Diagnosis of pes planus from X-ray images: Enhanced feature selection with deep learning and machine learning techniques

Abstract

Pes planus is a foot problem that occurs when the arch of the foot is lower than normal. Techniques such as radiography are actively used for its diagnosis. Artificial intelligence plays an important role in medical diagnoses when combined with traditional methods. It provides a faster and more accurate process in diagnosing diseases with its performance in image analysis. In the study, the use of artificial intelligence in diagnosing a foot deformity called pes planus from x-ray images is discussed. The stages of the study are feature extraction, feature selection and classification. After the feature extraction step, the most important 300 features were selected for each method from the features obtained using the Relief-F, Lasso and RFE methods. The classifier parameters were determined with the Optuna method and the K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Decision Tree (DT), XGBoost and CatBoost algorithms classified the data as normal and pes planus using these features. The results show that the XGBoost algorithm provides the highest performance with 97% accuracy for the features selected with the Relief-F method. This finding indicates a high success rate compared to similar studies. The limited number of studies in the literature on pes planus and artificial intelligence, and the fact that there is only one study that includes artificial intelligence-assisted diagnosis via X-ray images, reveal its innovative aspect. It is expected that artificial intelligence will be an important tool in the early diagnosis of the disease by optimizing the diagnostic processes of the study.