Whole Slide Image Analysis and Detection of Prostate Cancer using Vision Transformers
Kobiljon Ikromjanov, Subrata Bhattacharjee, Yeong-Byn Hwang, Rashadul Islam Sumon, Hee‐Cheol Kim, Heung‐Kook Choi
Abstract
Prostate cancer (PCa) is the most frequently diagnosed non-skin malignancy in men and the second leading cause of fatality from cancer. The most prognostic marker for PCa is the Gleason grading system on histopathology images. Pathologists examine the Gleason grade on stained tissue specimens of Hematoxylin and Eosin (H&E) based on tumor structural growth patterns from whole slide image (WSI). According to the Gleason grading system, prostate cancers are scaled into five grades based on glandular patterns of differentiation. It varies from grade 1 (normal tumor) to grade 5 (abnormal tumor). Cancer cells that look similar to healthy cells receive a low score. Recent developments in Computer-Aided Detection (CAD) using Artificial Intelligence (AI), mainly Deep learning (DL) have brought the immense scope of automatic detection and recognition at better accuracy in adenocarcinoma like other medical diagnoses. Automated DL systems have delivered promising results from histopathological images to accurate grading of prostatic adenocarcinoma. This study aims to classify multiple patterns of images extracted from the WSI of a prostate biopsy based on the Gleason grading system. First, extract patches from the detected region of interest (ROI), then applying Vision Transformers (ViT) model for classification. Finally, the classified patches are scored and graded. The proposed deep learning model in this research will be able to assist the pathologist and other researchers to identify and treat of prostate cancer.