Research On ECG Denoising Algorithm Based On Improved Guided Filter

According to the June 2017 report of the World Health Organization(WHO),cardiovascular disease caused 1.77 billion deaths each year,accounting for the global death toll of 31%.As a non-invasive technique,electrocardiograph is the most effective exam in the diagnosis of cardiovascular disease.Especially,multi-lead ECG signals are widely used in clinic for it can comprehensively reflect the activity of the heart,then accurately determine the specific location of the disease.However,most of these noises alter the morphological properties of ECG beats which are vital for physician's correct diagnosis.Therefore,filtering out the noises is the premise and guarantee for disease accurate diagnosis.In this paper,a new ECG signal denoising method with the improved guided filter is proposed.The main works are as follows:(1)Based on the finding that ECG signals share significant similarities in the morphology for a particular person,the paper construct a guided filter and reform it by a Butterworth highpass filter.The Butterworth high-pass filter is utilized to remove the baseline wander.We derive a template signal from the average of a selected number of ECG cycles,and then construct a guided signal by replacing the main part of each ECG cycle in the original signals with the template signal.Thus the guided signal can inherit most morphological characteristics from his/her physiome.The advantageous edge-preserving guided filter is applied to remove the rest noise,of which frequencies are between the ECG signals.Using the proposed method,the output SNR can reach as high as 19.28 dB,and the average RMSE is less than 0.41.Our experimental results validate that the proposed method can effectively denoise the complex noise.(2)Based on the nonlinear relationship of multi-lead ECG signals and the significant differences in the leading time of noise appearing on different leads,the paper proposed multilead model-based ECG signal denoising by guided filter.In the background of ECG big data,the proposed method constructs the statistical model between the multi-lead ECG signals using sparse autoencoder(SAE).In this way,both the normal and abnormal beats can be preserved in the statistical model.The guided signal is obtained by fusing the predicted signals,and contains comprehensive morphological features.The proposed method not only improves the accuracy of noise reduction,and conserves the detail features.Our experimental results demonstrate that the signal-to-noise ratio(SNR)improvement of the proposed method can reach as high as 21.54 dB,and the mean squared error(MSE)is less than 0.0401.The proposed algorithm outperforms previous algorithm with regard to conserve the detail features or the abnormal ECG beats.