2021 Fiscal Year Final Research Report
Extracting features of neuroimaging in pychiatric disorders using machine learning and multicenter datasets
| Project/Area Number |
18K07597
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 52030:Psychiatry-related
|
|---|
| Research Institution | Tokyo Medical and Dental University (2019-2021)
Kyoto University (2018) |
|---|
Principal Investigator |
Genichi Sugihara 東京医科歯科大学, 大学院医歯学総合研究科, 准教授 (70402261)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
大石 直也 京都大学, 医学研究科, 特定准教授 (40526878)
山下 祐一 国立研究開発法人国立精神・神経医療研究センター, 神経研究所 疾病研究第七部, 室長 (40584131)
孫 樹洛 京都大学, 医学研究科, 研究員 (60771524)
|
|---|
| Project Period (FY) |
2018-04-01 – 2022-03-31
|
| Keywords | 機械学習 / 精神疾患 / 脳画像解析 |
|---|
| Outline of Final Research Achievements |
The objective of this applied research is to construct a system that uses machine learning to remove differences between imaging facilities in brain image data, to extract the features of mental disorders in brain images, and to analyze the data with further heterogeneity in mind. Using publicly available datasets, we built a deep learning model to identify the imaging facility from MRI images from 6 facilities. With this model, we succeeded in building a model that identifies imaging facilities with a correct response rate of more than 99%, and we also succeeded in visualizing the characteristics of imaging facilities. Furthermore, by selecting two facilities from the dataset and using an adversarial generative network, a type of deep learning, we have succeeded in generating brain images in which subjects taken at one facility appear to have been taken at another facility, thereby verifying the effectiveness of this method.
|
| Free Research Field |
精神医学
|
| Academic Significance and Societal Importance of the Research Achievements |
本申請研究により構築された方法により、多施設で撮像された脳画像データセットの施設間差を除去することが広く可能となれば、多くのデータを統合し、サンプルサイズを増やした研究につながる。背景にある病態がさまざまな精神疾患の脳画像を用いた病態解明に向けた研究を遂行していく際には、こうした方法は研究の弱点を補い、さらに研究の促進に寄与することが期待できる。また、ここで構築したモデルは今後、さらに汎用性の高い深層学習モデルに応用される可能性を持っている。
|
