Yapay zeka kullanarak primeri bilinmeyen kanserlerde primer sistemin tahmini

Sümerkent, K. (2025). Using Artificial Intelligence to Predict the Primary System for Cancers of the Unknown Primary. Yeditepe University, Institute of Health Sciences, Department of Molecular Medicine. Doctorate Thesis. İstanbul. This dissertation presents an AI computational methodology using Whole Slide Images (WSIs) to determine the primary site of Cancers of Unknown Primary (CUP). Addressing CUP's diagnostic challenges where origin is elusive, it builds upon deep learning advancements in medical imaging. The project utilized over 1000 histopathological images from TCGA and ICGC databases representing diverse malignancies. Pre-processing included colour normalisation and patch extraction. A custom convolutional neural network (CNN), optimised via hyperparameter tuning, achieved >90% accuracy, outperforming baseline and transfer learning models in analysing morphological patterns. Validation on 300 cases from Yeditepe University Hospitals confirmed high sensitivity and specificity, particularly for breast and prostate cancers. Error analysis linked diagnostic challenges primarily to ambiguous tissue features or suboptimal image quality. While effective, integrating supplementary molecular and clinical data could further enhance diagnostic precision for CUP management. Implementing this AI-driven analysis in clinical workflows shows promise for improving patient care by facilitating more accurate identification of cancer origins.