A Collaborative Hybrid network with Aggregated Attention for Pulmonary Nodule Detection

Lung cancer remains one of the most lethal malignancies worldwide, underscoring the critical importance of early detection for improving patient survival. Current imaging-based methods for pulmonary nodule detection, however, are often limited by low efficiency and suboptimal accuracy, falling short of the demands for precise clinical diagnosis. While deep learning has markedly advanced medical image analysis, contemporary detection models still exhibit notable shortcomings—including inadequate recognition capability, high rates of missed detection, and elevated false positives—particularly when dealing with small and morphologically complex nodules.To tackle these issues, this paper proposes a collaborative hybrid network with aggregated attention for pulmonary nodule detection, termed YOLOv8-RTA. The model integrates an RT-DETR-based Transformer decoder head into the YOLOv8 architecture, replacing its original detection head. This hybrid design mitigates the suppression of true bounding boxes caused by non-maximum suppression (NMS) and significantly reduces missed detections. Furthermore, an aggregated attention mechanism is introduced within the YOLOv8 backbone to expand the effective receptive field and enhance the discriminative power of deep features, thereby improving the representation and classification of tiny nodules.Experimental results on the Tianchi dataset demonstrate that YOLOv8-RTA achieves a mean average precision (mAP) of 94.98%, substantially outperforming several baseline models. Additional validation using real-world CT images from a hospital PACS system confirms the model’s robustness and stability in clinical settings, highlighting its strong potential for practical deployment in early lung cancer screening.