Application of YOLOv12 for Diabetic Foot Ulcer Detection in Clinical Images

• M. Rivas Díaz ^[1] ; A. Oliver ^[1] ; J. Vehí ^[1] ; X. Lladó ^[1] ; R. Martí ^[1]
  1. [1] Universitat de Girona

     Universitat de Girona

     Gerona, España

     
• Localización: CASEIB 2025. Libro de Actas del XLIII Congreso Anual de la Sociedad Española de Ingeniería Biomédica: Innovando en tecnología para la salud y calidad de vida / David Hernando (ed. lit.), Jesús Lazaro Plaza (ed. lit.), Pablo Laguna Lasaosa (aut.) , Juan Pablo Martínez Cortés (aut.) , 2025, ISBN 978-84-09-80259-3, págs. 700-703
• Idioma: inglés
• Texto completo no disponible (Saber más ...)
• Resumen

  • Diabetic foot ulcers (DFUs) are among the most serious and costly complications of diabetes. Accurate and reliable detection in clinical images can support wound assessment and monitoring. Although YOLOv12 has shown strong performance in medical imaging tasks, its effectiveness for DFU detection remains untested. In this paper, we present a comprehensive study of YOLOv12n for DFU detection using the DFUC2020 dataset. We perform cross-validation over five random splits to assess the impact of different preprocessing methods, data augmentation techniques, and multi-label training based on ulcer size. Our best-performing configuration (strong data augmentation, multi-label training based on ulcer size, and no preprocessing) achieved a mAP@0.5 of 73.2% on the DFUC2020 test set. These results indicate that a lightweight YOLOv12n-based model provides competitive performance and practical utility for automated wound quantification and follow-up, with the advantage of real-time inference.