Search

BACKGROUND The electrocardiogram (ECG) is one of the most common non-invasive diagnostic tools that can provide useful information regarding the patient’s health status. Deep learning (DL) is a current area of intense exploration that leads the way in most attempts to create powerful diagnostic models based on physiological signals. OBJECTIVE This paper provides a systematic review of DL methods applied to ECG data for various clinical applications. METHODS We identified 230 relevant articles published between January 2020 and December 2021 and provided a complete account of the state-of-the-art DL strategies by reporting on the number and type of hidden layers, the ECG data sources, the data preprocessing techniques, and the data splitting strategies for each one of them. RESULTS We provided a complete account of the state-of-the-art DL strategies by reporting on the number and type of hidden layers, the ECG data sources, the data preprocessing techniques, and the data splitting strategies for each one of them. We also present open research problems and point out potential gaps regarding the design and implementation of DL models. CONCLUSIONS We expect this review will provide insights into the state-of-the-art DL methods applied to ECG data and point to future directions for research on DL to create robust models that can assist medical experts in clinical decision-making.

Respiratory diseases constitute one of the leading causes of death worldwide and directly affect the patient’s quality of life. Early diagnosis and patient monitoring, which conventionally include lung auscultation, are essential for the efficient management of respiratory diseases. Manual lung sound interpretation is a subjective and time-consuming process that requires high medical expertise. The capabilities that deep learning offers could be exploited in order that robust lung sound classification models can be designed. In this paper, we propose a novel hybrid neural model that implements the focal loss (FL) function to deal with training data imbalance. Features initially extracted from short-time Fourier transform (STFT) spectrograms via a convolutional neural network (CNN) are given as input to a long short-term memory (LSTM) network that memorizes the temporal dependencies between data and classifies four types of lung sounds, including normal, crackles, wheezes, and both crackles and wheezes. The model was trained and tested on the ICBHI 2017 Respiratory Sound Database and achieved state-of-the-art results using three different data splitting strategies—namely, sensitivity 47.37%, specificity 82.46%, score 64.92% and accuracy 73.69% for the official 60/40 split, sensitivity 52.78%, specificity 84.26%, score 68.52% and accuracy 76.39% using interpatient 10-fold cross validation, and sensitivity 60.29% and accuracy 74.57% using leave-one-out cross validation.

BACKGROUND Electrocardiogram (ECG) recording and interpretation is the most common method used for the diagnosis of cardiac arrhythmias, nonetheless this process requires significant expertise and effort from the doctors’ perspective. Automated ECG signal classification could be a useful technique for the accurate detection and classification of several types of arrhythmias within a short timeframe. OBJECTIVE To review current approaches using state-of-the-art CNNs and deep learning methodologies in arrhythmia detection via ECG feature classification techniques and propose an optimised architecture capable of different types of arrhythmia diagnosis using publicly existing annotated arrhythmia databases from the MIT-BIH databases available at PHYSIONET (physionet.org) . METHODS A hybrid CNN-LSTM deep learning model is proposed to classify beats derived from two large ECG databases. The approach is proposed after a systematic review of current AI/DL methods applied in different types of arrhythmia diagnosis using the same public MIT-BIH databases. In the proposed architecture the CNN part carries out feature extraction and dimensionality reduction, and the LSTM part performs classification of the encoded ECG beat signals. RESULTS In experimental studies conducted with the MIT-BIH Arrhythmia and the MIT-BIH Atrial Fibrillation Databases average accuracies of 96.82% and 96.65% were noted respectively. CONCLUSIONS The proposed system can be used for arrhythmia diagnosis in clinical and mHealth applications managing a number of prevalent arrhythmias such as VT, AFIB, LBBB etc. The capability of CNNs to reduce the ECG beat signal’s size and extract its main features can be effectively combined with the LSTMs’ capability to learn the temporal dynamics of the input data for the accurate and automatic recognition of several types of cardiac arrhythmias. CLINICALTRIAL Not applicable.

Atrial fibrillation is a heart arrhythmia strongly associated with other heart-related complications that can increase the risk of strokes and heart failure. Manual electrocardiogram (ECG) interpretation for its diagnosis is tedious, time-consuming, requires high expertise, and suffers from inter- and intra-observer variability. Deep learning techniques could be exploited in order for robust arrhythmia detection models to be designed. In this paper, we propose a novel hybrid neural model utilizing focal loss, an improved version of cross-entropy loss, to deal with training data imbalance. ECG features initially extracted via a Convolutional Neural Network (CNN) are input to a Long Short-Term Memory (LSTM) model for temporal dynamics memorization and thus, more accurate classification into the four ECG rhythm types, namely normal (N), atrial fibrillation (AFIB), atrial flutter (AFL) and AV junctional rhythm (J). The model was trained on the MIT-BIH Atrial Fibrillation Database and achieved a sensitivity of 97.87%, and specificity of 99.29% using a ten-fold cross-validation strategy. The proposed model can aid clinicians to detect common atrial fibrillation in real-time on routine screening ECG.