Development of Hallux Valgus Classification Using Digital Foot Images with Machine Learning

Area of Science:

• Orthopedics
• Medical Imaging
• Artificial Intelligence

Background:

• Hallux valgus is a common foot deformity requiring early detection to prevent progression.
• The medical and economic impact of hallux valgus necessitates efficient diagnostic tools.
• Current screening methods may not be rapid or widely accessible.

Purpose of the Study:

• To evaluate the accuracy of a machine learning tool for early hallux valgus screening.
• To compare the performance of two image preprocessing patterns (A and B) for hallux valgus detection.
• To assess the potential of AI in analyzing foot images for deformity identification.

Main Methods:

• A dataset of 507 foot images was utilized for machine learning model training.
• Two image preprocessing patterns, A and B, involving various transformations were applied.
• The VGG16 convolutional neural network was employed to classify foot images.

Main Results:

• Machine learning model with Pattern B preprocessing demonstrated higher accuracy (0.79) than Pattern A (0.62).
• Pattern B achieved superior precision (0.77), recall (0.96), and F1 score (0.86) compared to Pattern A.
• The developed tool showed sufficient accuracy in distinguishing hallux valgus from normal feet.

Conclusions:

• Machine learning analysis of foot images is a viable method for hallux valgus screening.
• Image preprocessing pattern B significantly improved the diagnostic accuracy of the AI tool.
• Further refinement of this AI tool could enable widespread, easy screening for hallux valgus.