ECG signal is one of the greatest initial sources to obtain diagnostic information. In recent research, an effective computational technique has been evolved in processing and analyzing the ECG signal. The major challenging research using the ECG signal is to provide a computational technique based on signal processing to diagnose a heart disease which includes Bundle Branch Block, Left Ventricular Hypertrophy (LVH), and Myocardial Infarction. The ECG with 12 leads helps to diagnose Myocardial Infarction and is useful in predicting complications in cardiovascular disease [1]. However, it has some limitations. The ECG signal with the presence of Myocardial Infarction is difficult to interpret with other electrocardiographic abnormalities such as Left Bundle Branch Block (LBBB) and Left Ventricular Hypertrophy disease [2]. The ECG patterns of Left Bundle Branch Block and Left Ventricular Hypertrophy may resemble the ECG findings related to myocardial Infarction. These patterns decrease the clarity of the ECG signals to detect Myocardial Infarction [2]. Bacharova et al. [3] reported that left ventricular hypertrophy induced modification in the QRS and T pattern, and based on Romhilt-Estes Score, the amplitude of R wave and S wave was greater than 2.0 mV and induced depression in the ST segment. Gubner reported that the sum of the amplitude of the R wave in lead I and S wave in lead III was greater than 25 mV indicated in the LVH. Sokolow Lyon reported that the sum of amplitude of S Wave in V1 lead and R wave in lead V5 caused an increase in voltage by 3.5 mV. Based on Cornell voltage the sum of amplitude of R wave in avL and S wave in V3 was greater than 2.0 mV [3]. Gosse et al. [4] reported that the amplitude of the R wave in lead aVL produced a huge change in the ECG signal for identifying LVH. This approach was simple and cost-effective. Increases in voltage and length of the QRS complex, which are suggestive of a rise in LVM and repolarization changes considered to indicate anomalies in myocardial perfusion, are the most common signs [4]. de la Garza-Salazar et al. [5] reported that the LVH could be diagnosed using ECG and that it was independent of the parameter called ventricular mass. The measurements of ECG parameters were complex and provided a low accuracy for the diagnosis of LVH. Efficient machine learning algorithms were required for the diagnosis of LVH with the highest accuracy [5].

The Electrocardiographic pattern of Left Ventricular Hypertrophy induces a great change in the ST segment and T waves. It shows depression in the ST segment with negative T waves and elevation in the ST segment with positive T waves. This changed pattern may mean that the patient with the ECG signal has a condition that is compatible with Myocardial Infarction [6–8]. The LVH based ECG signal is shown in Fig. 1. An EKG with an R wave in lead I (17 mm) that is greater than 14 mm is seen below. The R wave in lead V5 and/or V6 is about 24 mm, while the S wave in lead V1 is 21 mm. According to Sokolow-Lyon guidelines, the total is 45 mm, which is higher than 35 mm, suggesting left ventricular hypertrophy.

Myocardial Infarction (MI) is the origin of coronary artery disease. It is also known as a heart attack. This is because of the occlusion of any one of the above coronary arteries [9]. The ECG signal with Myocardial Infarction is shown in Fig. 2. It represents the normal ECG signal and the ECG signal with MI. The synthetic ECG [notice Fig. 2a] is only seen for wave morphologies comparison. In a clinical setting, all of the above modifications (Figs. 2b–2d) can be present during an infarction. This demonstrates the importance of evaluating a 12-lead ECG for MI identification and localization.

Another common observation in patients with the ECG signal associated with the LBBB and Myocardial Infarction is revealed that there is a discordance of the ST segment, QRS complex, and T waves. That is, the ECG signal shows ST-segment elevation with negative QRS complexes and depression in the ST segment along with negative T wave and positive QRS complexes [11]. The ECG with LBBB is shown in Fig. 3. Traditional methods to diagnose these diseases are cumbersome. The Electrocardiogram criteria for Left Bundle Branch Block, Left Ventricular Hypertrophy mimics the criteria for the diagnosis of acute Myocardial Infarction. This may lead to a wrong diagnostic and treatment procedure. Fig. 3 represents ECG with Left Bundle Branch Block. In leads I, aVL, V5, and V6, the R wave is wide and notched. The RS pattern can sometimes be shown in leads V5 and V6. In leads I, V5, and V6, there are no q waves. In lead aVL, a small q wave can be present. The ST and T waves are usually in the opposite direction of the QRS effect. Positive T waves in leads with positive QRS are likely normal (positive concordance). In leads with negative QRS (negative concordance), a depressed ST section and/or negative T wave are abnormal. As LBBB progresses, the mean QRS axis can move to the right or left.

