Predicting Severity of Spinocerebellar Degeneration Patients Using Machine Learning Model and Gait Video

2023 Fiscal Year Final Research Report

• Project/Area Number: 22K20843
• Research Category: Grant-in-Aid for Research Activity Start-up
• Allocation Type: Multi-year Fund
• Review Section: 0902:General internal medicine and related fields
• Research Institution: Hokkaido University
• Principal Investigator: Katsuki Eguchi 北海道大学, 医学研究院, 客員研究員 (20852635)
• Project Period (FY): 2022-08-31 – 2024-03-31
• Keywords: 脊髄小脳変性症 / パーキンソン病 / 歩行障害 / 機械学習 / 動画

Outline of Final Research Achievements

During the study period, gait videos of 60 SCD patients and 72 PD patients were collected.
First, using the gait videos of SCD patients, we had a deep learning model predict SARA from the gait videos using the SARA score measured by the physician as the supervised data. The model was able to predict SARA score from the gait videos with an accuracy of 2.7 mean squared error and 0.71 coefficient of determination. The performance of the model was evaluated by leave-one-out cross validation, and the model was able to distinguish between SCD and PD with an accuracy of 87% and ROC-AUC of 0.88.

Free Research Field

臨床神経学

Academic Significance and Societal Importance of the Research Achievements

本研究において、深層学習モデルが歩行動画からSCDの重症度の評価やSCDとPD患者の歩行の区別が一定の精度で可能であることを示した。歩行動画の撮影は患者に侵襲を与えることなく簡便に行うことができる点から、歩行動画に深層学習モデルを適用して重症度評価や疾患鑑別を行うことは簡便なスクリーニング手法として有用であることが示され、社会実装目指す意義にある手法であることを示すことができた。