See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/324219743

Unexpected Results: Embedded Information in Fingerprints Regarding

Diabetes

Article · April 2018

CITATIONS READS

0 212

6 authors, including:

Constantin Ionescu-Tîrgovişte Daia Alexandru

National Institute of Diabetes, Nutrition and Metabolic Diseases, N… UPWORK
295 PUBLICATIONS   11,051 CITATIONS    6 PUBLICATIONS   0 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Unexpected Results: Embedded Information in Fingerprints Regarding Diabetes View project

Acupuncture research View project

All content following this page was uploaded by Daia Alexandru on 05 April 2018.

The user has requested enhancement of the downloaded file.

 Unexpected Results: Embedded Information in Fingerprints Regarding
Diabetes
Author Block: Constantin Ionescu-Tirgoviste1, Alexandru Daia2, Nicoleta Dragana1,
Siddhant Bansal4, C. Vulpe3, L. Guja1
 1. University of Medicine and Pharmacy ‘’C. Davila’’ Bucharest, Romania
 2. Deep Factory Bucharest, Romania
 3. Institute of Anthropology ‘’Francisc Rainer’’, Bucharest, Romania
 4. Deep Factory Gujarat, India

Importance of Diabetes Prediction institutions in the early 2000 revealed that

more and more people in the USA were
For more than a decade now, actually
developing diabetes, with about 3.5
since the year 2000, diabetes has become
million people reported to have Type 2
one of the major health issues affecting
Diabetes Mellitus. In UK in the same
the general population worldwide.
year, the figures were reported to be
Reports from various health care
approximately 549,000. The situation is
not any better in other developed and
developing countries in fact, according to
the World Health Organization (WHO)
this situation is expected to worsen in the
near future. In a report released in 2015,
WHO stated that by 2030, more than 552
million people globally will be affected
by diabetes. This is based on the current
trends in the health condition of
individuals and continues changes in
lifestyles. Case in point is, that with
increased developments in technology and
the nature of the current community, less
people exercise regularly and eat healthy.
This means that more and more people
are prone to major health risk factors.
Poor diet and inactivity that are strongly
linked to type-2 Diabetes. Type-2
Diabetes is currently reported to account
for more than 90% of the people predicted
to have diabetes worldwide. Therefore,
there is need for stakeholders to focus on
research studies that aim to predict
possibilities of developing diabetes and
also provide measures that can be
undertaken to prevent diabetes. Whereas
in comparison type-1 diabetes is defined
as an auto-immune condition in which the
immune system of the patient is activated
to destroy the cells that produce insulin in
the pancreatic organ. Studies are yet to
reveal what causes the reaction of the
immune system and why it cannot be
cured, whereas type-1 diabetes can be
prevented.
Some recent studies have revealed
that change in lifestyles and use of drug
intervention measures can serve to incurred during the treatment of diabetes
prevent diabetes. Other scholars argue that can be heavy on the society as well as
that by studies accurately predict the on the individuals.
possibility of a given population,
With the early prediction of diabetes
developing diabetes would be more
and identification of the people at a high
beneficial to a community that is already
risk of developing the illness at the right
at a high risk of living poor quality lives.
time, the chances are elevated of making
The use of prediction strategies is very
sure that diabetes is no longer a chronic
important as it allows proper
disease of affluence that has become a
implementation of prevention strategies
major issue for the general public
that lower the risk of individuals
worldwide but more of an a fortuitous
developing diabetes. In addition, the use
event.
of diabetes prediction techniques, the
community and society at large is able to
avoid the burden that is associated with
application of mechanisms and cost of
preventing diabetes. This is according to Unexpected method for prediction
Kirwan and Sacks who further explain diabetes: Preliminary data from
that it also reduces the cost that could be dermatoglyphics analysis of type 1 and
 type 2 diabetic versus non-diabetic from the huge and various patterns
subjects present in dermatogliphes, could be easily
available and cost-effective method.

