Enhancing Bone Tumor X-Ray Classification with Hybrid Augmentation Based on YOLOv8

Le résumé fourni par la source

Bone tumors such as osteosarcoma and chondrosarcoma are rare but malignant tumors that require immediate and accurate diagnosis for effective treatment. Radiographic imaging is the first-line modality for assessing bone tumors due to its accessibility and cost-effectiveness; however, the diagnostic process remains challenging because of low contrast, uneven illumination, and overlapping anatomical structures. In this work, we propose a YOLOv8-based classification framework designed explicitly for bone tumor X-rays, enhanced with a hybrid augmentation pipeline that combines Contrast Limited Adaptive Histogram Equalization (CLAHE), wavelet denoising, and unsharp masking. We conduct extensive experiments on the BTXRD dataset and evaluate multiple augmentation scenarios, comparing the proposed framework with state-of-the-art CNN and Transformer-based models. Results show that our hybrid pipeline improves classification performance over the YOLOv8 baseline, achieving the highest accuracy of 88.6 % and superior F1-scores, with notable gains in malignant tumor recall. These findings demonstrate that augmentation strategies can play a pivotal role in boosting model robustness and highlight the potential of YOLOv8 with hybrid preprocessing as a reliable tool for supporting radiologists in bone tumor diagnosis. The implementation of our framework is publicly available at https://github.com/virgantara/AugYolo.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.