The ST segment in the ECG signal pattern indicates the period that occurs between ventricular depolarization and repolarization. The ST segment is the isoelectric part of the ECG signal between the terminal point of the S wave and the starting point of the T wave. The elevation in the ST segment indicates heart abnormalities. Fig. 4 denotes the normal and abnormal ECG signal patterns. This section describes the research work which investigates the heart disease abnormalities that cause changes in the ST-segment pattern of the ECG signal. Tab. 1 indicates diagnosis criteria for pathological conditions using ECG signals.

Noriega et al. [12] evaluated whether the existence of multivessel coronary artery disease (CAD) varied the ST-segment of the ECG signal in patients with acute coronary artery occlusion. It was noticed that both the cases with a single or multivessel CAD had inverse depression in the ST-segment, which was due to the occlusion of the LAD artery. It could lead to a major infarction [12]. Rossello et al. [13] predicted whether QRS and QT duration differentiated between pericarditis and acute myocardial ischemia. It was observed that the elongation of the QRS complex and the reduction in QT interval in the ECG signal was noticed in patients who suffered from acute STEMI but not with pericarditis. A diffuse elevation of the ST segment and an upright divergence of the PR segment with ST-segment depression in lead aVR is common electrocardiogram (ECG) findings in patients with acute pericarditis [13]. Yildirim et al. [14] reported different clinical pieces of evidence that the ECG signal of myocarditis was similar to that of Myocardial Infarction (MI). The reciprocal of the ST-segment elevation was observed in the ECG signal for acute myocarditis cases. It can have a variety of clinical symptoms and can be interpreted as myocardial infarction (MI) because patients typically have chest pain and electrocardiographic changes that are similar to those seen in acute ST-elevation MI [14]. Willemsen et al. [15] analyzed three cases of chest pain. In all these cases, the clinical diagnosis for the general practitioner was unambiguous for conditions such as Acute Coronary Syndrome and light pain in the chest. A General Practioner’s ability to distinguish between ACS and less serious sources of chest pain remains a challenge [15]. Coppola et al. [16] reported on various conditions that induced changes in ST-segment Elevation Myocardial Infarction (STEMI). When septal hypertrophy is occurring, the ECG reveals R waves with a high voltage in the anterolateral leads and Q waves in the anterior and inferior leads. The T waves in V2 and V4 are always very deep and inverted, resembling a non-Q AMI [16]. Smith et al. [17] reported that the elevation in ST-segment caused difficulty in differentiating Myocardial infarction from early repolarization of ECG signal. For a variety of causes, it can be impossible to determine the two entities apart. First, although upward ST-segment concavity is typically associated with normal ECG findings, it is also present in 30% to 40% of anterior STEMI (due to occlusion of the left anterior descending artery), particularly early after the onset of symptoms. Furthermore, 30% to 40% of anterior STEMIs have borderline ST-segment elevation [17]. Dodd et al. [18] analyzed Left Bundle Branch Block ECG signal cases with and without Myocardial Infarction and identified variations in the amplitude of QRS complexes, the morphology of ST-segment, and the T waves of ECG signal, and he compared them with modified Sgarbossa criteria to find them perform well against other variations [18]. Pollak et al. [19] reported that the ECG pattern minimized an accurate diagnosis of acute coronary syndrome because of the mystifying ECG patterns of Left Ventricular Hypertrophy, ventricular paced rhythms, and Left Bundle Branch Block. The confound electrocardiographic patterns are those that minimize the electrocardiogram’s ability to detect modifications relative to ACS, not because the results resemble STEMI, but because the syndrome obscures the identification of ST-segment elevation [19]. Nable et al. [20] reported that the changes in ST-segment elevation in ECG signal were a key factor for the diagnosis of ST-segment elevation Myocardial Infarction (STEMI). Particularly, this change in ST-segment elevation pattern could also be perceived in the other diseases which included Takotsubo cardiomyopathy, left ventricular hypertrophy, left bundle branch block, and early repolarization [20].

Several efficient algorithms have been evolved for disease identification using the ECG signal. The morphologies of the ECG signal play a vital role in its analysis. Various Signal Processing techniques have been adopted to grasp information from the ECG signal. Various techniques have been investigated to identify the characteristics related to coronary artery disease. The detection of Q, R, S, and T wave amplitudes using signal processing techniques is an important indicator to identify heart diseases.

However, it is troublesome to perceive the differences in the ECG signal to indicate a specific type of heart disease. Hence, an automated heart disease diagnosis system has been devised in the medical field to find a solution to this problem. The data employed in this system are required to be processed and classified accurately. For this purpose, various techniques have been reported in the literature so far.

The statistical diagnostic strategies for predicting cardiac diseases based on ECG signals are summarized in this paper. The technique of ECG signal analysis based on machine learning is explored from preprocessing, feature extraction, feature selection, and classification. End-to-end models for ECG analysis based on deep learning algorithms have been summarized, removing the need for feature extraction using hand-crafted techniques from the analysis process.

