysis and has non-planar views with-
out occlusions with the camera following a circular trajectory.
FVM managed to generate reliable mosaics with minimum
drift which can be observed from frame 681 in Fig. 7b, where
loop closure with minimal drift can be seen. Unlike the exist-
ing methods that use external sensors for minimizing the
drift [26], FVM relies only on image data and generates
meaningful mosaics with minimum drift even for non-planar
sequences.
PHN2 represents a challenging scenario as it contained
highly non-planar frames with heavy occlusions, low reso-
lution, and texture paucity. None of the existing fetoscopic
mosaicking literature investigated such a scenario. DIH and
FEAT failed to register this sequence (refer to supplemen-
tary), while FVM gave promising results. We observe from
Frames 250 and 350 (Fig. 7c) that although the generated
mosaic can serve well for increasing the FoV, there is a
noticeable drift due to heavy occlusions and highly non-
International Journal of Computer Assisted Radiology and Surgery (2020) 15:1807–1816
(a)
(b)
Fig. 7 Qualitative results of the proposed FVM
planar views. Such errors may be corrected by end-to-end
training using the photometric loss [20]. INVI is taken from
a TTTS fetoscopic procedure and contains occlusions due to
the appearance of the fetus in the FoV, reflection from float-
ing particles, illumination variation, and low resolution. DIH
failed to register consecutive frames in this sequence. FEAT
lost tracking around 50th frame due to inaccurate feature
matches (refer to supplementary). However, FVM (Fig. 7d)
managed to generate a meaningful mosaic for the complete
duration of the sequence with noticeable drift.
The quantitative and qualitative comparison on five
diverse fetoscopic test videos shows that the proposed FVM
is capable of handling several visual challenges such as
varying illumination, specular highlights/reflections, and low
resolution along with non-planar views with occlusions. The
proposed FVM solely relied on the image intensity data and
generated reliable mosaics with minimum drift even for non-
planar test videos.
Conclusion
We proposed a deep learning-based fetoscopic video mosaick-
ing framework which is shown to handle fetoscopic videos
captured in different settings such as simulation, phantoms,
ex vivo, and in vivo environments. The proposed method
extended an existing image homography method to handle
sequential data. This is achieved by introducing a con-
trolled training data generation stage which assumed that
there is only a small change in rotation and translation
between two consecutive fetoscopic frames. Homography
estimates slightly vary between two consecutive frames
when selecting patch location randomly during testing due
to texture paucity and visual variations; hence, this can
introduce drifting error. To handle this issue, we intro-
duced the consistent homography estimation stage that
pruned the homography estimate between multiple pair of
patches extracted at random from two consecutive frames.
1815
(c)
(d)
Quantitative and qualitative evaluations on five diverse feto-
scopic videos showed that, unlike existing methods that
are unable to handle visual variations and drift rapidly
in just a few frames, our method produced mosaics with
minimal drift without the use of any photo-consistency
(loop closure) refinement method. Such a method may
provide computer-assisted interventional support for TTTS
treatment to facilitate the localization of abnormal placen-
tal vascular anastomoses by providing an expanded FoV
image.
Acknowledgements This work is supported by the Wellcome/EPSRC
Centre for Interventional and Surgical Sciences (WEISS) at UCL
(203145Z/16/Z), EPSRC (EP/P027938/1, EP/R004080/1, NS/A0000
27/1), H2020 FET (GA 863146) and Wellcome (WT101957). Danail
Stoyanov is supported by a Royal Academy of Engineering Chair
in Emerging Technologies (CiET1819/2/36) and an EPSRC Early
Career Research Fellowship (EP/P012841/1). Tom Vercauteren is sup-
ported by a Medtronic/Royal Academy of Engineering Research
Chair (RCSRF1819/7/34).
Compliance with ethical standards
Conflict of interest The authors declare that they have no conflict of
interest.
Ethical approval For this type of study, formal consent is not required.
Informed consent No animals or humans were involved in this research.
All videos were anonymized before delivery to the researchers.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing, adap-
tation, distribution and reproduction in any medium or format, as
long as you give appropriate credit to the original author(s) and the
source, provide a link to the Creative Commons licence, and indi-
cate if changes were made. The images or other third party material
in this article are included in the article’s Creative Commons licence,
unless indicated otherwise in a credit line to the material. If material
is not included in the article’s Creative Commons licence and your
intended use is not permitted by statutory regulation or exceeds the
permitted use, you will need to obtain permission directly from the copy-
right holder. To view a copy of this licence, visit http://creativecomm
ons.org/licenses/by/4.0/.
123
1816 International Journal of Computer Assisted Radiology and Surgery (2020) 15:1807–1816
References
1. Baker S, Datta A, Kanade T (2006) Parameterizing homographies.
Technical Report CMU-RI-TR-06-11
2. Bano S, Vasconcelos F, Amo MT, Dwyer G, Gruijthuijsen C,
Deprest J, Ourselin S, Vander Poorten E, Vercauteren T, Stoy-
anov D (2019) Deep sequential mosaicking of fetoscopic videos.
In: International conference on medical image computing and
computer-assisted intervention, Springer, pp 311–319
3. Bano S, Vasconcelos F, Shepherd LM, Vander Poorten E, Ver-
cauteren T, Ourselin S, David A L, Deprest J, Stoyanov D (2020)
Deep placental vessel segmentation for fetoscopic mosaicking.
