AI-Enhanced Endometrial Cancer Diagnosis System

Endometrial cancer is the most prevalent type of uterine cancer, and early diagnosis is important for effective treatment. Traditional methods like biopsy and histopathological examination, are often time-consuming and prone to human error. Accordingly, this study proposes, implements and evaluates an Artificial Intelligence (AI)-based prototype utilizing deep learning algorithms for endometrial cancer diagnosis, which would increase the accuracy and the speed of cancer detection by acting as an assistant to the physicians. The proposed system leverages advanced object detection models, making use of multiple You Only Look Once (YOLO) versions, trained on a comprehensive dataset of hysteroscopy images annotated by physicians. Specialized AI models demonstrate high precision and recall rates, significantly enhancing diagnostic capabilities. Moreover, the developed system features a user-friendly interface that enable clinicians to easily upload images and employ existing trained models to detect endometrial abnormalities. After training the models, the results show that the YOLOv9c model achieves the highest mAP of 0.906 at IoU=0.5 and the highest precision of 0.894, and the YOLOv8s model achieves the best recall of 0.906. Also, expert evaluations confirm t he system's reliability, noting accuracy of models ranging from 94% to 98% and marking a substantial advancement in endometrial cancer diagnostics.