IMAGEANALYSIS OF CERVICAL CANCER

AIM
To analyse the cancerous cervix images in automated way for differentiating normal and the cancerous regions of the uterine cervix by image segmentation technique using matlab and to obtain severity level of the cancer

OBJECTIVES
To Convert input RGB image to CBCR format To do Intensity adjustment and displaying HSV image To extract S layer by image erosion To extract feature and entropy calculation To find out the spectrum of the image using FFT

INTRODUCTION
Cervical cancer is the second most common form of cancer among women, with nearly half a million new cases diagnosed annually. Much progress has been made in the prevention and control of cervical cancer . Cancer of the cervix is primarily caused by human papillomavirus (HPV) infection. Several studies showed the feasibility of using digital image processing techniques to automatically interpret colposcopic images. The proposed idea presents a domain-specific automated image analysis framework for the detection of pre-cancerous and cancerous lesions of the uterine cervix and so we can detect the moderate and severe cases of cervical cancer with the help of MATLAB. The input RGB image is first converted into YCBCR format for splitting. DCT is used for normalization. Morphological operations are performed on the image. The Back propagation classifier(neural network) is used for the classification purpose. FFT is used for finding out the spectrum of the image.

LITERATURE REVIEW
1. Title : Automated and interactive lesion detection and segmentation in uterine cervix images.
Amir Alush, Hayit Greenspan, and Jacob Goldberger ,IEEE transactions on medical imaging, vol. 29, no. 2, february 2010. This paper presents a procedure for automatic extraction and segmentation of a class-specific object (or region) by learning class-specific boundaries. We describe and evaluate the method with a specific focus on the detection of lesion regions in uterine cervix images. The watershed segmentation map of the input image is modeled using a Markov random field (MRF) in which watershed regions correspond to binary random variables indicating whether the region is part of the lesion tissue or not. Automated extraction and segmentation is the concept used in the project which is related to the base paper where DCT is used for splitting and normalization instead of watershed algorithm which is used for final lesion region segmentation.

LITERATURE REVIEW
2.Title :Domain-specific image analysis for cervical neoplasia detection based on conditional random fields
sun y. park, member, ieee, dustin sargent, member, ieee, richard lieberman, and ulf gustafsson ,IEEE transactions on medical imaging, vol. 30, no. 3, march 2011 This paper presents a domain-specific automated image analysis framework for the detection of pre-cancerous and cancerous lesions of the uterine cervix. Our proposed framework departs from previous methods in that we include domain-specific diagnostic features in a probabilistic manner using conditional random fields. Likewise, we provide a novel window-based performance assessment scheme for 2D image analysis which addresses the intrinsic problem of image misalignment .Domain-specific automated image analysis framework for the detection of pre-cancerous and cancerous lesions of the uterine cervix is the concept used in the project which is related to the base paper where back propogation of neural network is used for diagnosis and classification instead of the probabilistic manner for detecting cervical cancer or precancerous lesions using a CRF model.

FLOW CHART
start Input image Converting RGB into CBCR format Splitting 8*8 blocks

Normalization by DCT
Contrast and illumination balancing Display intensity image Display HSV image

Image erosion
Superimposed input and output images Binary image Eccentricity and entropy calculation Applying FFT stop

ALGORITHM
Step 1: Converting input RGB image into CBCR format Step 2: Splitting the converted image into 8*8 matrix for segmenting the image Step 3: Contrast and illumination adjustment to the segmented image Step 4: Displaying intensity image and hue saturation value(HSV) image Step 5: Image erosion is done to the HSV image Step 6: Super imposing of input and output images Step 7: Converting the superimposed image into binary form Step 8: Feature extraction includes eccentricity and entropy calculation Step 9: Finding out the spectrum of the image using FFT and classification of image by neural network algorithm

TECHNICAL DETAILS

The original image acquired in the RGB color space is transformed into the luminance(Y), chrominance blue(Cb), and chrominance red(Cr) color space. Discrete Cosine Transform for splitting the image into 8*8 matrix for detecting cancerous region minutely and effectively known as segmentation process. The Hue Saturation and Value(HSV) of the image is calculated and image erosion is done where the Saturation value of the image is extracted from HSV image which is called feature extraction. The superimposition is done to enhance the contrast and for good visualisation of the affected area. The eccentricity, entropy, variance and standard deviation of the image are found. Classification of different stages of cancer through neural network algorithm is done. Finally spectrum of the image is found using Fast Fourier Transform(FFT) .

