BCL-YOLO: A YOLO-Based Model for Medical Image Detection
Abstract
Medical image detection requires accurate localization of subtle lesions while maintaining efficient inference. Methods:We propose BCL-YOLO, a YOLOv8-based detector that integrates the BC2fB attention module,the C2fIRB inverted-residual module, LDSPPF with linear deformable convolution, and the Multi-InterpIoU loss. Results: On the Br35H brain-tumour MRI dataset, BCL-YOLO achieved 97.2% mAP@0.5 and 74.6% mAP@0.5:0.95, improving mAP@0.5 by 3.4 percentage points over YOLOv8. Ablation and comparative experiments confirmed the contribution of each component and showed competitive precision, recall, efficiency, and cross-dataset robustness on ChestX-Det10. Conclusions:BCL-YOLO improves medical lesion detection by combining attention-guided feature extraction, multi-scale fusion, and efficient bounding-box regression.