ECG Signal Analysis Based On Wavelet Transform And Deep Learning In Heart Disease Detection

Due to the rising mortality rate caused by heart diseases worldwide,research into electrocardiogram(ECG)signals has piqued the interest of many academics.Heart disease problems rise year after year due to increasing stress and an unhealthy diet.According to statistics,people with type 2 diabetes mellitus(T2DM)are twice as likely to develop heart disease than people without diabetes.Cardiovascular disease symptoms,like diabetes itself,can go unnoticed for years;many people with T2 DM will only show signs of heart disease complications when they are diagnosed.As a result,locating the victims ahead of time will save lives.The ECG test is simple to perform,safe,and relatively inexpensive.Therefore,it remains the gold standard in identifying heart diseases.However,obtaining quality ECG signals that lead to proper treatment is not easy.The ECG signals are tiny in amplitude and very sensitive to noise.Therefore,they can easily be affected by physiological artefacts and other interference while processing.Thus,a well-capable processing algorithm that guarantees quality ECG signals is necessary to avoid misinterpretation.This research offers less costly,effective,and precise methods for dealing with the mentioned problem.The study investigated ECG signals of patients subjected to type 2diabetes mellitus and non-diabetic patients,and then a comparison was made.The workflow is as follows: Firstly,we de-noised all the selected ECG records to enhance the excellent performance of the proposed model.Then,the proposed algorithms were used to detect and confirm the presence of the QRS complex,which is a vital feature in obtaining the heart’s structure,function,and condition.Finally,build a deep learning model that will detect cardiac disorders.The research embodied three main aspects as follows:(1)The purification of all the ECG signals that are used in this study ensures quality signals,which are key to proper analysis and treatment.The Discrete Wavelet Transform(DWT)filtering approach,in particular the Daubechies(db6)wavelet,was used to remove all artefacts from the selected ECG signals.The db6 wavelet was selected because it resembles the real ECG’s morphology.The ECGs were passed through multiple levels of decomposition and recontraction based on their signal length.And with the help of MATLAB code,the noise such as baseline wander,electrode motion,and power frequency interference were eliminated.(2)To identify and detect the QRS complex’s presence,which would be utilized to compute the heart rate.This section’s main goal was to find the association between heart rate and T2 DM.The R-peak method based on the Discrete Wavelet Transform(DWT)called Symlet4 was used to calculate the heart rate of both T2 DM and non-T2 DM patients.When executed in MATLAB,the experimental results exposed that diabetes mellitus is linked to elevated heart beat(Tachycardia Arrhythmia).The heartbeats of people subjected to T2 DM were higher above 100 bpm than those who did not have T2 DM.Therefore,concluded that fast heart rate is linked to type 2 diabetic Mellitus.However,heart rate itself is not enough criteria to judge if a patient has severe heart disease or not.Thus,this work went further by proposing the deep learning method for auto-detection of cardiac diseases.(3)To utilize the strength of deep learning to detect the presence of heart diseases.The regression analysis of two different Convolutional Neural Networks(CNNs): Alex Net and Google Net,was done.The Continuous Wavelet Transform(CWT)was utilized to convert 1-dimensional ECG images into 2-dimensional images called scalograms.The Convolution Neural Network(CNN)extracted ECG signal features and diagnosed heart illness in the study.The obtained scalograms were fed into two transfer learning networks(Alex Net-CNN and Google Net-CNN)and then subjected to regression analysis.The MATLAB platform was used in designing,executing,and testing all the experimental work in this section.The observations from proposed deep learning models for heart disease detection are as follows:The performance of Alex Net-CNN was better by 1%,whereby 99.00 and 98.00 were the general accuracy for Alex Net-CNN and Google Net-CNN,respectively.Despite the proposed methodologies being less expensive and straightforward to execute,they gave the most accurate results.Therefore,the proposed methods can be considered in helping doctors to diagnose heart diseases.