YOLO26-Based Framework for Automated Liver Cancer Detection and Localization in CT Images

Abstract

Liver cancer remains a major global health challenge due to the difficulty of early and accurate diagnosis from medical images. This study presents a YOLO26-based deep learning framework for automated liver cancer detection using Computed Tomography (CT) scan images. The proposed framework performs simultaneous lesion localization and classification within an end-to-end object detection pipeline. Approximately 12,000 annotated CT images were utilized for model training and evaluation following preprocessing and annotation preparation. Experimental results showed that the model achieved a Precision of 0.9600, Recall of 0.9257, F1-score of 0.9425, and mAP@0.5 of 0.9600, indicating strong detection and localization performance. Qualitative evaluation further demonstrated effective localization of suspicious liver lesions across different CT image samples. Compared with several existing approaches, the proposed framework showed improved localization consistency and enhanced small lesion detection capability. A Gradio-based deployment environment was also implemented to validate real-time inference on unseen CT images. The findings suggest that YOLO26 provides a promising approach for automated liver cancer detection and computer-aided medical imaging applications.