Related Experiment Video
Updated: Aug 6, 2026
08:23
A Multimodal Imaging- and Stimulation-based Method of Evaluating Connectivity-related Brain Excitability in Patients with Epilepsy
Published on: November 13, 2016
A Convolutional Neural Network Combining Multi-scale Convolution and Functional Connectivity for Predicting rTMS
Yongcong Li1,2, Jun Ma3, Tian Shu1
1Department of Rehabilitation, Southwest Hospital, Third Military Medical University (Army Medical University), Chongqing, 400038, China.
Annals of Biomedical Engineering
|July 20, 2026
Summary
A deep learning model accurately predicts response to repetitive transcranial magnetic stimulation (rTMS) for methamphetamine use disorder (MUD). This neuroscience-based approach can guide treatment decisions and optimize healthcare resource allocation for MUD patients.
Area of Science:
- Neuroscience
- Machine Learning
- Neuromodulation
Background:
- Repetitive transcranial magnetic stimulation (rTMS) shows promise for treating methamphetamine use disorder (MUD).
- Treatment outcomes for rTMS in MUD are highly variable.
- Predicting individual patient response is crucial for optimizing therapy.
Purpose of the Study:
- To develop a deep learning model for predicting rTMS treatment response in MUD patients.
- To leverage resting-state electroencephalography (EEG) data for predictive modeling.
- To guide personalized therapeutic decisions and reduce healthcare waste.
Main Methods:
- Resting-state EEG data were collected from 17 MUD subjects before rTMS treatment.
- A convolutional neural network (MSFC-CNN) integrating multi-scale (MS) and functional connectivity (FC) modules was developed.
- The MS module used multi-scale kernels and self-attention; the FC module analyzed connectivity matrices.
Main Results:
- The MSFC-CNN model achieved 88.06% accuracy in predicting treatment responders.
- The model significantly outperformed existing methods.
- Feature visualization and ablation studies confirmed the model's effectiveness and component contributions.
Conclusions:
- The MSFC-CNN model offers a viable method for predicting MUD treatment response using EEG.
- This approach provides methodological support for personalized neuromodulation therapies.
- Further research can refine this predictive tool for clinical application.