A Deep Learning Method for Foot Progression Angle Detection in Plantar Pressure Images.

Area of Science:

• Biomechanics
• Computer Vision
• Medical Imaging

Background:

• Foot progression angle (FPA) analysis is crucial for identifying gait abnormalities like in-toeing and out-toeing, which can lead to foot injuries.
• Deep learning object detection offers a potential method for precise FPA measurement using plantar pressure images.

Purpose of the Study:

• To develop a highly accurate deep learning model for FPA detection.
• To evaluate the efficacy of different You Only Look Once (YOLO) network versions (v3, v4, v5x) for FPA measurement.
• To provide precise kinematic data for assessing physical therapy outcomes in knee pain and osteoarthritis.

Main Methods:

• Analysis of 1424 plantar images using YOLO v3, v4, and v5x deep learning models.
• Comparison of FPA measurements obtained from YOLO models against ground-truth data.
• Evaluation of model performance based on precision, average precision, and statistical significance (p-values).

Main Results:

• YOLOv4 demonstrated superior performance in detecting foot profiles, achieving 100.00% average precision for the left foot and 99.78% for the right foot.
• FPA measurements from YOLOv4 closely matched ground-truth values (5.86 ± 0.09° vs. 5.58 ± 0.10°, p = 0.013).
• YOLOv3 and YOLOv5x showed significant deviations from ground-truth FPA measurements (p < 0.001 for both).

Conclusions:

• Deep learning, specifically using the YOLOv4 model, significantly enhances the precision of foot progression angle detection from plantar images.
• Accurate FPA measurement using YOLOv4 can support the evaluation of physical therapy interventions for knee conditions.
• This technology offers a promising tool for objective gait analysis and injury prevention strategies.