The Effectiveness of a Deep Learning Model to Detect Left Ventricular Systolic Dysfunction from Electrocardiograms

Abstract

Deep learning models can be applied to electrocardiograms (ECGs) to detect left ventricular (LV) dysfunction. We hypothesized that applying a deep learning model may improve the diagnostic accuracy of cardiologists in predicting LV dysfunction from ECGs. We acquired 37,103 paired ECG and echocardiography data records of patients who underwent echocardiography between January 2015 and December 2019. We trained a convolutional neural network to identify the data records of patients with LV dysfunction (ejection fraction < 40%) using a dataset of 23,801 ECGs. When tested on an independent set of 7,196 ECGs, we found the area under the receiver operating characteristic curve was 0.945 (95% confidence interval: 0.936-0.954). When 7 cardiologists interpreted 50 randomly selected ECGs from the test dataset of 7,196 ECGs, their accuracy for predicting LV dysfunction was 78.0% ± 6.0%. By referring to the model's output, the cardiologist accuracy improved to 88.0% ± 3.7%, which indicates that model support significantly improved the cardiologist diagnostic accuracy (P = 0.02). A sensitivity map demonstrated that the model focused on the QRS complex when detecting LV dysfunction on ECGs. We developed a deep learning model that can detect LV dysfunction on ECGs with high accuracy. Furthermore, we demonstrated that support from a deep learning model can help cardiologists to identify LV dysfunction on ECGs.

Automatic Detection of Left Ventricular Dilatation and Hypertrophy from Electrocardiograms Using Deep Learning

“…We implemented a multi-input neural network in Python version 3.7.4 17) using the open-source PyTorch version 1.6.0 deep learning library and the NVIDIA Tesla V100 32 Gb graphics processing unit (NVIDIA Corporation, Santa Clara, CA, USA). For all 3 models, the following parameters were used: loss function, binary cross-entropy with logits loss (BCEwithLogitsLoss); optimizer, Adam; learning rate, 0.00005; and batch size, 128 (based on the grid search of previous studies 13,14) ). Machine learning models: Logistic regression and random forest methods, which have been used in previous studies, 18,19) were used as machine learning models.…”

“…Deep learning can detect significant aortic regurgitation and left ventricular systolic dysfunction from a 12-lead ECG. 13,14) However, there are no established LVD diagnostic criteria or screening methods based on ECG data, and the usefulness of deep learning for the detection of LVD has not been investigated. For the detection of LVH, a detailed comparison of deep learning with conventional criteria and other machine learning methods needs to be made.…”

Artificial intelligence enabled ECG screening for left ventricular systolic dysfunction: a systematic review

“…Attia et al [12] from the Mayo Clinic created an algorithm that was tested in eight studies [12][13][14][15][16][17], one group from Seoul (the republic of Korea) developed and tested their algorithms in two studies [18,19]. Five research groups (Sbrollini et al) developed in-house algorithms and tested them in separate studies [20][21][22][23][24]. A detailed overview of each study population and outcome can be found in "Supplementary results" (Supplementary Appendix 3), where details are provided for AUC, type of outcome, model adjustments and comparison to other test (e.g., BNP/NT-proBNP).…”

“…A retrospective design was used in eleven studies [12][13][14][17][18][19][20][21][22][23][24], one was a case series [15] and three were prospective cohort studies [16,25,26]. Apart from those 15 studies that fulfilled eligibility criteria for development and testing of algorithms, we identified one randomized controlled trial [1] and one cost-effectiveness study [27].…”

“…The diversity of study cohorts reflected a broad clinical spectrum, spanning from unselected digital ECG databases [12,15,16,20,[22][23][24], emergency department patients [13], cardiac intensive care unit patients [17], unspecified hospitalized patients [18,19,21], general populations [14,25] and patients diagnosed with Chagas disease [26].…”