ADRAN: Adaptive Deep Residual Aggregation Network-based Cervical Region Segmentation in Cervix Images

A colposcopy is a crucial imaging method for detecting abnormalities in the cervical region. Clinical applications benefit from the examination of colposcopic images, especially when it comes to precise cervical place segmentation. To segment the cervical region efficiently, specular reflections need to be removed first. Because acetic acid causes precancerous and cancerous areas to turn white and is able to be used as an indicator of abnormalities, it can be used to detect cervical cancer. Furthermore, reflections from specular sources ought to be eliminated because they hinder the detection of Aceto white areas. Therefore, it is suggested in this work to eliminate reflections from specular sources while segmenting the cervix area of a cervix image. Pre-processing as well segment are the two key phases of the suggested methodology. The thresholding technique is used to identify and get rid of reflections from specular sources during the pre-processing phase. After that, cervical regions are segmented using an adaptive deep residual aggregation network. The deep residual aggregation network parameter is appropriately chosen for segmentation accuracy using the ERO algorithm. The proposed method's performance is evaluated using precision, sensitivity, particularity, Jaccard index, and dice coefficients, and its effectiveness is contrasted with that of alternative processes.