Adaptable Noise Reduction of ECG Signals for Feature Extraction

An Approach Based On Wavelet Decomposition And Neural Network For ECG Noise Reduction

“…In the modeling application, [29] had shown that a learning of a function that [time ma1 ma2] = textread('data/noises_ma.txt','%f %f %f'); random_noise = randn (1,size(time,1))'/5; %data files data_full_length = size(time,1); data_train_start = wInput; data_train_end = train_length; data_test_start = train_length + wInput; data_test_end = train_length + test_length; % -layer (1).node x = 1:data_full_length; ECG_clean = ecg1; wn1 = 0.5*(rand(data_full_length,1)-0.5*ones(data_full_length,1)); pw1 = 1/2 * sin(2*pi*x*1/360*60)'; ECG_noisy = ecg1 + ma1 + wn1 + bw1 + pw1 + em1; %assign data input_data = ECG_noisy; %(1:data_train_end+layer(1).node+1); reference_data = ECG_clean; %(1:data_train_end+layer(1).node+1); %% Normalization dmin = min( min(input_data), min(reference_data)); dmax = max( min(input_data), max(reference_data)); input_data = normalization(input_data,dmin,dmax); reference_data = normalization(reference_data,dmin,dmax); %% Initialize the parameter last_layer = size(layer,2); input_length = layer(1).node; layer (1).weight (1:layer(2).node) = 1; %always kept as 1 layer (1).output = zeros (1,layer(1).node); layer (1).bias = 0; %rand(); %no bias for first layer for i=2:size(layer,2)-1, layer(i).weight = rand(layer(i).node,layer(i-1).node)-0.5; layer(i).output = zeros (1,layer(i).node); layer(i).bias = rand (1,layer(i).node)-0.5; end %layer(total_layer).input (1:layer(total_layer-1).node) = 0; layer(last_layer).weight = rand(layer(last_layer).node,layer(last_layer-1).node)-0.5; layer(last_layer).output = zeros (1,layer(last_layer).node); layer(last_layer).bias = rand (1,layer(last_layer).node)-0.5; error(1) = 0; %% Learning tic; for i=1:iteration, %iter = i for jj=data_train_start:data_train_end, j = ceil((data_train_end-wInput). *rand(1,1))+wInput; X = input_data(j-wInput+1:j)'; cof = wavelet(W, L, X); layer (1).output = cof( Ainput(WaveParaChoice).len ); reference = reference_data(j-layer(last_layer).node+1:j)'; for k=2:last_layer, for m=1:layer(k).node, layer(k).weight_sum(m) = sum(layer(k-1).output. *layer(k).weight(m,:))-layer(k).bias(m); layer(k).output(m) = activation_function(layer(k).weight_sum(m)); end %m end %k output_train(j-layer(last_layer).node+1:j) = layer(last_layer).output; error(j-layer(last_layer).node+1:j) = referenceoutput_train(j-layer(last_layer).node+1:j); layer_old = layer; layer(last_layer).delta = error(jlayer(last_layer).node+1:j) .…”

“…*layer_old(k+1).weight,1); clear delta_tmp; for m=1:layer(k-1).node, %expanding dimention by repeat for calc next line delta_tmp(:,m) = layer(k).delta; end %m layer(k).weight = layer_old(k).weight + learning_rate * output_tmp . * delta_tmp; layer(k).bias = layer_old(k).bias + learning_rate * (-1) * layer(k).delta; end %k end %j subplot(totalplot,1,1); plot(i,top) title('Noise-free ECG');xlabel('Sample');ylabel('Amplitude (mV)');grid on; subplot(totalplot,1,2); plot(i,mid) title('Noisy ECG');xlabel('Sample');ylabel('Amplitude (mV)');grid on; subplot(totalplot,1,3); plot(i,bot) title('Filtered ECG');xlabel('Sample');ylabel('Amplitude (mV)');grid on; noisy = calcSNR(input_data(1:data_test_end),reference_data(1:data_test_end),dat a_test_start+wInput,data_test_end-1,'data/atr_103.txt') filtered = [permute(Xe -xfir(Seq{1,1},Seq{1,2},Xo,Ext), [2,1,3]), ... permute (Xo,[2,1,3] …”

“…where ‫ݎ‬ is a random real number uniformly distributed between [0,1]. Adjustable parameters ‫ݍ‬ ∈ [0, 1] determine the probability of the search direction.…”

“…The very top signal is the original noisy ECG. The next following 8 plots (d [1][2][3][4][5][6][7][8] are decomposed signals of the wavelet coefficient. Lastly, the very bottom signal is a scaling signal.…”

“…Moreover, several electrical and mechanical noise components are also added to the signal, making it difficult to extract key features. In general, measured ECG signal data contains white noise, muscle artifact noise, baseline noise, electrode moving artifact noise, and 60Hz power line noise [1]. In the effort to remove these noises, there has been little success when employing traditional methods such as linear filers, signal averaging, and their combination.…”