kinda does the job

Author: hamzaMahdi

EEG_SeizureDetector_teamA.m

@@ -0,0 +1,40 @@
+function gmarker = EEG_SeizureDetector_teamA(eeg)
+%The following is training of KNN clarification model with the EEG data 
+%given by Professor online
+
+features = load('features.mat'); %Importing already created feature matrix
+features = features.X; %access the matrix from struct
+gtrue = load('GT.mat');  %Importing already created GT marix
+gtrue = gtrue.Y; %access the matrix from struct
+%}
+%Finally Train the KNN model with the created GLOBAL feature & GT marixs 
+mdl = fitcknn(features,gtrue,'NumNeighbors',11,'Standardize',1);
+%mdl = load('mdl500.mat');
+
+%mdl = knn(); %This is creating the KNN model from scratch...20min for 1000
+%width...suggest not to do it.
+
+band = 5; %0.5-30 Hz
+window = 500; %Fast Processing
+
+%Extract the features from the testing signal (Also the remainder after
+%windowing
+[features,modulus] = eegmeasure(eeg,band,window); 
+
+%Run the test signal's features through the KNN clasification model
+result = predict(mdl,features); 
+
+%Reconstruct a full sized auto GT vector from the windowed version
+gmarker = [];
+for i = 1:length(result)
+    %If value is 0, then add 'window' number of 0 to the full sized version
+    if result(i) == 0 
+        gmarker = [gmarker;zeros(window,1)];
+    %If value is 1, then add 'window' number of 1 to the full sized version
+    else 
+        gmarker = [gmarker;ones(window,1)];
+    end
+end
+%Add zeros to fill the spaces that were cut when the signal was windowed
+gmarker = [gmarker;zeros(modulus,1)]; %The remainder that was removed (Better FP than FN) that is why it is zero
+end

GT.mat

@@ -0,0 +1,37 @@
+function [eeg] = eegfilter(signal,band)
+%This function is essentially the preproccesing portion 
+%Select the appropriate Frequency Band
+if(band == 1) %Delta
+    fc1 = 4;
+    fc2 = 0.5;
+elseif(band == 2) %Theta
+    fc1 = 8;
+    fc2 = 4;
+elseif(band == 3) %alpha
+    fc1 = 14;
+    fc2 = 8;
+elseif(band == 4) %beta
+    fc1 = 30;
+    fc2 = 14;
+else %Appropriate band for Seizures
+    fc1 = 30;
+    fc2 = 0.5;
+end
+
+fs = 256; %Sampling Freq
+
+eegraw = signal;
+
+%Normalize the signal
+eeg0 = eegraw - mean(eegraw); 
+eegnorm = eeg0./max(eeg0);
+
+%Create the Bandpass with the specified Frequencies
+[b1,a1] = butter(4,fc1/(fs/2),'low');
+[b2,a2] = butter(4,fc1/(fs/2),'high');
+
+%Pass the inputted Signal through the Bandpass filter
+eegfilthold = filter(b1,a1,eegnorm);
+eegfilt = filter(b2,a2,eegfilthold);
+eeg = eegfilt;
+end

eegfilter.m

@@ -0,0 +1,39 @@
+function [features,modulus] = eegmeasure(signal,band,w)
+[eeg] = eegfilter(signal,band); %Filter the inputted signal
+
+fs = 256;
+
+modulus = mod(length(eeg),w); %Remainder total size when divided by window size
+eeg = eeg(1:(length(eeg) - modulus));
+
+eegwin = zeros(w,length(eeg)/w); %Make matrix: Row=window size col=specific window
+
+for i = 1:(length(eeg)/w) %iterate through each window
+    eegwin(:,i) = (eeg(1+(i-1)*w:i*w)); %segment signal into the specific segments
+end
+
+%Measure each feature
+
+%RMS
+eegmean = zeros(length(eegwin(1,:)),1);
+for i = 1:length(eegmean)
+    eegmean(i,1) = rms(eegwin(:,i));
+end
+
+%Medium Freq
+eegfreq = zeros(length(eegwin(1,:)),1);
+for i = 1:length(eegfreq)
+    eegfreq(i,1) = medfreq(eegwin(:,i),fs);
+end
+
+%Entropy
+eegent = zeros(length(eegwin(1,:)),1);
+for i = 1:length(eegent)
+    eegent(i,1) = entropy(eegwin(:,i));
+end
+
+features = [eegmean,eegfreq,eegent]; %Output all 3 features together
+
+end
+
+

eegmeasure.m

@@ -0,0 +1,12 @@
+function ent  = entropy(signal)
+% Generate the histogram 
+nbins = 25;
+[n x] = hist(signal,nbins); 
+
+% Normalize the area of the histogram to make it a pdf 
+n = n / sum(n); 
+
+% Calculate the entropy 
+indices = n ~= 0; 
+ent = -sum(n(indices).*log2(n(indices))); 
+end

