Development of Oral Cancer Segmentation Framework with Histopathological Images and Dilated Trans- Mobile-Unet++

In order to detect Oral Squamous Cell Carcinoma (OSCC), a pathologist must segment the tumor regions on HandE-stained slides. Since labeling histological images is a highly skilled, intricate, and time-consuming operation, the availability of labeled training data affects histopathological image segmentation. Accurate segmentation of microvessels and nerves is needed for an effective oral cancer diagnosis process. Thus, an efficient oral cancer segmentation framework is designed with deep learning techniques to tackle the limitations of the existing mechanism. Several phases presented in the developed oral cancer segmentation framework are (a) Image Collection, (b) Image Denoising, and (c) Image Segmentation. At first, histopathological images needed for the experiment are accumulated from the benchmark sites. Next, the attained images are provided for the image denoising phase. Here, median filtering and Contrast Limited Adaptive Histogram Equalization (CLAHE) mechanisms are employed to perform denoising in the collected images. Next, the denoised images are given to the image segmentation region. In this stage, effective segmentation is performed using the Dilated Trans- Mobile-Unet++ (DTM-Unet++) model. Later, different experimental analyses were carried out to validate the effectualness rate of the developed system.