YOLO in Healthcare: A Comprehensive Review of Detection Architectures, Domain Applications, and Future Innovations

The growing adoption of deep learning in medical imaging has propelled object detection frameworks into clinical relevance, with YOLO (You Only Look Once) standing out for its real-time inference and high accuracy. This survey offers a comprehensive review of YOLO-based medical object detection, synthesizing findings from 123 peer-reviewed papers published between 2018 and 2024. Across oncology, radiology, pathology, surgery, and PPE detection, YOLO models have demonstrated strong performance in tumor localization, lesion segmentation, tool tracking, and anomaly detection. YOLOv3 and YOLOv5 remain the most frequently deployed variants, while YOLOv8 and YOLO-NAS are emerging in recent studies with improved mAP and lower latency. Findings reveal that mAP scores exceeding 85% were commonly achieved in breast cancer and pulmonary nodule detection tasks, while lightweight versions such as YOLOv5s maintained detection speed above 50 FPS in surgical environments. Despite its clinical promise, persistent challenges include poor sensitivity to small or rare lesions, absence of benchmark datasets across modalities, and variability in annotation practices. This survey introduces a structured taxonomy of YOLO architectures, benchmarks their comparative performance, highlights key limitations, and proposes forward-looking trends involving multimodal fusion, transformer-enhanced backbones, and privacy-preserving learning. The work consolidates YOLO’s growing role in healthcare AI and serves as a foundational reference for researchers and practitioners alike.