Seizure detection device

Context

During my master's degree, I successfully developed a wearable device for multimodal biosignal acquisition and epilepsy monitoring. This project aimed to detect seizures using a glove/smartwatch prototype equipped with biomedical sensors, namely acceleration data, photoplethysmography (PPG), and electrodermal activity (EDA). Using state-of-the-art algorithms and cloud services, I employed a real-time seizure detection device with remote monitoring.

Project Highlights

1. Prototype Development. Design and assembly of a glove/smartwatch prototype, integrating the biomedical sensors for the chosen biosignal modalities (acceleration data, PPG, and EDA). The device was lightweight and user-friendly, ensuring comfortable long-term wear for patients.

2. Signal Acquisition and Processing. Implementation of state-of-the-art signal processing techniques based on spectral features extracted from acceleration data. These features are known to be accurate in seizure detection in patients with tonic-clonic seizures. 

3. Data collection. Multimodal signal collection under different non-seizure motion conditions (eating, writing on laptop, walking and running) and simulated tonic-clonic seizures*.

4. Getting the pipeline up-and-running. The was pipeline developed and then trained on the collected data. It involved some steps such as data partitioning (i.e. to rotate the trained and test data and check for differences in classification performance), followed by threshold optimization using a grid-search approach with a refining step. The trained model achieved an accuracy of 91.60% (discriminating seizure vs. non-seizure hand movement) when trained with all motion conditions.

5. Real-time and remote implementation. By using cloud technologies (Firebase), a remote monitoring system was established, using an adapted version of the detection algorithm with the trained threshold. The device worked correctly and, as a demonstration, I showcased the efficacy of the prototype by simulating seizures by shaking the hand. The device accurately detected the simulated seizure, triggering an audible alarm. 

*Although our research institute has obtained succesful approval for performing acquisitions in hospital settings with real patients, during COVID, it was hard to obtain it within the project's time length, thus seizures had to be simulated.