On March 14, 2023, a company based in the U.S. and Poland—Neatsy—released its doctor dashboard on the world’s first app for the detection of orthopedic and podiatry-related problems. Working in conjunction with the Massachusetts General Foot & Ankle Research and Innovation Lab, Neatsy has created patented AI algorithms that can help patients understand foot, knee and hip pain—all from their smartphones.
After taking a 3D scan of the foot, the user sends the image to the healthcare provider via the Neatsy app. In some cases, the patient is presented with a diagnosis and a care plan, such as physical therapy or orthotics. In others, the user responds to questions about sneaker fit preference and then takes a 3D scan of their feet using a smartphone. An array of possible sneakers, available for sale in the app, is then presented to the user.
Soheil Ashkani-Esfahani, M.D. is an orthopedic surgeon who directs the Massachusetts General Foot & Ankle Research and Innovation Lab. Dr. Ashkani-Esfahani told OTW, “We think that technology of this nature may improve the precision of diagnosis among providers, help them with the process of decision-making, and help the patients report their outcomes and feedback. Moreover, this can help democratize knowledge and expertise globally, especially in underprivileged settings.”
Artem Semjanow, the founder and CEO of Neatsy.AI, brings years of experience as former head of the machine learning team at Prism Labs. He told OTW, “With Neatsy, individuals can use their smartphone to scan their feet and get a personalized analysis of their foot length, width, and arch. The app’s machine learning technology compares the scans to anonymous 3D scans with doctor-verified diagnoses. If there’s a possibility of a podiatric issue, the app will notify the user to consult a specialist. Neatsy’s AI computer vision facilitates a self-service patient diagnostic tool with accuracy comparable to a visual examination by a doctor.”
“Neatsy is the first to use a 3D model autoencoder in its phone application. The program is a neural network tasked with creating low-dimensional, informative embeddings to compress a point cloud from tens of thousands of points to hundreds of numbers without losing information. We adapted foldingNet, added regularizations and changed the architecture so that the final model fits in a mobile phone and extracts maximum accuracy without big medical data. At the moment, users’ scans are compared to over 65,000 3D scans.”
Looking forward, the company has plans to launch full-body scans, thus allowing users to track body volume index, overall body measurements, and spinal/posture issues.