This article describes the work reported in various pieces of literature for the diagnosis of cardiac diseases. The structure of the literature review is organized in this chapter as follows: ECG signal database, various pathological-condition-based ECG signals, ECG signal preprocessing, Feature Extraction techniques, and classification techniques. The advantages and disadvantages of various research works are also discussed in detail to identify the appropriate technique to diagnose and differentiate between various cardiac signals indicating heart diseases such as Myocardial Infarction, Bundle Branch Block, and Left Ventricular Hypertrophy. The flow of ECG signal processing technique to diagnose these diseases indicated in Fig. 5

The vast majority of research used computational approaches for Ischemic heart disease and myocardial infarction (MI) are detected automatically using publicly available datasets containing ECG waveforms to test the efficacy of their methods. The Physikalisch-Technische Bundesanstalt (PTB), The European ST-T (EST) and MIT-BIH arrhythmia Database as well as others, are all accessible via the Physionet data repository [21]. Fig. 6 indicates the database mainly used for the diagnosis of Myocardial Infarction, Bundle Branch Block, and Left Ventricular Hypertrophy.

Preprocessing is the preliminary stage involved in the implementation of any proposed approach. ECG signals are commonly affected by noises such as muscle artifacts, electrode motion, powerline interference, and baseline wander [25]. Muscle artifacts are caused by muscle activity. The movement in the position of the electrode introduces an electrode motion [26,27]. Powerline Interference is generally present in the ECG signal that contains a 50/60 Hz sinusoidal signal and harmonics [28]. Baseline wander causes modifications in the baseline of the ECG signal during respiration. These noises could affect the P, Q, R, S, and T segments of the ECG signal, the frequency resolution, and signal quality, and generate amplitudes in the ECG signals that could mimic PQRST waveforms. It made the analysis of the ECG signal more difficult, often misdirecting the physician to make a false diagnosis [29,30]. Hence, the cancelation of these noises present in the ECG signal was an essential task for further processing.

Different techniques have been presented in the literature for removing the noise present in the signal. The adaptive filtering technique based on discrete Wavelet Transform and artificial neural network is applied to the ECG signal, and it shows a significant improvement in Signals to Noise Ratio (SNR). But the disadvantage is that it is restricted to remove only certain types of noises. This technique reduces computational complexity [31]. The hybrid denoising techniques such as Ensemble Empirical Mode Decomposition (EMD) along with Block Least Mean square technique and Discrete Wavelet Transform (DWT) along with Neural Network were investigated and compared with DWT Thresholding method for eliminating the noise present in the signal. These techniques showed better performance than DWT thresholding did [32]. The presence of baseline wanders induced noise within 1 Hz which caused changes in the ST-segment. To reduce the noise Framelet Transform was applied and compared with DWT. The framelet transform showed better performance than DWT did [33]. The wavelet decomposition with level 2 was applied to the ECG signal and it showed that the approximation coefficient was similar to the low-frequency ECG signal. The higher-order decomposition coefficients emerged as the approximation coefficient with distorted ECG signals. So, it minimized the use of Non-Local Means (NLM). The effectiveness of both DWT and NLM was combined and found to be more efficient to remove the noise than other techniques were [34]. The ECG signal processing technique was applied based on Empirical Mode Decomposition (EMD) and the improved approximation envelope method. The Butterworth lowpass filter was implemented before applying it to EMD to eliminate high-frequency noise. This method eliminated the baseline wander and power line interference present in the ECG signal [35]. Jenkal et al. [36] proposed a method to denoise the ECG signal affected by various noises due to EMG signal, high frequency, and power line interferences. This technique depended on the DWT decomposition, and an Adaptive Dual Threshold Filter (ADTF). This technique produced a better result for the noise with higher density [36]. Sharma et al. [37] implemented a wavelet-based technique for the removal of the noise present in the electrocardiogram signal. This technique involved thresholding the coefficients of wavelets at various sub-bands. It was noticed that the utmost energy present in the signal existed in cD4, cD5, and cA5 sub-bands. The noise existed in the lower order sub-bands [37]. AlMahamdy et al. [38] presented various algorithms includes Savitzky-Golay filtering, adaptive filters, and discrete wavelet transform to remove the noise in the ECG signal. The NeighBlock wavelet method outperforms the other techniques. In some mid-range SNRs, however, the RLS and Savitzky-Golay filters work better [38]. Fig. 7 denotes the different methods with SNR improvement in decibel. Tab. 2 represents the different denoising techniques using ECG signals.

