Radiographic Angle-Based Machine Learning Models for the Diagnosis of Pes Planus and Pes Cavus: A Large-Scale Study...

Area of Science:

• Orthopedics and Sports Medicine
• Radiology
• Biomedical Engineering

Background:

• Pes planus and pes cavus are common foot deformities impacting biomechanics and causing pain.
• Traditional diagnostic angles (calcaneal pitch, talar declination, Meary) lack a gold standard and are prone to measurement error.
• Objective and reliable methods are needed for accurate diagnosis of foot arch deformities.

Purpose of the Study:

• To evaluate machine learning algorithms for diagnosing pes planus and pes cavus.
• To determine the most accurate predictive model using radiographic foot angles.
• To automate and standardize the classification of foot arch deformities based on established angle thresholds.

Main Methods:

• Retrospective analysis of 697 male patients' weight-bearing lateral foot radiographs.
• Measurement of calcaneal pitch, Meary, and talar declination angles.
• Application and comparison of Random Forest, XGBoost, Logistic Regression, SVM, and KNN algorithms.

Main Results:

• XGBoost achieved perfect (1.000) and near-perfect (0.996-1.000) accuracy for left and right feet, respectively.
• Random Forest also showed high performance (0.986-1.000).
• Ensemble methods outperformed traditional models; KNN showed the weakest performance, especially for pes cavus.

Conclusions:

• Machine learning, particularly XGBoost and Random Forest, offers high accuracy for diagnosing foot arch deformities from radiographic angles.
• These models can serve as objective, rapid decision support tools.
• External validation is crucial for clinical generalizability of these ML-based diagnostic approaches.