Ambulatory REACT: real-time seizure detection with a DSP microprocessor

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neurology

Background:

• Automated seizure detection using electroencephalography (EEG) is crucial for patient monitoring.
• Real-Time EEG Analysis for event detection (REACT) is a Support Vector Machine-based technology effective for seizure detection in adults and neonates.
• Implementing advanced algorithms on efficient hardware is essential for practical clinical applications.

Purpose of the Study:

• To implement the REACT algorithm on a commercial Digital Signal Processing (DSP) microprocessor, the Analog Devices Blackfin®.
• To develop a prototype system for ambulatory or in-ward automated EEG analysis.
• To analyze the computational complexity of REACT stages on the Blackfin processor and optimize for reduced complexity and power.

Main Methods:

• Implementation of the REACT algorithm on the Analog Devices Blackfin® DSP microprocessor.
• Analysis of the computational complexity of EEG feature extraction stages within the REACT algorithm.
• Selection of a reduced, platform-aware feature set based on hardware profiling.
• Evaluation of seizure classification accuracy for a lower-complexity, lower-power REACT system.

Main Results:

• Successful implementation of the REACT algorithm on the Blackfin DSP.
• Identification of computational bottlenecks in EEG feature extraction.
• Demonstration of a reduced feature set's viability for optimized REACT performance.
• Evaluation of the trade-off between system complexity, power consumption, and seizure detection accuracy.

Conclusions:

• The Blackfin DSP is a suitable platform for implementing automated EEG analysis systems like REACT.
• Algorithm optimization through hardware-aware feature selection can lead to lower-complexity, lower-power seizure detection systems.
• This work paves the way for practical, ambulatory EEG monitoring systems for seizure detection.