Background <p>Charcot neuroarthropathy (CN) is a rare but severe complication of diabetes mellitus, characterized by progressive bone and joint destruction that may lead to foot deformity and limb loss. Accurate anatomical classification using Brodsky’s system is essential for clinical decision-making but often depends on specialist interpretation, which may not be accessible in all healthcare settings. Artificial intelligence (AI), particularly deep learning, has emerged as a promising tool to support clinicians by automating the detection and classification of complex radiological patterns.</p> Methods <p>Neural network (NN) models with varying architectures were developed using a retrospective dataset comprising 234 X-rays from consecutive patients diagnosed with and without CN who received treatment at the Foot and Ankle Department of a tertiary care institute. All X-rays were retrieved from the hospital’s PACS and annotated under the supervision of a senior foot and ankle consultant. The algorithm was trained to automatically detect the affected region and classify CN according to the Brodsky anatomical classification. The resulting algorithm was then validated on two sets of unseen datasets of 23 X-rays (internal test) and 15 X-rays (external test). The reference standard for classification was the assessment made by a senior foot and ankle surgeon, and model performance was evaluated against this reference standard.</p> Results <p>Among the 23 X-rays included in the internal test, 18 had Charcot neuroarthropathy (target condition present) and 5 did not have the condition (target condition absent). The classification model achieved an accuracy of 93.39% on unseen data, while the segmentation model demonstrated 96.85% accuracy in both validation and testing. The AI predictions showed 91.3% agreement with expert-assigned Brodsky classifications, successfully identifying and localizing affected anatomical regions across diverse cases. For external validation, the model was evaluated on an independent dataset comprising 15 X-rays, of which 14 were correctly classified, corresponding to a 93.3% agreement with expert labels.</p> Conclusion <p>The proposed AI system offers a reliable and accurate method for diagnosing and classifying CN from radiographs, even in resource-limited environments. By recognizing global versus localized destruction patterns, it can help distinguish CN from conditions such as osteomyelitis, supporting earlier diagnosis and improved clinical outcomes. However, given the limited dataset, the current model should be regarded as a feasibility study, and larger multicenter datasets with increased number of observers for validation are needed to enhance its generalizability and clinical&#xa0;applicability.</p>