AI-based diagnosis of hallux valgus: evaluation of measurements derived from foot X-ray images

Objective: This study aimed to diagnose hallux valgus (HV) using machine learning algorithms (MLAs) applied to angular measurements obtained from foot X-ray images. Materials and Methods: The study was a retrospective observational investigation. The sample comprised 50 patients diagnosed with HV and 50 controls. The intermetatarsal angle (IMA), interphalangeal angle (IPA), talometatarsal angle (TMTA), metatarsal tangent angle (MTTA), metatarsus adductus angle (MTAA), and metatarsal parabola angle (MTPA) were measured on the anteroposterior X-ray image, and calcaneal inclination angle (CIA) and the lateral talocalcaneal angle (LTCA) were measured on the lateral X-ray image. The following algorithms were used as machine learning algorithms (MLAs): Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting (GB), XGBoost (XGB), LightGBM (LGBM), AdaBoost (ADA), Extra Trees (ET), K-Nearest Neighbor (KNN), and Linear Discriminant Analysis (LDA). Statistical analyses were performed using the open-source RStudio software. Results: The diagnostic accuracy rates of the algorithms (CV) were 90% for XGB, ET, RF, LGBM, and LDA; 85% for ADA, LR, SVM, and GB; and 80% for KNN. The mean ± SD values of the variables for the control and patient groups, respectively, were calculated as 21.71 ± 1.83 and 22.52 ± 3.32 for IMA; 17.86 ± 1.79 and 18.35 ± 2.28 for MTAA; 143.43 ± 9.20 and 143.25 ± 9.14 for MTPA; and 50.96 ± 5.27 and 49.42 ± 6.98 for LTCA. The median (min–max) values were: for IPA, 36.48 (25.73–45.82) and 45.41 (29.60–76.90); for TMTA, 7.60 (5.91–11.45) and 12.54 (6.62–19.50); for MTTA, 7.93 (1.07–13.20) and 7.13 (2.60–12.07); and for CIA, 28.52 (22.31–37.65) and 27.19 (10.85–47.46). Conclusions: Angular measurements derived from X-ray images enabled MLAs to diagnose HV with an accuracy of up to 90%. This finding demonstrates the applicability of artificial intelligence for diagnostic purposes in imaging-based applications.