The Use of AI For Hallux Valgus Assessment via Mobile Phone-Based 3D Camera Scan

Category: Bunion; Other Introduction/Purpose: Hallux valgus (HV) is a prevalent forefoot deformity that leads to pain, stiffness, and difficulties with footwear. Conventional diagnosis of HV relies on clinical and radiographic examination, which can be cumbersome and expose patients to ionizing radiation. The emergence of 3D scanning technologies, such as structured light, integrated into smartphones, presents a novel approach to early detection and diagnosis of HV, potentially allowing for earlier intervention and subsequent reduction in the progression of the deformity. Methods: This prospective non-randomized clinical trial recruited patients from the foot and ankle clinic, identifying cases and controls based on the presence or lack of clinically diagnosed HV. Patients older than 18 years of age and able to consent were included. The 3D scanning of the foot utilized structured light technology integrated into smartphones, with an algorithm developing a surrogate angle for HV detection. This angle was correlated with conventional radiographic measurements, including hallux valgus angle (HVA) and intermetatarsal angle (IMA) from weightbearing radiographs. A cluster analysis assessed the algorithm’s capacity to identify deformity severity according to radiographic severity grading. The algorithm's accuracy and performance were assessed using the area under the receiver operating characteristic curve, precision-recall Curves, and the leave-one-out cross-validation method. Results: 120 patients were enrolled in this trial. Out of 240 feet examined, HV was present in 29.1% of cases. The algorithm's surrogate angle showed a robust correlation with the clinical hallux valgus angle (HVA), with a correlation coefficient of 0.91 and a specificity of 0.882. Notably, the surrogate angle's correlation with the intermetatarsal angle (IMA) was also significant, albeit lower (r = 0.65). The area under the ROC curve (AUC) was 0.947, and the precision-recall AUC scores for positive and negative classes were 0.89 and 0.92, respectively. These metrics confirmed the algorithm's potential as a reliable diagnostic tool for detecting HV and determining its severity through non-invasive, commonly available smartphone technology. Conclusion: This study validates the accuracy of a smartphone-based 3D scanning algorithm in detecting and grading HV deformity, providing a low-cost, accessible decision-support tool for physicians and patients. This approach could improve early detection, reduce the need for clinical visits, and potentially expand to remote monitoring of HV and other pathologies, aligning with the growing trend of telemedicine and personalized healthcare.