Cardiac disease is a primary cause of human death and disability. Nearly everyone suspected of cardiac disease is imaged by echocardiography, i.e., cardiac ultrasound, "echo". Complete evaluation of cardiac anatomy including chambers, valves and function is a complex, costly test requiring lengthy training to perform/interpret.
We envision a “basic echo” performed by a lightly-trained user on inexpensive, portable equipment, to focus on LVEF. AI would assist the user to obtain high-quality images by providing immediate feedback as the user sweeps the probe over the chest, identifying when they reach the correct probe position. Then AI would automatically compute LVEF from images obtained at this location.
Wrist and elbow fractures in children are currently examined using x-rays. The actual occurrence of fractures is rare but children are still exposed to radiation. Ultrasound (US) is a safe, cost-effective, and portable alternative to x-rays. It is also highly sensitive in visualizing cortical disruption. With the advent of handheld ultrasound devices like Lumify, V-Scan, and Clarius, ultrasound is even more portable making it ideally suited for use in pediatric emergency departments. However, the effectiveness of such ultrasound-based examination depends heavily on image quality and sonographer expertise which is required both in acquiring and interpreting the image. We aim to improve the image quality of wrist images and to provide an AI tool for automated image analysis that could be used to detect fractures quickly and accurately in emergency triage.
Early diagnosis of Developmental Dysplasia of Hip (DDH) using ultrasound can result in simpler and more effective treatment options. However, universal screening programs using ultrasound examinations are often criticized for overdiagnosis and lack of reliability since its interpretation is manual and subjective. With our industrial collaborator MEDO.ai Inc, we have developed the world's first AI tool for automatically diagnosing DDH from ultrasound scans. Currently, we aim to expand our AI analysis to scans acquired from handheld ultrasound probes that are low-cost and pocket-sized. Using AI, we aim to enhance the quality of images acquired from these devices and apply geometric deep learning techniques on surface models of various hip structures to understand the development of normal and abnormal hips.