Intelligent Neural Interfaces

Gerard O'Leary (1,2,3), David M. Groppe (2), Roman Genov (1), Taufik A. Valiante (1,2,3,4)

1. The Edward S. Rogers Sr. Department of Electrical & Computer Engineering, University of Toronto

2. Krembil Research Institute

3. Institute of Biomaterials and Biomedical Engineering, University of Toronto

4. Division of Neurosurgery, Department of Surgery, University of Toronto

The biggest challenge in the treatment of neurological disorders are their patient-specific nature, as each brain is different and each electrode placement configuration unique. By recording neural activity from implanted electrodes, neurologists can understand a disorder and determine a course of treatment. However, brain signal recording systems generate large volumes of complex data which are infeasible to process fully. We have seen rapid progress in the field of artificial intelligence (AI) which has opened the door to decoding these data to detect brain states such as the onset of seizures in epilepsy. By finding hardware-friendly variants of brain-state classification algorithms, it could be possible to detect subtle pathological activity within implantable medical devices to support an immediate intervention.

There has also been a great advancement in identifying electrographic biomarkers in neural signal recordings during abnormal brain states such as seizures. Epilepsy can be viewed as a disorder of brain synchrony, as hypersynchronous avalanches of neural activity are exhibited during a seizure. Our research group has shown that the onset of seizures in focal epilepsy is heralded by a sharp decrease in synchrony between the seizure-generating and non-seizure-generating region. Combining such biomarkers with AI approaches for brain state classification has the potential for highly accurate seizure detection within an implant.

In response to detected pathological brain states, an electrical stimulus can be applied to the brain to alter the underlying behavior. Programmable stimulation waveforms aim to produce patient-tailored stimuli to control neural populations more precisely and reduce the side-effects associated with excessive stimulation. However, more programmability involves more complex configuration. Finding a rational approach to selecting the stimulator’s parameters could revolutionize the quality of treatment for those with neurological disorders. Steps in this direction are being taken using multi-electrode array (MEA) neural interfaces with reinforcement learning approaches. A customizable platform to test these devices with brain slice epilepsy models is being developed to enable the investigation of the most effective preventative stimulation parameters. The platform could also lead to fundamental discoveries about the underlying mechanisms of electrical stimulation and how it can be used to interact with neural tissue.