Epilepsy is a group of neurological disorders identifiable by infrequent but recurrent seizures. Seizure prediction is widely recognized as a significant problem in the neuroscience domain. Developing a Brain-Computer Interface (BCI) for seizure prediction can provide an alert to the patient, providing a buffer time to get the necessary emergency medication or at least be able to call help, thus improving the quality of life of the patients. A considerable number of clinical studies presented evidence of symptoms (patterns) before seizure episodes, this project aims to exploit those patterns in the brainwaves and recognize them as anomalies and predict seizure.
Features
- Quick Preprocessing of data for real-time use, preprocesses 10 mins worth EEG recording sampled at 400Hz with an average of 1.02s i.e processes 10 (mins) X 400 (freq) x 60 (sec per min) = 2,40,000 data points in 1.02 sec
- Artifacts analysis from EEG to automatically remove noise due to Eye Blinking, Heart Beats and Motor Movement.
- Extensive feature extraction from multiple domains:
Energy Based: Entropy, Higuchi Features, Fractal Dimensions
Frequency: Band powers using Fast Fourier Transform within different frequency bands
Time Based: Statistical features skewness, standard deviation
Channel Based: Eigen Matrix of a correlation matrix
- Ensembled Statistic Machine learning model for classification on the basis of feature extracted
Workflow
- Scalp Raw EEG recorded serves as an input which is passed to the preprocessing and artifact removal module. The input EEG is sampled at 258 Hz, scaled and is made free from noise, sampled into windows of 2 min with a 1 min overlap
- The preprocessed window is passed to the feature extraction module where the above-mentioned information is extracted which serve as a feature vector for our model
- Each window is annotated based on the fact that was it extracted from file corresponding to Seizure recording or not.
- The features are then passed to an ensemble model made from a simple statistical model for improving sensitivity to the anomalies. The ensemble was made using the Boosting technique
Knowledge Gained
- Signal Processing and handling noise generated while data recording.
- EEG processing and artifacts analysis and understanding to what part of the brain they correspond to.
- Different information hidden within the voltage recordings of the brain.
- Difference between Bagging and Boosting techniques of creating ensemble models and when which technique should be used.