Analysis and design of a patient specific seizure detection processor using linear support vector machine

Abstract

Epilepsy is a chronic disorder of the central nervous system (CNS) characterized by recurrent seizures. Seizures are caused by abnormal electrical discharges in the brain, which can vary from just a lapse of attention to whole body convolutions and even to death. It's difficult to determine the occurrence of seizure event, especially in children. While hard to detect at young age, the seizure event becomes more severe with age. In the US it affects approximately 3 million children under the age of 15. As of yet there is no patient-friendly seizure detection solution for this alarming large population. This thesis presents the implementation of a multichannel patient-specific seizure detection processor using linear Support Vector Machine (SVM) which can be integrated on a wearable patch sensor to form an ambulatory medical application. This is the first on-chip implementation of feature extraction, classification and recording integrated with a storage function for seizure detection. To integrate 8 channels, an area- and energy-efficient filter architecture using Distributed Quad-LUT (DQ-LUT) is proposed, which reduces area by 64.2% with minimal overhead in powerdelay product. The on-chip patient specific classification with a linear (SVM) results in 82.7% seizure detection rate for 206 hours of EEG data from 23 patients. The overall energy efficiency is measured as 1.52μJ/classification. Finally, the test results of implemented chip using Children's Hospital Boston-MIT EEG data base are also provided.

Date of Award	2012
Original language	American English
Supervisor	Jerald Yoo (Supervisor)