Improvement of decoding accuracy of visual stimulus using deep learning and ECoG big data

2021 Fiscal Year Final Research Report

• Project/Area Number: 20K16466
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 51010:Basic brain sciences-related
• Research Institution: Osaka University
• Principal Investigator: FUKUMA RYOHEI 大阪大学, 医学系研究科, 特任助教(常勤) (20564884)
• Project Period (FY): 2020-04-01 – 2022-03-31
• Keywords: Brain-Machine Interface / 皮質脳波

Outline of Final Research Achievements

The development of machine learning technology enabled the decoding of human visual perception from brain activity. However, the accuracy of existing methods for decoding is still insufficient. In this study, we attempted to improve the decoding accuracy by using a deep learning model with electrocorticograms recorded while the subjects perform daily activity. By using the trained model as a feature extractor from the electrocorticograms recorded during visual stimulus task, the decoding accuracy was better than that of existing methods. When the model was applied to a behavioral task, the decoding accuracy was also improved.

Free Research Field

Brain-Machine Interface

Academic Significance and Societal Importance of the Research Achievements

脳活動パターンから情報を読み出す脳情報デコーディング技術は，色々な疾患や外傷で身体機能が損なわれてしまったヒトの日常生活の質の改善に役立つと考えられている．しかし，現在のところ脳情報の読み出し精度は十分ではない．また，一般に脳情報の読み出しには脳活動を計測しながら患者が課題を行うことが必要である．脳情報の読み出し精度を改善するためには課題を長時間行うことが望ましいが現実的には困難である．そこで，本研究では課題を行っていない自由行動下での脳活動を用いることで，精度が改善できることを明らかにした．即ち，患者への課題を増やすことなくより高い精度が得られる．