AI(Aritificial Intelligence) Determination of Functional Ambulatory Category of Stroke Patients

We classified gait functions using AI by extracting and analyzing coordinate data acquired from viewing gait performance videos of stroke patients. We collected gait videos of normal people and stroke patients from 6 medical institutions to learn the AI model for gait function classification of stroke. The subjects who participated in the study were normal 278, mean age 64.9, stroke 218, mean age 62.2. As a result of verifying the performance of the AI model based on gait video, the accuracy of stroke classification for normal persons was 95.09% and AUC was 0.94. The functional gait classification accuracy of stroke patients was 83.44% and AUC was 0.95. In this study, walking video files only were applied without using various types of sensors. The AI model based on image data had an accuracy of over 80%. This tool served as support data for functional gait evaluation and diagnosis performed by physiatrists.