Data Collection for the Validation of an Artificial Intelligence Software to Support Musculoskeletal Ultrasound Examination

Overview

This study conducted at three locations, aiming to collect ultrasound images from volunteers and assess the performance of an AI software, MSK Go, using these images. The software to be assessed is designed to assist physicians and healthcare professionals in performing ultrasound exams by classifying scan views and identifying key anatomical structures during a musculoskeletal ultrasound examination. The main question the study aims to answer is whether the AI software performs safely and effectively for future clinical use.

This study is designed to validate the performance of the MSK Go software in classifying musculoskeletal ultrasound examination views and segmenting relevant anatomical structures in a U.S. population. Data collection method and analysis: A total of 79 subjects were enrolled and undergone shoulder, elbow, wrist/hand, knee, and foot/ankle scans at three clinical sites. Ultrasound examinations were performed by MSK ultrasound experts without using MSK Go. Each participant spent approximately 30 minutes in the ultrasound scanning session. Demographic data were collected for all subjects. An FDA-cleared, commercially available ultrasound device was used at each location. The collected ultrasound scans are then annotated by healthcare professionals with expertise in musculoskeletal sonography. In a post hoc analysis, the ultrasound clips were processed using the AI software, and the software-generated classifications of views and segmentations of anatomical structures were compared with the reference annotations . Convenience sampling was applied during data collection to ensure that participants represented a balanced distribution across age, gender, BMI, and ethnicity.