Bimodal CNN for cardiovascular disease classification by co-training ECG grayscale images and scalograms

Abstract

This study aimed to develop a bimodal convolutional neural network (CNN) by co-training grayscale images and scalograms of ECG for cardiovascular disease classification. The bimodal CNN model was developed using a 12-lead ECG database collected from Chapman University and Shaoxing People's Hospital. The preprocessed database contains 10,588 ECG data and 11 heart rhythms labeled by a specialist physician. The preprocessed one-dimensional ECG signals were converted into two-dimensional grayscale images and scalograms, which are fed simultaneously to the bimodal CNN model as dual input images. The proposed model aims to improve the performance of CVDs classification by making use of ECG grayscale images and scalograms. The bimodal CNN model consists of two identical Inception-v3 backbone models, which were pre-trained on the ImageNet database. The proposed model was fine-tuned with 6780 dual-input images, validated with 1694 dual-input images, and tested on 2114 dual-input images. The bimodal CNN model using two identical Inception-v3 backbones achieved best AUC (0.992), accuracy (95.08%), sensitivity (0.942), precision (0.946) and F1-score (0.944) in lead II. Ensemble model of all leads obtained AUC (0.994), accuracy (95.74%), sensitivity (0.950), precision (0.953), and F1-score (0.952). The bimodal CNN model showed better diagnostic performance than logistic regression, XGBoost, LSTM, single CNN model training with grayscale images alone or with scalograms alone. The proposed bimodal CNN model would be of great help in diagnosing cardiovascular diseases.

ECG Heartbeat classification using Progressive Moving Average Transform

“…While many of these methods have demonstrated good performance, they have often been applied primarily to the MIT-BIH database, making it challenging to verify their stability across different databases. Two methods presented in Huang et al 6 and Yoon et al 25 offer viable comparisons with our method as they utilize STFT and CWT transforms, respectively, which are similar to our PMAT transform. However, the first method 6 , employing STFT, was tested on a very limited subset of size 2520 and not on the entire MIT-BIH database, thus resulting in comparisons that diverge from our results.…”

“…However, the first method 6 , employing STFT, was tested on a very limited subset of size 2520 and not on the entire MIT-BIH database, thus resulting in comparisons that diverge from our results. On the other hand, the second method 25 , utilizing the CWT transform, was applied to a larger database and yielded promising results. Nonetheless, the authors did not assess the method's stability across different databases, raising questions about its generalizability.…”

2D Materials in Flexible Electronics: Recent Advances and Future Prospectives

“…In addition, CNN may also be used to analyze spectrograms, a two-dimensional representation of the frequency response of time-series data. 1015,1016 Spectrograms can be generated from various biosignals, such as ECG and EEG. On the other hand, RNN models are fit for analyzing continuous time-series data from wearable sensor arrays.…”

“…CNN models are quite useful in analyzing steady-state data, such as PD arrays, where the data can be presented as a grid of numerical values. − CNN models will extract features from such arrays via convolutional layers. In addition, CNN may also be used to analyze spectrograms, a two-dimensional representation of the frequency response of time-series data. , Spectrograms can be generated from various biosignals, such as ECG and EEG. On the other hand, RNN models are fit for analyzing continuous time-series data from wearable sensor arrays.…”

Estimation of Left and Right Ventricular Ejection Fractions from cine-MRI Using 3D-CNN

“…Deep learning techniques, specifically CNNs, have been applied to various medical image analyses [ 8 , 9 , 10 , 11 ]. CNNs are widely used for image classification [ 12 , 13 , 14 ] regression [ 15 , 16 , 17 ], object detection [ 18 , 19 ], super resolution [ 20 , 21 ], and semantic segmentation [ 22 , 23 , 24 ]. Recent studies have proposed automatic segmentation of the left ventricle lumen to reduce tracing time and interobserver errors in the study of cardiac function [ 25 ].…”