In: International conference on medical image computing and
computer-assisted intervention. Springer. arXiv:2007.04349
4. Bano S, Vasconcelos F, Vander Poorten E, Vercauteren T, Ourselin
S, Deprest J, Stoyanov D (2020) FetNet: a recurrent convolutional
network for occlusion identification in fetoscopic videos. Int J
Comput Assist Radiol Surg 15(5):791–801
5. Baschat A, Chmait RH, Deprest J, Gratacós E, Hecher K,
Kontopoulos E, Quintero R, Skupski DW, Valsky DV, Ville Y
(2011) Twin-to-twin transfusion syndrome (TTTS). J Perinat Med
39(2):107–112
6. Bay H, Tuytelaars T, Van Gool L (2006) SURF: Speeded up robust
features. In: Proceedings of the European conference on computer
vision, Springer, pp 404–417
7. Blinn J (2002) Consider the lowly 2×2 matrix, chap 5. In: Jim
Blinn’s corner: notation, notation, notation. Morgan Kaufmann
8. Brown M, Lowe DG (2007) Automatic panoramic image stitching
using invariant features. Int J Comput Vis 74(1):59–73
9. Chadebecq F, Vasconcelos F, Lacher R, Maneas E, Desjardins A,
Ourselin S, Vercauteren T, Stoyanov D (2019) Refractive two-view
reconstruction for underwater 3d vision. Int J Comput Vis 128:1–17
10. Daga P, Chadebecq F, Shakir DI, Herrera LCGP, Tella M, Dwyer G,
David AL, Deprest J, Stoyanov D, Vercauteren T (2016) Real-time
mosaicing of fetoscopic videos using SIFT. In: Medical Imaging:
Image-Guided Procedures
11. Deprest J, Van Schoubroeck D, Van Ballaer P, Flageole H, Van
Assche FA, Vandenberghe K (1998) Alternative technique for Nd:
Yag laser coagulation in twin-to-twin transfusion syndrome with
anterior placenta. Ultrasound Obstet Gynecol J 11(5):347–352
12. DeTone D, Malisiewicz T, Rabinovich A (2016) Deep image
homography estimation. RSS Workshop on Limits and Potentials
of Deep Learning in Robotics
13. Dusmanu M, Rocco I, Pajdla T, Pollefeys M, Sivic J, Torii A,
Sattler T (2019) D2-Net: A trainable CNN for joint description and
detection of local features. In: Proceedings of the IEEE conference
on computer vision and pattern recognition, pp 8092–8101
14. Dwyer G, Chadebecq F, Amo MT, Bergeles C, Maneas E, Pawar V,
Vander Poorten E, Deprest J, Ourselin S, De Coppi P (2017) A con-
tinuum robot and control interface for surgical assist in fetoscopic
interventions. IEEE Robot Autom Lett 2(3):1656–1663
123
15. Gaisser F, Peeters SH, Lenseigne BA, Jonker PP, Oepkes D (2018)
Stable image registration for in-vivo fetoscopic panorama recon-
struction. J Imaging 4(1):24
16. Hartley R, Zisserman A (2003) Multiple view geometry in com-
puter vision, Chapter 4: Estimation - 2D Projective Transforma-
tions. Cambridge University Press, Cambridge
17. Javaux A, Bouget D, Gruijthuijsen C, Stoyanov D, Vercauteren
T, Ourselin S, Deprest J, Denis K, Vander Poorten E (2018) A
mixed-reality surgical trainer with comprehensive sensing for fetal
laser minimally invasive surgery. Int J Comput Assist Radiol Surg
13(12):1949–1957
18. Lewi L, Deprest J, Hecher K (2013) The vascular anastomoses in
monochorionic twin pregnancies and their clinical consequences.
Am J Obstet Gynecol 208(1):19–30
19. Malis E, Vargas M (2007) Deeper understanding of the homogra-
phy decomposition for vision-based control. Ph.D. Thesis, INRIA
20. Nguyen T, Chen SW, Shivakumar SS, Taylor CJ, Kumar V (2018)
Unsupervised deep homography: a fast and robust homography
estimation model. IEEE Robot Autom Lett 3(3):2346–2353
21. Peter L, Tella-Amo M, Shakir DI, Attilakos G, Wimalasundera R,
Deprest J, Ourselin S, Vercauteren T (2018) Retrieval and registra-
tion of long-range overlapping frames for scalable mosaicking of
in vivo fetoscopy. Int J Comput Assist Radiol Surg 13(5):713–720
22. Reeff M, Gerhard F, Cattin P, Gábor S (2006) Mosaicing of
endoscopic placenta images. INFORMATIK 2006–Informatik für
Menschen, Band 1
23. Shen X, Darmon F, Efros AA, Aubry M (2020) RANSAC-
Flow: generic two-stage image alignment. arXiv preprint
arXiv:2004.01526
24. Simonyan K, Zisserman A (2015) Very deep convolutional net-
works for large-scale image recognition. In: International confer-
ence on learning representations
25. Srivastava S, Chopra A, Kumar AC, Bhandarkar S, Sharma D
(2019) Matching disparate image pairs using shape-aware con-
vnets. In: IEEE winter conference on applications of computer
vision, IEEE, pp 531–540
26. Tella-Amo M, Peter L, Shakir DI, Deprest J, Stoyanov D, Iglesias
JE, Vercauteren T, Ourselin S (2018) Probabilistic visual and elec-
tromagnetic data fusion for robust drift-free sequential mosaicking:
application to fetoscopy. J Med Imaging 5(2):021217
27. Tella-Amo M, Peter L, Shakir DI, Deprest J, Stoyanov D, Ver-
cauteren T, Ourselin S (2019) Pruning strategies for efficient
online globally-consistent mosaicking in fetoscopy. J Med Imaging
6(3):035001
Publisher’s Note Springer Nature remains neutral with regard to juris-
dictional claims in published maps and institutional affiliations.
 DAFTAR PUSTAKA

1. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5405501/
2. Http://www.surgeryencyclopedia.com/Ce-Fi/Fetoscopy.html&prev=search
3. https://link.springer.com/article/10.1007/s11548-020-02242-8