Machine learning for brain stimulation and neuroimaging

Background. Uncertainty quantification, optimization, and closed-loop applications of brain stimulation require thousands of repeated field computations, and clinical neuroimaging and neurophysiology data call for fast, reliable classifiers. Learned emulators and low-order models can deliver these at a fraction of the cost of physics-based simulation.

Objective. Develop deep-learning emulators and classifiers that are fast enough for repetitive use while retaining the accuracy of physics-based methods.

Approach and main results. DeeptDCS is an Attention U-Net emulator that takes head volume conductor models, with electrode configurations embedded without additional input channels, and outputs the three-dimensional current density induced by transcranial direct current stimulation. One emulation takes a fraction of a second, at least two orders of magnitude faster than a physics-based open-source simulator, with satisfactory accuracy and, after fine-tuning, good generalization to untrained electrode configurations. In collaboration with Prof. Luciano Castillo’s group, proper orthogonal decomposition coupled with a convolutional neural network detected brain tumors in MRI scans with 95.9% accuracy at about one third of the computational time of a full convolutional network (99.2%).

Ongoing work. Machine learning approaches for electroencephalography analysis to classify responses to repetitive transcranial magnetic stimulation (manuscript under review).

Publications.

X. Jia, S. B. Sayed, N. I. Hasan (G), L. J. Gomez, G. Huang, and A. C. Yucel, "DeeptDCS: Deep Learning-Based Estimation of Currents Induced During Transcranial Direct Current Stimulation," IEEE Transactions on Biomedical Engineering, vol. 70, no. 4, pp. 1231-1241, 2023. link

R. Appiah, V. Pulletikurthi, H. A. Esquivel-Puentes, C. Cabrera, N. I. Hasan (G), S. Dharmarathne, L. J. Gomez, and L. Castillo, "Brain Tumor Detection Using Proper Orthogonal Decomposition Integrated with Deep Learning Networks," Computer Methods and Programs in Biomedicine, vol. 250, pp. 108167, 2024. link

A. Castillo Jimenez (G), L. Castillo, and L. J. Gomez, "Machine Learning Approaches for Electroencephalography (EEG) Analysis in Repetitive Transcranial Magnetic Stimulation (rTMS) Response Classification," Scientific Reports, under review.