RESULTS
MODERATE STAGE OF IMAGE 1
Input image
Ycbcr image
Contrast Enhanced Output

HSV Image Intensity Image

S Layer

Segmented Output

Super Imposed Input and Output Images

12

x 10

Spectrum of the Image

10

6 x 10

7
4

Super Imposed Input and Output Images

MODERATE STAGES OF IMAGE 1

12 x 10
6

Spectrum of the Image

10

6 x 10

7
4

SEVERE STAGE OF IMAGE 2

Super Imposed Input and Output Images
10 9 8 7 6 5 4 3 2 1 0 x 10
6

Spectrum of the Image

6 x 10

7
4

MODERATE STAGE OF IMAGE 3 AND 4

Super Imposed Input and Output Images

14

x 10

Spectrum of the Image

12

10

6 x 10

7
4

12

x 10

Spectrum of the Image

Super Imposed Input and Output Images

10

6 x 10

7
4

Super Imposed Input and Output Images

MODERATE STAGES OF IMAGE 5 AND 6

12 10

x 10

Spectrum of the Image

Super Imposed Input and Output Images

0 0
6

6 x 10

7
4

9 8 7 6 5 4 3 2 1 0

x 10

Spectrum of the Image

6 x 10

7
4

SEVERE STAGES OF IMAGE 7 AND 8

Super Imposed Input and Output Images
14 x 10
6

Spectrum of the Image

12

10

12

x 10

60

Spectrum of the Image

6 x 10

7
4

Super Imposed Input and Output Images

10

6 x 10

7
4

SEVERE STAGES OF IMAGE 9AND 10 Super Imposed Input and Output Images
10 9 8 7 6 5 4 3 2 1 0 x 10
6

Spectrum of the Image

Super Imposed Input and Output Images

12

x 10

3 4 Spectrum of the Image

6 x 10

7
4

10

6 x 10

7
4

FEATURE EXTRACTION OF 20 COLPOSCOPIC IMAGES

FEATURE EXTRACTION IMAGES
AREA ECCENTRICIT Y 0.1236 0.9891 0.9857 0.9869 0.9860 0.9971 0.9951 0.9942 0.9985 0.9973 ENTROPY VARIANCE STANDARD DEVIATION 3.2298e-004 2.8884e-004 2.8149e-004 2.6861e-004 2.4397e-004 3.1399e-004 2.8718e-004 2.9530e-004 2.9041e-004 2.9062e-004 STATUS

1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

6.3950 2.2618 6.1980 4.9309 4.4410 0.9789 4.4676 5.7816 2.5991 3.6105

5.8179 7.0688 5.4869 6.1892 5 .7811 7.2717 6.8751 6.4042 7.1415 7.1342

5.4869 6.1892 5 .7811 7.2717 5.9521e-008 9.8587e-008 8.2474e-008 8.7204e-008 8.4339e-008 8.4457e-008

Moderate severe Moderate Moderate Moderate Moderate severe severe severe severe

FEATURE EXTRACTION IMAGES

AREA ECCENTRICIT Y 0.9862 0.9811 0.9860 0.9937 0.9948 0.9984 0.9945 0.9696 0.9682 0.9971 ENTROPY VARIANCE STANDARD DEVIATION 3.0284e-004 2.8390e-004 3.0607e-004 3.1361e-004 3.2898e-004 3.0129e-004 2.9939e-004 2.8482e-004 2.9658e-004 2.8334e-004 STATUS

11. 12. 13. 14. 15. 16. 17. 18. 19. 20.

