Attention-Driven Enhancements in YOLOv8s for Improved Detection of Respiratory Conditions From HRCT Scans

Respiratory diseases, including COVID-19 and related pulmonary conditions, often manifest with complex lung abnormalities such as ground-glass opacities, consolidations, and crazypaving patterns, which pose significant diagnostic challenges. This study presents a novel automated detection framework that integrates the Convolutional Block Attention Module (CBAM) into YOLOv8s, a state-of-the-art object detection algorithm, to enhance the analysis of high-resolution computed tomography images. The incorporation of CBAM enables YOLOv8s to refine feature extraction and classification processes, ensuring precise localization of subtle and overlapping lesions that are typically difficult to discern. Comparative analysis against YOLOv5 and YOLOv7 demonstrates the superiority of this approach. The CBAM-augmented YOLOv8s achieves a mean average precision (mAP) of 65.02% and [email protected]:0.95 of 48%, significantly outperforming the standalone YOLOv8s baseline, which achieved an mAP of 41%. The research highlights the transformative potential of attention mechanisms in medical imaging, offering an innovative and robust solution for improving diagnostic accuracy and efficiency in respiratory disease management. By addressing key challenges in lesion detection, this approach has the potential to advance clinical workflows and improve patient outcomes.