Page 112 - Ebook-Book JCMS 2025

AI Machine Learning Assistances for Detection and Severity
Grading of Hallux Valgus
Arphakorn Wellchan, Chayanin Angthong*
Faculty of Medicine, King Mongkut’s Institute of Technology Ladkrabang
*Corresponding Author E-mail: : chayanin.an@kmitl.ac.th
Background: Objectives: Abstract
Hallux Valgus, commonly known as bunions, is a widespread foot deformity. Early detection
can lead to early treatment with better outcomes than delayed treatments. However, little is
known about the benefits of image analysis using artificial intelligence (AI) for the detection
and classification of this disability.
To develop the AI models for visual detection and classification of Hallux Valgus via foot
clinical image, and evaluate their performances.
Methods: Results: AI Roboflow and Teachable Machine platforms are used to develop machine learning
models, trained with 285 images from an open-access database. Roboflow data are divided
into three groups:train set (70%), valid set (20%) and test set (10%). Teachable Machine
divides the data into two groups: train set (85%) and test set (15%). Each foot image was
labeled as none or mild or moderate or severe degree of Hallux Valgus deformity based
on the Manchester scale using an expert (a fellowship-trained foot and ankle surgeon)’s
supervision. The models’ performances are then tested to assess their accuracy values.
Roboflow demonstrates strong performance in detecting mild Hallux Valgus deformity,
achieving 78% of accuracy on the test dataset. In contrast, Teachable Machine excels in
identifying moderate-severe Hallux Valgus deformity, with accuracies of 69-86%.
Conclusion: Both AI models currently show promising potential for aiding in Hallux Valgus screening
within clinical settings. Roboflow shows superior performance for mild deformity
detection. On the other hand, Teachable Machine demonstrates superior performances for
moderate-severe deformity detection. The combined use of these AI models may assist
patients and non-expert physicians in the early detection of Hallux Valgus at its initial to
severe stages, improving the chances of timely intervention and better treatment outcomes.
110 Joint Conference in Medical Sciences 2025