Medical computer vision research
YOLOv12 Hand Fracture Detection with Registered IP
A medical-imaging research system developed with researchers and clinical collaborators to detect hand fractures from X-ray imagery.
- reported mAP50
- 98%final YOLOv12 model
- X-ray images
- 2K+research dataset
- registered IP
- HKINo. 001053135
01Problem
Fracture patterns can be subtle and medical image datasets require careful annotation, evaluation, and collaboration with domain specialists.
02Approach
- Prepared and evaluated more than two thousand hand X-ray images with the research team and clinical collaborators.
- Compared candidate detection models before selecting and optimizing YOLOv12.
- Tuned model parameters and assessed the final detector using mAP50.
03Contribution
- Implemented the detection program and model pipeline.
- Optimized training parameters and compared the final approach with alternative models.
- Contributed to the system registered as intellectual property.
04Outcome
- The final YOLOv12 detector reached a reported mAP50 of 98%.
- The resulting computer program received registered copyright protection.
- The project demonstrated an applied collaboration between software, AI research, and clinical expertise.
05Limitations
- This is a research decision-support system and is not presented as an autonomous diagnostic device.
- Clinical deployment would require broader external validation, governance, and regulatory review.