Background and aims:

The Holly Grail of diabetological Subjects and methods:
science is that to prevent diabetes. For
This study included 314 patients with
that we need a good prediction, before
type-2 diabetes (156 M/158 F), 137
occurrence the autoimmune markers or a
patients with -1 diabetes (78M/59F) and
progressive decrease of the B-cell
287 controls (130M/157F). In order to
function. The first degree subjects of a
record the dermatogliphs we used the
patient with type-1 or type-2 diabetes
classical ink method. Among the
could be one of the criteria, who's value
numerous quantitative and qualitative
can be increase by a genetic risk score,
dermatoglyphic traits, we assessed
which however have a high cost/efficient
variables including: finger patterns
ratio and difficult to be applied at a large
(whorl, loop and arch), palmar line a, b, c,
scale. Among the non invasive methods
d and palmar axial triradius position (t, t’,
and with a high potential in prediction of
t’’).
diabetes, the extraction of the information
 The analyses of the dermatogliphic
images has been done by two methods:
one according to the classical visual
comparison of various fields between
normal and type-1 and type-2 diabetic
patients and another using the new
information technology adapted to image
analysis.
Results and discussions:
Preliminary data obtained by manual
analysis carried out by an experienced
anthropologist, generally overlap with
some of those published on small group
of diabetic versus non-diabetic patients.
Areas identified as different in shape or
frequency in diabetic to non-diabetic is
numerous and depends on analyzed
regions, fingers or palms, indicating their
location according to the agreed
nomenclature. The major disadvantage of
the manual method is that it is time
consuming even in the case of an
experienced investigator. In order to
overcome this obstacle, on the same
images obtained on the paper, we applied
a new original method of automatic
analysis using software based on Neural
Networks built for this purpose, whose
validation could lead to the possibility of
faster and more accurate identification of
the prediction areas indicating the
predisposition for diabetes. As method we
have chosen to use Deep Neural Nets with
Convolutional filters, because our initial
opinion was that using multiple
convolutional filters we could detect
different patterns in the fingerprints.
 Network API, written in Python language
and uses Tensor Flow as backend
framework developed by Google.

Convolutional
Neural Networks are hyper-contextually
aware - they know what’s going on above,
below, to the right and to the left, even
what is going on in front and behind if
you use image depth.
We have also used a particular type of
Deep Neural Network library called
Keras. Which is a high-level Neural
Flow of Neural Network
Using this method on our diabetic patients character of diabetic syndrome by the
and the control group, we have obtained common final target: the decrease in the
an average accuracy of prediction from 50 β-cell function/mass. We expect that by
different random cross-validation splits analyzing the more complex
and obtained a high accuracy of around dermatoglyphes of the palm we may
76% for type 1 using also the dropout indicate the autoimmune versus non-
regularization method for improving autoimmune mechanism affecting the β-
performmance and reducing overfitting . cells.
In this way after using dropout
regularization the standard deviation
dropped to 0.9 from 1.2 which proved
experiment was succesful
Neural Networks reported 73% accuracy
on only type-2 patients data when
compared to control. An important
observation was that this method did not Convolutional Neural Network
discriminate between type-1 and type-2 Conclusions:
diabetes, probably indicating the unitary
 Our preliminary analysis using humanity to develop a prediction model
convolutional neural networks show that for such issues.
even in the 4 fingerprints there is a large
amount of information embedded for
discrimination between diabetic vs. non-
diabetic subjects. We are now exploring
other artificial intelligence methods like
Residual Neural Networks, in order to
evaluate behavior on our data or to
visualize the convolutional filters in the
last hidden layer of neurons in order to
indicate the particular regions from finger
prints where the influential information
patterns exists. Such method can be
applied successfully in population studies
with the goal to select subjects having a
high risk of developing type-1 or type-2
diabetes. It is an important issue of

View publication stats