Performance Analysis of Dysplasia Diagnosis for Cervical Cancer
B. Shanmuga Priya, Sathis Kumar · International Journal For Science Technology And Engineering · 2016
Cancer cervix is the second most common cancer in women in the world, while it is the leading cancer in women in the developing countries. Globally, 15% of all cancersin females are cervical cancers', while in Southeast Asia, cancer cervix accounts for 20%-30% of all cancers'. Cancer of cervix is a major cause of death in women living in developing countries. Unlike most other malignancies, cancer of cervix is readily preventable when effective programs are conducted to detect and treat its precursor lesions. Since the introduction of Pap test, a dramatic reduction has been observed in the incidence and mortality of invasive cervical cancer worldwide. Conventional methods used Multimodal Entity Co-reference for combining various tests to perform disease classification and diagnosis. Its performance is degraded due to its low sensitivity and specificity. In this project, we propose watershed segmentation algorithm to segment the abnormal region and this abnormal regions are classified into normal or cancer using feed forward neural network classifier. The performance of the proposed system is analyzed in terms of sensitivity, specificity and accuracy. CANCER CERVIX death rate is reduced by combination of several screening and diagnostic procedures in western countries. And also due to the lack in screening methods and diagnostic procedures there is a need, for automated screening methods are increased. In Digital cervicography, the photographs of cervix is taken by cerviscope as a diagnostic device. The cervix is visualized with a vaginal speculum and 5% acetic acid is applied to the cervix. The image of the acetic-acid treated cervix is processed onto film and projected on a white screen for analysis. Due to the poor correlation in digital cervicography, computer- assisted diagnoses are used to detect cervical cancer in early stage. In that computer algorithms are used to detect cervical region. Accuracy in detection is improved by using images of cervigram, the normal clinical Pap test and HPV test images are combined with automated Pap and HPV test images. The effectiveness of the combination of several screening methods is poor in terms of sensitivity and accuracy, so that we further improves, the accuracy by cervigram images taken by scanning. The cervix is visualized with a vaginal speculum and optical rays are passed over that and the treated cervix is projected on a white screen for analysis. In cervix region, the region of interest is detected by Local Binary Pattern and Gray scale Co-occurrence matrix, after that image is classified. The image classification is done by color features and also Local Binary Pattern and Gray scale Co- occurrence matrix are used to classify different cervix regions. After that region classification texture features are used to classify vascular patterns in cervix region. Totally classifier is used to classify the abnormal cervix regions from normal cervix region. In that Feed Forward Neural Network classifier that classifies the cervigram images. To classify the vascular tissues and neural networks in cervix image this classifier is efficient. The classified cervix image, is further segmented by segmentation algorithm. Water Shed Segmentation algorithm is efficient for Neural networks. This algorithm segments the cancer regions in the cervical image. The image analysis result is combined with physician marked image to evaluate the performance of disease classification. International Federation of Gynecology Obstetrics is the system that staging the cervical cancer. The stages are Normal, Early, Late and Final stage. We develop a computerized program that taken these factors in to consideration for the treatment that should be much easier.