Lesion Detection in Cervigrams based on Boundary Cues

In this paper, I have focused on the extraction and segmentation of a specific tissue within the cervix, known as the AcetoWhite(AW) tissue that turns white after the application of 3% to 5% of acetic acid during cervicography. The AW is a major indicator of cervical cancer. This paper presents an automated approach to detect and segment AW lesions by learning class-specific boundaries. The approach is a multi-step scheme that includes the watershed transform that converts the image into an edge map[10] that contains the lesion boundary and viewing the watershed map as a Markov random field (MRF), in which each watershed superpixel corresponds to a binary random variable indicating whether the superpixel is part of the lesion. By applying a belief-propagation (BP) algorithm on the loopy MRF, the final lesion region segmentation is obtained.