A Computer-Assisted Pap Smear Screening System Based on HSV Color Segmentation

• Main Authors: Hsu,Nai-Jen, 許乃仁
• Other Authors: Fang-Hsuan Cheng
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/07692527845147081710

碩士 === 中華大學 === 資訊工程學系碩士班 === 102 === Malignant tumor, also known as carcinoma, is the first of top 10 causes of death. In which the cervical carcinoma is the top 5 common cancer of women. With the popularize of Pap test, the rank of cervical carcinoma have a declining trend. The prevention of cervical carcinoma depends on early detection and treatment. Pap test is the most effective method for early screening. For many years, our country had promoted the free Pap test for the women who over the age of 30. According to the statistics from Ministry of Health and Welfare, the screen rate of 30+ women has increased from 35% to 56.2%. With the increase of the screening rate, there were more and more workload to the doctors and medical staff. Subjective view, heavy duty and overworked were causing the mistakes of the screening. Therefore, the automatic computer-aided system for Pap test is the new trend for solve it. In this study, we used the Bethesda system, a system for reporting cervical or vaginal cytological diagnoses, as the basis of screening, the image processing, and computer vision method to retrieve the feature of abnormal cells. At first, we segment the smear image into nucleus and cytoplasm in HSV color space and calculate the global nuclear-cytoplasm ratio. Next, we find the contour of nuclei by expansion and erosion methods in morphological. We also record the deformation features in this step. Finally, we estimate the Syncytium-like cell from area features and the Hyperchromasia cell from color characteristics. Combine all of these features, we mark the tumor location with color circle and block as a reference for doctors and medical staff.   The result shows that the Accuracy of our system is 0.975, Sensitivity is 0.974 and the Specificity is 1 of screening the image into normal and abnormal. Furthermore, the classification Accuracy of Normal is 1, LSIL is 0.7 and HSIL or Cancer is 0.72. From the result, this method could screen the tumor cell and support the staffs in their work.