A micropower support vector machine based seizure detection architecture for embedded medical devices.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Implantable neurostimulators offer novel epilepsy treatment options by sensing seizures.
• Detecting seizures is difficult due to patient-specific intracranial EEG signal variability.

Purpose of the Study:

• To develop and evaluate a patient-specific seizure detector for implantable neurostimulators.
• To compare the performance and power consumption of patient-specific versus non-specific seizure detection algorithms.

Main Methods:

• A machine learning-based architecture was developed for seizure detection.
• The algorithm was implemented in the micropower domain for embedded systems.
• Performance was evaluated using a seizure library, comparing patient-specific and non-specific approaches.

Main Results:

• The patient-specific detector demonstrated superior performance compared to the non-specific detector.
• The machine learning approach achieved lower power consumption.
• The architecture proved feasible for micropower, embedded applications in implantable devices.

Conclusions:

• Patient-specific, machine learning-based seizure detection is a viable and effective strategy for implantable neurostimulators.
• This approach offers improved performance and energy efficiency for epilepsy treatment devices.
• The developed architecture supports the feasibility of embedded, low-power seizure detection systems.