Akademik Veri Yönetim Sistemi
Multimedia Tools and Applications , cilt.83, sa.34, ss.1-18, 2024 (Hakemli Dergi)
Colorectal cancer stands as the world’s second most prevalent cause of cancer-related fatalities. Traditional histopathological image analysis, the primary tool for pathologists to categorize colorectal cancer types, is hindered by its subjectivity and susceptibility to misdiagnosis. To address this challenge, a pioneering computer-aided diagnosis (CAD) approach utilizing supervised contrastive learning is proposed. It is the first application of supervised contrastive learning to colorectal cancer detection. This study explores the efficacy of different encoder architectures within the proposed supervised learning-based method, highlighting variations in performance based on the chosen encoder network. The proposed framework was evaluated on the Enteroscope Biopsy Histopathological H&E Image Dataset (EBHI), the first publicly available dataset for colorectal histopathology enteroscope biopsies. Experimental findings demonstrate the superiority of the proposed model over state-of-the-art methods, with classification accuracy, precision, recall, F1-score, and specificity for colorectal cancers achieving 97.3%, 98.87%, 98.32%, 98.59%, and 98.92% respectively. This model serves as a valuable supplementary diagnostic tool for pathologists, alleviating their workload and helping to formulate precise diagnoses and treatment plans.