Rapid Detection and Diagnosis of Patients with Plantar Fasciitis Based on Integrated YOLOv12n and ResNet34 Framework Using Magnetic Resonance Imaging.

Background: Plantar fasciitis (PF) is the primary cause of heel pain. We aimed to develop a fully automated, computationally efficient deep learning-based system for the PF identification using magnetic resonance imaging (MRI) images. Methods: A dataset of MRI images from 123 PF patients and 150 controls was collected. Data augmentation methods were applied during training. Four YOLO algorithms (YOLOv8n, YOLOv11n, YOLOv12n, and YOLOv13n) were applied to train object detection models for locating relevant anatomical structures in MRI images. The convolutional neural network, ResNet14, ResNet18, ResNet34, and ResNet50 were used for classification model construction. The optimal models were integrated to form an intelligent diagnostic pipeline. Results: For object detection models, YOLOv12n model presented the best performance, achieving a mAP50 of 0.907. The YOLOv13n, YOLOv11n and YOLOv8n models achieved mAP50 of 0.904, 0.896 and 0.887, respectively. For classification models, the ResNet34 model outperformed the others with the highest accuracy of 0.9740. Then, YOLOv12n model, as the object detection model, and ResNet34 model, as the classification model, were integrated to construct the intelligent diagnostic process for the automatic identification of PF. Conclusion: In this study, we innovatively propose an automatic detection process integrating YOLOv12n and ResNet34 to efficiently and automatically identify PF, which demonstrates high potential for streamlining the diagnostic workflow and supporting clinical decision-making. However, the single-center nature of the dataset warrants further external validation in multi-center cohorts to confirm the generalizability of our model.