Shrivastava Awarded NIH R01 Grant for Creating Smart Implants for Managing Brain Disorders
ECE Associate Professor Aatmesh Shrivastava was awarded a $1.2M grant from the National Institute of Neurological Disorders and Stroke for “Transforming Brain Disorder Management through Ultra-Low-Power ML Implants.”
Abstract Source: NIH
The project focuses on developing an ultra-low power system for on-chip machine learning (ML)-enhanced sub-scalp EEG devices, aiming for real-time medical intervention and state-of-the-art performance. It addresses the challenge of large intersubject EEG variability by innovating in ML algorithm design and small form-factor system development. The goal is to create an efficient, real-time solution compared to conventional approaches, enabling advancements in brain-computer interfaces (BCI) and neural engineering for understanding, diagnosing, and managing neurological conditions. The overall system will be realized using cross domain innovation ranging from ML algorithm, to circuit design, to system design. It will develop and validate a new family of transient ML algorithms to enhance information extraction from sub-scalp EEG (ssEEG) data. These algorithms use a representation-based approach that leverages neurophysiological principles to mitigate volume conduction effects, improving ssEEG signal resolution and enabling ultra-low power hardware implementation. Additionally, the project will design an ultra-low power application-specific integrated circuit (ASIC) for ssEEG processing, utilizing analog computing to minimize power consumption and device size by eliminating the need for ADC/DAC stages and reducing memory usage. Analog computing improves the power consumption and reduces the area of the hardware by several fold. A proof-of-concept prototype will be created, focusing on epilepsy as a test case, with a custom ssEEG device fabricated and tested on rat models to detect induced seizures. The goal is to demonstrate the effectiveness of integrating ultra-low power ML into ssEEG devices. The overall system will be validated using in-vivo testing on rats and available EEG databases.