Kolon polipleri için kolonoskopi ve histopatoloji görüntülerinden yapay zekâ destekli prognostik belirteç tespiti

According to World Health Organization statistics for 2023, colorectal cancer is the third most common type of cancer worldwide, accounting for approximately 10% of all cancer cases. Most colon cancer begins with polyps that form as a result of abnormal cell proliferation in the colon mucosa. There are two types of colon polyps: neoplastic and non-neoplastic. Neoplastic polyps have malignant potential. Colonoscopy is the most common method for detecting polyps. It is possible to detect and remove polyps (polypectomy) with the tool at the end of the colonoscopy. Pathologists prepare and examine almost all removed polyps under a microscope using hematoxylin and eosin (H&E)-stained tissue slides. When uncertainty arises, pathologists perform immunohistochemical (IHC) analyses to demonstrate significant expressions of cancer-specific antigens (proteins). This thesis achieved four main outcomes: First, the automatic determination of polyp type or subtype, stage, and malignant potential using colonoscopy videos, images, and frames and using pathology reports and IHC analysis results as labels were investigated. Secondly, the automatic characterization of colon polyps from histopathology images using features obtained from colonoscopy images, pathology reports, and IHC analysis results were examined. Statistical approaches were used to analyze polyp type or subtype, stage, and possible prognostic features (biomarkers) that may indicate malignant potential. Finally, we created a comprehensive database and shared it with the scientific community as an open-source repository. The database contains colonoscopy and histopathology images of more than 400 polyps, along with information on polyp type, location, stage, and IHC analysis results of Ki-67 (clone 30-9), CD34 (clone QBend/10), p53 (clone bp53-11), BRAF (clone V600E), VEGF (clone SP125), and PD-L1 (clone SP142) markers.