YOLOv8-Based Deep Learning System for Liver Tumor Detection

Abstract

Early and accurate detection of liver tumors remains a major challenge in medical imaging, This study develops a YOLOv8-based deep learning system for automated liver tumor detection using CT images. A total of 16,404 liver CT slices from the Medical Segmentation Decathlon (MSD) dataset were used to train the proposed model, which was trained and evaluated using several metrics. A callback function was employed to monitor and terminate the training. After 50 epochs, the proposed system achieved a precision, Recall, F1-score and mAP of of 0.96, 0.815, and 0.88, @0.5 of 0.884, demonstrating strong detection accuracy across heterogeneous liver textures and tumor sizes. The model also achieved a fast inference speed of 8.50 ms per image with a lightweight 11.4M-parameter architecture, confirming its suitability for real-time or near-real-time deployment. Qualitative outputs further validated accurate tumor localisation with high confidence scores. These results show that the YOLOv8-based system provides reliable, sensitive, and computationally efficient liver tumor detection, making it a practical decision-support tool for healthcare settings. The study contributes to improving early diagnosis and strengthening clinical imaging workflows.