A deep learning method for foot-type classification using plantar pressure images.

Area of Science:

• Biomechanics
• Medical Imaging
• Artificial Intelligence

Background:

• Flat foot deformity is a common condition with significant clinical implications.
• Accurate identification of foot types is crucial for effective treatment and intervention.
• Plantar pressure imaging offers a valuable tool for foot assessment.

Purpose of the Study:

• To evaluate the efficacy of deep learning models for identifying flat foot types.
• To compare the performance of You Only Look Once v5 (YOLO-v5) and multi-label classification models.
• To introduce an improved deep learning approach for enhanced foot type detection.

Main Methods:

• A dataset of 1573 plantar pressure images from 125 individuals was utilized.
• The You Only Look Once v5 (YOLO-v5) and multi-label classification models were trained and evaluated.
• An improved YOLO-v5 model with attention mechanisms (SE, CBAM) and a ResNet-50 based multi-label model were developed.
• A separate dataset was used for model verification and comparison.

Main Results:

• The ResNet-50 based multi-label classification model achieved the highest accuracy (0.826).
• Improved YOLO-v5 models with SE and CBAM modules showed enhanced performance (0.652 and 0.717, respectively).
• The proposed models significantly outperformed standard YOLO-v5 and ordinary plantar-pressure systems.

Conclusions:

• Deep learning-based multi-label classification using ResNet-50 demonstrates superior performance in detecting flat foot types.
• This model holds potential for assessing the clinical rehabilitation status of patients with abnormal foot conditions.
• Further training with diverse patient data could expand its application to various foot pathologies.