Feature extraction in the ECG signal is one of the essential processes for classification and different types of features are extracted to describe the ECG signal. The extracted features from ECG signals play a significant role in diagnosing most cardiac diseases, according to Sujan et al. [46] These features include temporal, statistical, and morphological features [47–50]. The feature extraction stage aims to find the simplest number of features that result in acceptable suitable classification rates [51]. It extracts valuable data from ECG signals [52]. Fig. 8 indicates various feature extraction techniques to diagnose MI, BBB, and LVH.

The performance of the classification is improved by extracting a combination of features mentioned in the previous section [112]. The major issue in the diagnosis of heart disease is that normal ECG signals vary from person to person, and the same disease induces various signs in the ECG signal for different patients. The diagnosis of heart disease using the ECG signal is difficult. Hence, the classification of the ECG signal plays a significant role in the diagnosis of various heart diseases. Many researchers implemented various types of conventional classifiers and artificial Neural Network classifiers for the diagnosis of heart diseases. Different feature extraction and classification techniques have been presented in the literature for obtaining better results from the ECG signal.

This study imparts details of various techniques that are associated with heart disease which induces changes in the ST segment. It describes the performance of each method and provides the scope for further development to explorers. Initially, various signal processing algorithms were implemented to diagnose various heart diseases. Later, different machine learning and deep learning approaches have been proposed to diagnose various diseases. These approaches provide a better result than the conventional techniques.

The various techniques implemented to diagnose Bundle Branch Block, Myocardial infarction, and Left Ventricular Hypertrophy. Most researchers implemented a hybrid technique that involves feature selection and a neural network approach. The feature selection technique along with neural network results shows that convergence speed is high. Generally, the performance of various feature selection and neural network classifiers to diagnose these diseases is nearly 98% accurate. For high-dimensional data, the computation complexity is high in machine learning classifiers. The deep learning approaches extract features implicitly and integrate them. The main disadvantage is higher computational complexity. The existence of a greater number of layers may increase computational time. Several factors may affect the performance of classifiers, which includes several data’s in the training set, and the existence of weights and biases in the neural network. Figs. 10–12 show the percentage of accuracy of feature selection, deep learning, and neural network techniques to diagnose. The figures represent that the neural network-based approach provides better results than the feature selection approach along with the conventional classifier.

Previous research on the diagnosis of these heart diseases was based on statistical and morphological features, wavelet transform, neural networks, and deep learning approaches. The deep learning approach requires huge data for analysis. It leads to time consumption and computational complexity. Tab. 7 indicates the shortcomings and advantages of various features to diagnose these diseases. Only very few researches have been implemented in the diagnosis of LVH. Most researchers proposed neural network approaches to diagnose LVH [135]. The method does not necessitate any advanced knowledge of the ST section. The estimated features can be used for automated LVH labeling, reducing the need for manual annotation and allowing for faster LVH diagnosis. The limitations are that it requires more computational time.

Tab. 5 indicates the performance analysis to diagnose these heart diseases using the MIT-BIH dataset. The PTB diagnostic database is used widely to diagnose Myocardial Infarction, Bundle Branch Block, and Left Ventricular Hypertrophy. Tab. 6 indicates the classification analysis of these heart diseases based on ResNet and VGGNet. This article benefits researchers to seek various hybrid approaches to differentiate these diseases.

This paper describes the overview of heart diseases that induce changes in the ST segment of the ECG signal. The process of analysis of ECG signal is reviewed in different sections which include diagnosis of Myocardial Infarction, Bundle Branch Block, and Left Ventricular Hypertrophy. Left Bundle Branch Block and Left Ventricular Hypertrophy are two similar conditions. Since patients normally complain of chest pain and the electrocardiographic changes mimic those seen in acute ST-elevation MI, it is called a myocardial infarction (MI). It has the potential to mislead the diagnostic process. A General Practitioner’s task is made more difficult by the fact that distinguishing between Acute Coronary Syndrome and less serious causes of chest pain. The majority of scientific work is focused on the automatic diagnosis of Myocardial Infarction using ECG signal, with a few works based on the automatic diagnosis of Left Ventricular Hypertrophy using ECG signal. Several issues with the automated classification of these diseases have been posed by researchers. The MIT-BIH and PTB diagnostic databases are used to present results in the literature. The limited number of databases available is a significant impediment to progress in research based on the fully automated classification of these diseases in ECG. To enhance the observed performance, some investigators used hybrid approaches. This is further described in various subsections such as different signal processing and classification techniques. Signal processing and classification techniques for the diagnosis of Myocardial infarctions in specific positions of coronary arteries in the heart, such as Left Anterior Descending Artery (Anterior position), Left Circumflex (Lateral position), and Right Coronary Artery (Inferior and Posterior position), can be investigated. Another essential direction is the investigation of these heart diseases using hybrid approaches.