Epileptic Seizure Detection With Patient-Specific Feature and Channel Selection for Low-power Applications.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• Epileptic seizures require accurate detection for timely intervention.
• Existing seizure detection methods often lack efficiency for low-power wearable or implantable applications.
• Intracranial electroencephalography (iEEG) provides high-resolution data for seizure monitoring.

Purpose of the Study:

• To develop an accurate and energy-efficient epileptic seizure detector using iEEG recordings.
• To optimize the detector for low-power wearable and implantable applications.
• To improve patient-specific seizure detection through advanced feature selection and data reduction.

Main Methods:

• Utilized eleven time-domain features with low hardware complexity.
• Developed a novel two-step feature ranking algorithm (MDMR) for patient-specific feature selection.
• Implemented a two-stage, energy-efficient architecture with a controller to minimize power consumption.
• Proposed a patient-specific iEEG channel selection technique for data reduction (average 68%).
• Implemented the algorithm on a Cyclone V FPGA and tested on a dataset from twelve patients.

Main Results:

• Achieved 100% sensitivity across all tested patients.
• Demonstrated significant improvements in specificity and detection delay compared to state-of-the-art methods.
• Reported an average dynamic power consumption of 0.49 mW, suitable for low-power applications.
• Achieved an outstanding figure-of-merit (FoM_SD) of 0.464, outperforming existing techniques.

Conclusions:

• The developed epileptic seizure detector is highly accurate and energy-efficient.
• The patient-specific feature and channel selection methods enhance performance and reduce computational complexity.
• The system is well-suited for low-power wearable and implantable applications, offering a significant advancement in epilepsy monitoring.