Integrated deep learning model for personalized transcranial magnetic stimulation
Project/Area Number |
22K12765
|
Research Category |
Grant-in-Aid for Scientific Research (C)
|
Allocation Type | Multi-year Fund |
Section | 一般 |
Review Section |
Basic Section 90110:Biomedical engineering-related
|
Research Institution | University of Hyogo |
Principal Investigator |
Rashed Essam 兵庫県立大学, 情報科学研究科, 教授 (60837590)
|
Co-Investigator(Kenkyū-buntansha) |
平田 晃正 名古屋工業大学, 工学(系)研究科(研究院), 教授 (00335374)
ゴメスタメス ホセデビツト 千葉大学, フロンティア医工学センター, 准教授 (60772902)
|
Project Period (FY) |
2022-04-01 – 2025-03-31
|
Project Status |
Granted (Fiscal Year 2022)
|
Budget Amount *help |
¥3,250,000 (Direct Cost: ¥2,500,000、Indirect Cost: ¥750,000)
Fiscal Year 2024: ¥650,000 (Direct Cost: ¥500,000、Indirect Cost: ¥150,000)
Fiscal Year 2023: ¥780,000 (Direct Cost: ¥600,000、Indirect Cost: ¥180,000)
Fiscal Year 2022: ¥1,820,000 (Direct Cost: ¥1,400,000、Indirect Cost: ¥420,000)
|
Keywords | Deep learning / Brain stimulation / TMS / Segmentation / brain stimulation |
Outline of Research at the Start |
Transcranial magnetic stimulation (TMS) is commonly used in several clinical procedures. Due to large variability of efficacy, planning of personalized stimulation is highly desired. However, personalized TMS requires a complicated data processing pipeline for individual head model generation to provide target-specific stimulation. This project aims at the development of accurate and reliable deep learning model to handle data processing, subject variabilities and other physical effects with integrated deep learning framework.
|
Outline of Annual Research Achievements |
In this year, we have developed a new deep learning model for the segmentation of all head tissues using multi-modal MRI scans. The developed model segmented MRI scans into 16 different tissues. A dataset of multi-modal MRI (T1w, T2w, PD and MRA) are collected from 600 subjects. The images are pre-processed through bias-correction, registration and normalization to be used for human head dataset. A segmentation of 20 subjects are obtained through semi-automatic method and used to train the deep learning model. The developed method is applied to the remaining MRI dataset to generate segmented head models. After visual validation, a set of 196 fully segmented human head models with variabilities in gender and age was approved. The use of MRA leads to significant improvement of identification of brain vessels and arteries. This dataset will be used next year for large-scale TMS study. Initial TMS study was conducted using two subjects (through collaborators) to compute the induced electric field using different coil positions, orientation and location around motor cortex. Another deep learning model was developed for the estimation of the induced electric field in human brain directly from the anatomical images. Training of the new model is scheduled for next year.
|
Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
The research is progress smoothly than planned. We initially plan to complete three work packages in the first year (development of the deep learning model, data collection, and TMS simulation). Which are all successfully completed. Furthermore, we have reported results through several presentations and invited talks. Work package 4 that include network training (scheduled for FY2023) was partially completed in FY2022.
|
Strategy for Future Research Activity |
The research plan for FY2023 include two work packages (network training and validation, optimization, and network architecture verification). We plan to conduct several training scenarios using the MRI data we already prepared to optimize the training parameters and achieve ultimate accuracy for both anatomical segmentation and induced electric field estimation. Also, we are currently drafting the research results of FY2022 to be submitted as a journal paper. In FY2024, two additional work packages will be completed (reporting results in journal/conference publications and deployment of open-source software of the developed deep learning models).
|
Report
(1 results)
Research Products
(3 results)