1.4516 5.3914 0.7409 1.5498 1.1409 2.6985 5.6844 5.6588 5.5119 2.1448

7.1872 6.3335 7.0257 7.4317 7.6235 7.3746 6.6000 6.1523 6.6118 6.9434

9.1709e-008 8.0598e-008 9.3677e-008 9.8353e-008 1.0823e-007 9.0776e-008 8.9632e-008 8.1120e-008 8.7962e-008 8.0280e-008

Moderate severe Moderate Moderate Moderate Moderate severe severe severe severe

CORRELATION ANALYSIS FOR MODERATE CASES

Variables

Correlation r value -0.5 -0.92662 0.778732 -0.54037 -0.91274

Area Vs eccentricity Area Vs entropy Area Vs variance Eccentricity Vs variance Entropy Vs variance

CORRELATION ANALYSIS FOR SEVERE CASES

Variables Correlation rvalue -0.52999 -0.84676 0.65301

Area Vs eccentricity Area Vs entropy Eccentricity Vs entropy

INFERENCE
Spectrum of moderate stage of cancer
12 x 10
6

Spectrum of severe stage of cancer

10 9 x 10
6

Spectrum of the Image

Spectrum of the Image

10

8 7

8
6

5 4

3 2

2
1

6 x 10

7
4

6 x 10

7
4

Spectrum Value: 445.0000 616.4828 206.2242 568.3506 534.0374

Spectrum Value: 0.0570 2.3865 1.8791 3.5600 4.2452

Using FFT algorithm the spectrum of the image is found out for both moderate and severe stages of cancer. By comparing the spectrum values of both the cases, it can be inferred that the values for moderate case is much higher than the values of severe case.

CONCLUSIO N

The goal is to automatically locate the abnormal regions within the image, 20 subsequent automated classification of precancerous regions has the potential to detect precancerous lesions in cervical cancer more effectively. The segmentation procedure and algorithmic ability discussed in this thesis to segment abnormal regions and extract characteristic features are significant steps in the process of developing a fully automated diagnostic tool for cervical cancer. The ways of illumination and contrast levels adjusting are visually effective and provide better platform for subsequent processing steps. The fast segmentation algorithm developed in this thesis greatly reduced the computation time in segmenting cancerous regions to 12.26 seconds. The statistical results demonstrate the correlation values. More clinical validation studies and analysis of uncertainties involved are currently being investigated.

LIMITATIONS

Few features are only taken into account. It is also not a diagnostic tool as it is just used to calculate the stages of cancer. In depth study of all the different type of stages inside a normal or severe stage has not been done.

FUTURE WORK
Automated methods of segmentation and classification of some of the major precancerous cervix lesions have been presented. Currently these lesions are evaluated manually by physicians, involving time-consuming and subjectively variable methods.However, the success of such an automated method is highly dependent on the quality of the cervix images acquired by cervicographic or colposcopic devices and on the ability of the image analysis algorithms to extract the most relevant features for segmentation and classification of various lesions associated with cervical cancer. Besides cancerous regions discussed in this thesis, there are other biomarkers which need to be segmented and more features to be extracted in the future to make the whole analysis process more efficient. Our final goal is to develop a fully automated and efficient diagnostic screening tool for early detection of cervical cancer.

REFERENCES
Automated and Interactive Lesion Detection and Segmentation in Uterine Cervix Images(IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 29, NO. 2, FEBRUARY 2010) Domain-Specific Image Analysis for Cervical Neoplasia Detection Based on Conditional Random Fields(IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 30, NO. 3, MARCH 2011) Graph cuts and efficient N-D image segmentation by y.boykov and G.funka-lea(2006) Shape matching and object recognition using shape contexts (IEEE APRIL 2002) A computational approach to edge detection(IEEE NOVEMBER 1986) Human Papillomavirus (HPV) Vaccines: Q & A - National Cancer Institute". http://www.cancer.gov/cancertopics/factsheet/risk/HPV-vaccine. Retrieved 2008-07-18. Canavan TP, Doshi NR (2000). "Cervical cancer". Study on solder joint image segmentation techniques based on matlab(ieee paper, 03 november 2011) The usage of neural networks for the medical diagnosis("information science and computing" by kateryna malyshevska) Medeiros, LR; Rosa, DD, da Rosa, MI, Bozzetti, MC, Zanini, RR (2009 Oct). "Efficacy of human papillomavirus vaccines: a systematic quantitative review.". International journal of gynecological cancer official journal of the International Gynecological Cancer Society. Shiri Gordon, Gali Zimmerman, Hayit Greenspan, Image Segmentation of Uterine Cervix Images for Indexing in PACS, Proceedings of the 17th IEEE Symposium on Computer-Based Medical System, 2004.