Explainable AI-based clinical decision support system for flatfoot classification using deep learning.

INTRODUCTION: Flatfoot, also known as pes planus, is a deformity of the foot characterized by a decrease or absence of the medial longitudinal arch, which may lead to postural and locomotion abnormalities. METHODS: This study presents a comparative analysis of handcrafted feature-based machine learning (LBP with Random Forest, Decision Tree, Logistic Regression) and deep learning models (InceptionResNetV2, ResNet101V2, DenseNet201, DenseNet169, InceptionV3, Xception) for flatfoot classification. Monte Carlo cross-validation was employed with subject-wise splitting. SHAP and Grad-CAM were used for explainability. RESULTS: Random Forest achieved 70% accuracy with LBP features. ResNet101V2-RMSprop achieved 92.86% ± 2.82% mean accuracy and 98.21% best single-run accuracy. Deep features with Decision Tree achieved 97% accuracy, 95% recall, and 100% specificity. DISCUSSION: SHAP and Grad-CAM confirmed that the model focuses on clinically relevant regions: the medial longitudinal arch and calcaneus for pes planus, and the talus-navicular region for normal feet. The hybrid approach is suitable for clinical screening.