entropy.m

@@ -0,0 +1,27 @@
+function mdl = knn()
+%Machine Learning
+gtrue = []; %Initilize GT matrix of all samples
+features = []; %Initilize features matrix for all samples
+for subject = 0:20    %Iterate through each patient
+    
+    [EEG,seizureGT] = loadfile(subject);  %Load the specific patient
+    
+   for channel = 1:23 %Iterate through each channel 
+    band = 5; %All 0.5- 30Hz range
+    window = 500; %Fast Processing (Could change later) (took ~20 min)
+    
+    %Extract 3 features (RMS, MedFreq,Entropy)
+    [ftemp] = eegmeasure(EEG(channel).ch,band,window); 
+    %Extract the segmented GT of specific subject
+    [gtemp] = truthsegment(seizureGT,window);
+    
+    %Add the specific subject features to the GLOBAL feature matrix
+    features = [features;ftemp];
+    %Add the specific subject GT to the GLOBAL GT matrix
+    gtrue = [gtrue;gtemp]; 
+  end    
+end
+
+%Finally Train the KNN model with the created GLOBAL feature & GT marixs 
+mdl = fitcknn(features,gtrue,'NumNeighbors',5,'Standardize',1);
+end

features.mat

@@ -0,0 +1,17 @@
+function [EEG,seizureGT] = loadfile(subject)
+ n = num2str(subject);  %Convert the number into a string
+ 
+  %If statement corrects the numbering notations for subjects larger than 10
+   if subject < 10 
+    file = strcat('EEG_subject00',n,'.mat'); %Create the name of the subject file
+    truthfile = strcat('seizureGT_subject00',n,'.mat'); %Create the name of subject's GT file
+    else
+    file = strcat('EEG_subject0',n,'.mat'); %Create the name of the subject file
+    truthfile = strcat('seizureGT_subject0',n,'.mat'); %Create the name of subject's GT file
+   end
+   
+    EEG = load(file);  %Input the subject file
+    EEG = EEG.EEG;  %Input the subject file
+    seizureGT = load(truthfile); %Input the GT file
+    seizureGT = seizureGT.seizureGT; %Input the GT file
+end

features1000.mat

@@ -0,0 +1,22 @@
+
+subject = input('\nEnter Subject #:\n');    %Input the patient #
+channel = input('\nEnter Channel #:\n');    %Input the channel to test
+
+[EEG,seizureGT] = loadfile(subject);  %Load the EEG and GT files of the subject
+
+seizureMarker_auto = EEG_SeizureDetector_teamA(EEG(channel).ch); %Produce GT_auto
+
+performanceMetrics = validation(seizureMarker_auto, seizureGT); %Output the metrics (tp,fp,fn,tn,accuracy)
+
+performanceMetrics(5)
+fs = 256; %Sampling Frequency
+t = (0:1:length(EEG(channel).ch)-1)/fs;   % create time variable
+%Plot binary seizure variable on top of EEG.ch data
+%Note scaling of SeizureGT to make visible in plot
+figure; plot(t, EEG(channel).ch); 
+hold on; 
+plot(t,(max(EEG(channel).ch))*seizureGT, 'r');
+plot(t,seizureMarker_auto*(max(EEG(channel).ch)), 'k')
+xlabel('Time(s)');
+ylabel('Amplitude');
+legend('EEG Signal','GT','Auto');

gtrue1000.mat

@@ -0,0 +1,22 @@
+function [seize] = truthsegment(truth,w)
+%This function is exclusivly for the GT files
+modulus = mod(length(truth),w); %Remainder total size when divided by window size
+truth = truth(1:(length(truth) - modulus)); 
+truthwin = zeros(w,length(truth)/w);%Make matrix: Row=window size col=specific window
+
+for i = 1:(length(truth)/w) %iterate through each window 
+    truthwin(:,i) = (truth(1+(i-1)*w:i*w)); %segment signal into the specific segments
+end
+
+seize = zeros(length(truthwin(1,:)),1); %Check if there is a seizure in the window 
+for i = 1:(length(truthwin(1,:)))
+    for j = 1:w
+        if (truthwin(j,i) == 1)
+            seize(i,1) = 1;
+            break;
+        end
+    end
+end
+
+
+end

knn.m

@@ -0,0 +1,20 @@
+function  [metrics] = validation(auto,gt)
+window = 100;
+auto = truthsegment(auto,window);
+gt = truthsegment(gt,window);
+
+tp = 0; fp = 0; fn = 0; tn = 0;
+for i = 1:length(auto)
+   if  (gt(i) == 1) && (auto(i) == 1)   %TP
+       tp = tp+1;
+   elseif (gt(i) == 0) && (auto(i) == 1)    %FP
+       fp = fp+1;
+   elseif (gt(i) == 1) && (auto(i) == 0)    %FN
+       fn = fn+1;
+   else %TN
+       tn = tn+1;
+   end
+end
+    accuracy = ((tp+tn)/(tp+tn+fp+fn))*100;
+    metrics = [tp,fp,fn,tn,accuracy];
+end