EEG EYE BLINK CLASSIFIER
main.py taken in arguments -f for filename to work with and -fx for location where to export new annotated edf file
pip install -r requirements.txt
Example
python main.py -f eeg/test.fif -fx eeg/test_anot.edf
For Batch-operation and metric calculation use mnbat.py. The module reads all fif files in mentioned directory. And predicted values are stored in a metrics.csv file in the same directory. with chan 0 representing Fp1 and chan 1 representing Fp2.
python main_batch.py -fp "fif directory path"
Although the annotations are more visually appealing with EDF Browser in case a matplot.plt is needed simply call plotter.py with with edf file in argument
python plotter.py -f "File.edf"
Metrics are based on standard formula for accuracy, precision, F1. Since all blinks labels havent been annotated Precision is calculated based on no of predictions in closed eye duration. Any predictions are considered false. Accuracy is predicted based on How many blinks have been detected that are labelled.
Accuracy = (Total Blink labels found in duration of predicted blink label)/Total Blink Labels Annotated
Precision = (Total Predictions - Predictions Made in Closed Eye Interval)/Total Predictions
Recall = (Total Predictions that can be counted as Accurate)/(Accurate Predictions + False Predictions)
BLINK Algorithm used as reference for building a eye_blink detection module based on temporal characterstics on Eye blinks. This module uses single channel data for extracting blink characterstics over time domain and detecting potential blink candidates. Fp1 and Fp2 both are great channels for detecting eye_blinks. However Fp1 showed a greater accuracy hence it was selected as preffered channel Although using relationship matrix for potential blinks in both channels can provide more refined selection of blink duration.
edf files are not supported use edfTofif.py to convert edf files to fif and use them.
edfTofif detects all edf files in a directory and saves them in fif format in a sub-directory called fif MNE shows codec errors while performing certain functions on EDF If still need to use edf use import pyedflib for reading annotations
Incase of batch_files if metrics.csv is not present in directory add one