A Lung Lesion Detection Algorithm Based on YOLOv7 and Self-Attention Mechanism

In recent years, non-invasive medical imaging techniques such as X-ray have become increasingly important in medical diagnosis. However, traditional lesion detection algorithms on chest X-ray using neural deep learning techniques often have low detection rates. To address this issue, we propose a new chest X-ray lesion detection algorithm based on the YOLOv7 object detection algorithms, which incorporates with self-attention mechanism module to enhance the detection ability and reduce computation time. Our algorithm achieved the best results on the VinDr-CXR dataset of
$640 \times 640$
images, using the YOLOv7-X and Swin Transformer module, with a mAP@0.5 of 41.13% and 17fps detection speed. These results demonstrate that our algorithm achieves high detection accuracy while maintaining fast detection speed, making it a promising approach for chest X-ray lesion detection.