/EDFA

EDFA implementation

Primary LanguageMATLABGNU General Public License v3.0GPL-3.0

Automatic Epilepsy Spike Detection

Dataset • Environment • Download • How to Run • License

Algorithm diagram Sample Epilepsy Spikes
1 2

Dataset

Project team is currently considering the UCI machine learning repository, once dataset is uploaded users can export MEG/EEG data from the updated link. Each patient data contains the following elements.

  • data (A Matrix element with each row containing waveform of a specific channel. )

  • events (A struct element with M spike label imformation)

    Field Data Types Value
    label string "spikes"
    times array (double) M doubles
    samples array (double) M doubles

  • smpfreq (A integer element with sampling frequency)

  • chantype (An array element with each array element showing data type of the corresponding row in data element)

  • channame (An array element with each array eleemnt showing channel type of the corresponding type in chantype elements)

    Note: For EEG data, the channel waveform is the difference of voltage between the electrode and the ground, with the type of electrode shown in channame variable. EEG waveform can be generated by substracting two data waveforms thus eliminating the ground voltage. More basics about EEG signal can be found in this tutorial.

Environment

Our application is developed in Matlab 2020a. An equaling or higher version is recommended and no other dependencies are needed to implement our method. To implement other methods and utilities, it is recommended to install the following Matlab toolbox:

Download

This application can be downloaded by built-in git tool in Matlab. screenshot

How to Run

The repo provides a wide variety of methods and utilities which can help researchers to study epilepsy spike detection methods. Both detection methods and feature extraction methods in existing literature publications are provided.

Table of content

  • Our method
    • Pre-defined system parameters
    • Automatic detection function
    • Utilities

Our method

Both pre-defined system parameters and automatic detection function and utilities are described. The first subsection is listed for research purposes with some system parameters shown in variable location and variable name allowing easier tunning process. Then usages of automatic detection function and utilities are covered for implementations.

Pre-defined system parameters

Parameter Description Variable Location Variable Name
number of windows in TestBuffer /functions/main/FD_analysis.m KK
data scaling parameter /functions/main/FD_analysis.m ds
Initial u value before search /functions/main/threshold2.m c
non spike ratio R thresholds /functions/main/ratiojudge.m -
exponential search base number /functions/main/StateTrans.m searchD
number of linear scanning intervals /functions/main/StateTrans.m IndMax
ratio value for judging C_1, C_2 conditions /functions/main/SCSASpikeDetect.m ThK, ThN
SCSA C_1 , C_2 conditions thesholds /functions/main/SCSASpikeDetect.m Ts1, Ts2
maximum sample length in a region esRegionsExtract.m NL
number of channels in data esRegionsExtract.m chan_index
window sample size esRegionsExtract.m frame_size
moving window step size esRegionsExtract.m step

Automatic detection function

This section provides implementation of algorithm 3 in our paper. You can get extracted regions R_F from an input MEG/EEG patient data D (D is matrix with each row containing channel waveforms) simply by running the following command.

[R_F] = esRegionsExtract(D)

Note: Computing time for this function depends on the length of the data. For instance, with a sampling frequency of 500, it typically takes several minutes to analyze input with 1-5 minutes of patient data, while it can take several hours to analyze the input with 15-30 minutes of patient data. This non-linear increase in computing time lies in the threshold selection since longer data will involve more efforts in computing FD threshold. Empirically, it is recommended to keep the patient data at around 15 minutes to achieve maximum performance.

Utilities

To perform performance comparisons or to implement seperate modules for debugging purposes, use the following functions with the specifications:

Utilities Description
P_Matrix P_Matrix will output the patient's performance matrix specified in epilepsy review. The data workspace must contain all events informations specified in Dataset. Running this script will return a performance matrix summarized in a command line table.
FD The fractal dimension of signals be estimated with Katz’s algorithm. Kat'z Fractal Dimention (FD) calculation of a window can be implemented by F_D = FD(y). Note: scaling of y can affect F_D value given FD definitions.
FDM Implementation of FD module only. Input is the data matrix with rows containing all the channels. Parameters can be adjusted in the constant definitions section. This module can be implemented by [R] = FDM(D,framesize,step,chan_index), where D is the input data matrix and R is extracted regions
FD_Analysis FD analysis in a single iteration. This involves more detailed implementation such as state definitions and transitions. It can also adjust system parameters such as number of elements in TestBuffer. This function can be run by [regions] = FD_Analysis(frame_size,step,chan_index,data,coefficient).
SCSAM Implementation of SCSA detection module only. Input is the regions with each row containing boundaries. Ouput is the filtered regions with each row containing boundaries. This function can be run by [R_F] = SCSAM(D, regions, frame_size,step,chan_index).
SCSA SCSA reconstruction. Users can decompose the signal with a positive h value, by [yscsa,Nh,EigV,EigF] = scsa_build(h,y). function Input: h is a positive value, also known as the SCSA semi-classical constant. It can be selected by Quantum-based interval selection. y is the input signal to be decomposed. function Output: yscsa is the decomposed signal, Nh is the number of eigenfunctions. EigV is a matrix with eigenvalues on its diagonal (you can export the eigenvalues by diag(EigV). EigF is the matrix of eigenfunctions. UI interface of this tool is also available.

License

The application library (i.e. all code inside of the functions directory) is licensed under the GNU General Public License v3.0, also included in our repository in the COPYING file.