Artificial Intelligence and Smartphone Integration Create Improved Diagnostic Access for Foot and Ankle Deformities

Foot and ankle malalignments, particularly in the hindfoot area, present in pediatric and primary care settings, but making the correct diagnosis can be elusive for non-orthopaedic practitioners evaluating pes planus (flat foot) or pes cavus (cavus foot). Radiographs can be a screening tool to help orthopaedic surgeons assess potential patients, but acquisition of weight-bearing films critical for diagnosing foot and ankle deformity are not typically available in non-orthopaedic x-ray suites. Artificial intelligence (AI) may hold promise for improving diagnostic accuracy of these deformities outside of the orthopaedic surgeon exam room.

Soheil Ashkani-Esfahani, MD, MPH, and coauthors presented their paper titled “Utilization of AI in the Diagnosis of Pes Planus and Pes Cavus with a Smartphone Camera” at the AAOS 2025 Annual Meeting and met with AAOS Now to discuss the findings and implications. Dr. Ashkani-Esfahani, assistant professor of orthopaedic surgery at Harvard Medical School, is the director of the Foot & Ankle Research and Innovation Lab (FARIL) at Massachusetts General Hospital and the principal investigator of the project. The lab explores the applications of AI in orthopaedics, including prevention, diagnosis, and treatment decision-making processes for orthopaedic surgeons. Samir Ghandour, MD, was the postdoctoral researcher leading the study, and Christopher DiGiovanni, MD, professor of orthopaedic surgery at Harvard and chair of FARIL, and Lorena Bejarano-Pineda, MD, assistant professor of orthopaedic surgery at Harvard, were senior authors.

“AI is kind of an analytical tool that you can use to interpret large and complex data, sometimes even better than human,” Dr. Ashkani-Esfahani noted. “In this study, we used AI to help improve the accuracy of diagnosis of foot malalignments and deformities through the lens of smartphones. This will be helpful for clinicians to screen the patients for such deformities, even from remote, without the need to obtain further radiological imaging, and that can reduce the side effects of radiation on the patients. Moreover, it will provide a more accessible tool for patients to assess their foot and ankle for possibilities of deformities and to find the root cause of their inconvenience.”

He added that the team undertook this study in order to “leverage AI technology to help patients have a more accessible while accurate diagnosis for their foot and ankle problems, specifically here flat foot and cavus foot.”

The study was a prospective evaluation of 246 patients. Flat foot was present in 70 patients, cavus foot existed in 20, and the remainder were classified as neutral alignment, as diagnosed by an expert foot and ankle clinician. In controversial and subtle conditions, further radiologic imaging was used to confirm the diagnosis. An AI algorithm was developed via deep-learning methods, specifically a convolutional neural network architecture trained to classify foot-arch deformities. The model included several convolutional and pooling layers followed by fully connected layers, optimized to extract visual features from foot images and assign diagnostic labels.

Standard cell-phone cameras were used to capture images of the foot from multiple angles—lateral, medial, posterior, and anterior. These images were used to train the AI model based on the visual characteristics of pes planus, pes cavus, or normal arches. After the initial training, the model’s predictions on new images were compared with surgeon-confirmed diagnoses to assess its accuracy.

Dr. Ashkani-Esfahani noted that the quality of data used for training an AI algorithm is one of the biggest challenges of this technology. “AI can reach near-perfect accuracy in some scenarios, but the more honest reality is that every algorithm still needs time, patience, and rigorous validation,” he said. “Even if it’s trained on a million patients, it may still need two million—the more data an algorithm receives, the more accurate and reliable its outcomes are likely to be.”

The FARIL study had more than 90 percent accuracy in diagnosing cavus or planus deformities, but it is unclear whether the algorithm would perform at the same level in different patient populations. “Variations in skin tone, a higher prevalence of foot and ankle deformities, or even socioeconomic factors could all influence the outcomes,” Dr. Ashkani-Esfahani explained. A high accuracy for an AI algorithm is not enough; expanding the database, improving the algorithms, and continual retraining are needed. Confirmation of reliability and validity in different populations will be necessary before AI-based tools can be used as screening or decision-support tools or potentially be approved as alternative diagnostic technologies.

Given that pes planus and pes cavus often present in the pediatric population, it is a notable limitation that the FARIL study used patients older than age 18 to train the algorithm. Additionally, Dr. Ashkani-Esfahani acknowledged the downside of a misdiagnosis from AI, which could lead to a patient referral for a deformity and potential patient frustration if treatment is expected but not delivered.

At FARIL, Dr. Ashkani-Esfahani and colleagues are expanding the reach of this technology internationally, aiming to bridge the gap in musculoskeletal care across the globe. He highlighted their work in underserved countries where there is a shortage of academically trained orthopaedic surgeons and where access to accurate diagnosis and timely treatment is severely limited. In those settings, AI-powered tools—particularly those accessible through standard smartphone cameras—offer a transformative solution. These tools not only offer value for prevention, early detection, and diagnostic purposes, but they also play a critical role in guiding treatment decision making and tailoring post-treatment rehabilitation plans. By empowering frontline healthcare workers and even patients themselves to gather and interpret clinical data, AI can reduce dependency on specialized infrastructure, support continuity of care, and bring high-quality orthopaedic services to remote or resource-limited communities worldwide.

Still, these initiatives require thoughtful oversight. AI tools cannot simply be deployed in underserved regions with an expectation of flawless operation. Expert supervision is essential during the early stages—not only to ensure the safety and reliability of clinical decisions but also to build trust among local providers. Even once an AI system has proven its effectiveness, ongoing oversight remains necessary, though it may become less intensive over time. Maintaining high-quality, accurate, and granular data is critical for the algorithm’s performance, especially when dealing with diverse populations. Social determinants of health, cultural factors, and patient diversity must be carefully considered, as well as the capabilities, training levels, and infrastructure available to the local healthcare staff and surgical teams. Responsible implementation means adapting AI solutions to real-world contexts—not the other way around.

Alexandra E. Page, MD, FAAOS, is a foot and ankle specialist in private practice in San Diego, California, and the deputy editor of AAOS Now.