2022 Fiscal Year Final Research Report
Extraction of new electrophysiological feature for the prediction of epileptic seizures
| Project/Area Number |
19K18427
|
|---|
| Research Category |
Grant-in-Aid for Early-Career Scientists
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Basic Section 56010:Neurosurgery-related
|
|---|
| Research Institution | Asahikawa Medical College (2021-2022)
Osaka University (2019-2020) |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2019-04-01 – 2023-03-31
|
| Keywords | epilepsy / deep learning / data-driven |
|---|
| Outline of Final Research Achievements |
A deep learning model named Epi-Net using raw iEEG signals detected seizures of different types of epilepsy with better accuracy than the SVM model using power and PAC features. Moreover, we evaluated how each frequency amplitude contributed to the seizure likelihood inferred by the trained Epi-Net, and proposed the data-driven epileptogenicity index, d-EI, based on the relative contribution of each frequency. The proposed d-EI succeeded in classifying the seizures and the interictal states better than other previously known features such as the power, PAC, and ER.
|
| Free Research Field |
Neurosurgery
|
| Academic Significance and Societal Importance of the Research Achievements |
これまでに報告されてきた波形解析は個人ごとに行われていることが多かったが、今回の研究では波形を数値データとして直接深層学習の入力として使用し、患者間で共通のてんかん発作時に現れる波形特徴量を抽出できることを示した。てんかん発作を波形からより正確に同定することは、発作を検知して脳を刺激してんかん発作の伝播を防ぐシステムを作成する際に非常に重要である。今回報告した手法や、新しい指標が今後てんかん患者診療に応用されることが期待される。
|
