YOLO-LXR: An Enhanced Model for Pathology Detection in Chest X-Rays

Algorithms for object detection, particularly those based on YOLO, have demonstrated impressive effectiveness in achieving a balance between accuracy and speed. Their use in medical imaging has shown promise, but remains underexplored. In this study, a YOLO-like model is proposed that focuses on detecting lung pathologies in X-ray images, is based on YOLOv5, and uses Squeeze-and-Excite and Residual Blocks. It outperforms most YOLO and YOLO-like architectures in terms of accuracy, and compared to YOLOv5, experiments showed an increase in recall and mAP@50 of 3.3% and 1.5%, respectively. The suggested architecture yields encouraging results, establishing itself as a cutting-edge solution for accurate and efficient detection of pathologies in chest X-rays. The code implementation of this project is publicly available at: https://github.com/GKatsagannis/YOLO-LXR.