The code used by Karoly et. al. (2016) Slow Rhythms of Seizures Improve Prediction
Define a log regression classifier using basic features of NeuroVista data for seizure prediction, with output weighted by time of day.
NB most of this code won't run, as it requires my personal login files for the iEEG portal https://www.ieeg.org/, as well as information about patients seizure times that is not publicly available. The code is here as a reference for the methods used. Training data and forecast results are also provided.
getInterictal: get interictal segments of data from iEEG portal
getSeizures: " " preictal " "
NV_filters: make filters
NV_seizure_prob: sets up circadian profiles
NV_validate_classifier: 10 fold cross-validation of LR model
NV_train_classifier: train LR model on all data
NV_eval_forecast: works out the Brier scores from probbaility vector
NV_calibrate_forecast: adjusts probability vector based on calibration curve
logistic_regression_fit: fits the LR model to training data (code from Andrew Ng, Machine Learning Coursera, https://www.coursera.org/instructor/andrewng)
logistic_regression_run: gets output of LR model given an input feature vector
calculate_features: gets the correct features given an index (from 1-16)
time_of_day_pdf_estimate(WRAPPED): calculates kernel mixture of von-Mises distribution. Relies on circular statitsics toolbox in MATLAB
(https://www.mathworks.com/matlabcentral/fileexchange/10676-circular-statistics-toolbox--directional-statistics-?requestedDomain=www.mathworks.com)
pt_Forecast: contains the probability vector calculated for the entire trial and indicator function
[1] Cook, Mark J., et al. "Prediction of seizure likelihood with a long-term, implanted seizure advisory system in patients with drug-resistant epilepsy: a first-in-man study." The Lancet Neurology 12.6 (2013): 563-571.