F027 An AI-assisted video telemedicine system for assessment of movement disorder in HD patients

<h3>Background</h3> Huntington’s disease (HD) is a rare chronic progressive neurodegenerative disease and assessment of movement disorder is of great significance for the care of HD. Development of a convenient and effective method for clinical analysis of HD severity is in urgent need. <h3>Methods</h3> In the present study, all participants were required to provide 6 1-minute-videos, demonstrating patients in seated, standing, and walking positions. 27 patients participated in the video reliability test to evaluate the consistence of scores derived from videos taking at home and scores derived in the clinic visit. 48 HD patients and 48 healthy controls were enrolled in the Artificial intelligence (AI) study. Firstly, we designed a 3D convolutional module based on channel-spatial attention to extract data features. We then used an excitation-based 3D residual convolution module for feature learning. Finally, we used a fully connected layer based on channel-spatial attention to complete feature compression and obtain video classification results. An AI-assisted video telemedicine system was developed based on these results. <h3>Results</h3> The consistence between video analysis and in-person visit was excellent (r = 0.969, P &lt;0.001). The ACC of our AI model to distinguish HD patients and Healthy controls was 85%. <h3>Conclusions</h3> Our video telemedicine platform is a feasible and reliable option for HD patient care. Our AI models further facilitate HD care, especially at the remote area.