Anantha Chandrakasan, John Guttag
Epilepsy is a neurological disorder affecting 50 million people worldwide. It is caused by abnormal neuronal firings that lead to undesired clinical reactions, including loss of motor control/coherence, convulsions, and even death. Since neuronal firings generate electrical field potentials, we can sense them, non-invasively, via on-scalp EEG. The EEG signatures of seizures, however, are highly diverse, especially from patient to patient, and they must be separated from seemingly abnormal EEG patterns, that can occur frequently, but don't directly lead to clinical seizures.
As a result, machine learning techniques must be applied to detect the onset seizures with minimal latency, so that a therapeutic response can be generated as early as possible before a catastrophic clinical reaction. This project aims to develop very low-power and integrated EEG acquisition circuits, digitization circuits, and processing circuits to derive and classify feature-vectors for a continuous, wearable seizure detection system suitable for everyday use by epilepsy patients. The implemented algorithms are based on 536 hours of patient tests to ensure efficacy, and the system is intended to be highly portable and low-power for very long-term use.
Part of the Energy Efficient Integrated Circuits & Systems Group at MIT
Platforms for Ultra-Low-Power
Massachusetts Institute of Technology
50 Vassar St. 38-107
Cambridge MA 02139 USA
617-253-0016 (main), 617-253-5053 (fax)