Cervical Cancer Screening in Pap Smear Images Using Improved Distance Regularized Level Sets

Pap smear test was introduced in 1940 and has proved to be an effective method for cervical cancer screening. Manual interpretation of pap smear images is a challenging task and is prone to a lot of human induced errors. It also consumes considerable amount of time. To overcome all these problems, an automatic segmentation and classification algorithm is proposed in this work. Pre-processing is achieved by enhancing the contrast of the image and smoothening the image using a median filter. Small regions on the nucleus are also detected as initial points for segmentation in the next step. The nucleus from the images is segmented using distance regularized level sets evolution. After segmentation, 22 texture features of the images are calculated using gray level co-occurrence matrix (GLCM). 7 classes of cervical cancer are considered for this work. After segmentation of nucleus and feature extraction, the feature matrix is passed through a neural network to classify the images into their respective category. The neural network is made using the backpropagation algorithm with 2 hidden layers. The accuracy achieved after segmentation, feature extraction and classification is 92 % which is significant for clinical interpretation.