Prediction of Transcranial Magnetic Stimulation Responses from Structural and Functional Magnetic Resonance Images using Deep Neural Networks

Transcranial magnetic stimulation (TMS) is a non-invasive neuromodulation technique that has been approved for treating neurological and psychiatric disorders. To be effective, one needs to induce sufficient electric fields in critical regions of the brain, which can be estimated using complex and time-consuming finite element analysis (FEA) simulations using spatial distribution of electrical properties of the patient’s brain inferred from anatomical slices of magnetic resonance imaging (MRI). We showed a Deep Encoder-Decoder Convolutional Neural Network can be trained to predict the electric field distribution using T₁-weighted and T₂-weighted magnetic resonance imaging (MRI) based anatomical slices with high accuracy in healthy [1] as well as mild to moderate traumatic brain injury patients [2], which will be discussed in detail.

This will be followed by a discussion of the possibility of using such Deep Convolutional Neural Networks (DCNNS) to predict resting motor threshold (RMT) for a given TMS protocol based on both anatomical and functional MRIs that also include functional information from a randomized clinical trial involving healthy participants and schizophrenia patients.

Finally, we will go beyond the use of AI for imaging, and discuss how future AI hardware [3] that can implement such large deep neural networks on edge devices, where both hardware resources and power available are severely constrained, can potentially enable new therapeutic paradigms such as responsive neuromodulation.