{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"medRxiv"}],"indexed":{"date-parts":[[2026,10,13]],"date-time":"2026-10-13T11:33:35Z","timestamp":1791891215993,"version":"4.3.4"},"posted":{"date-parts":[[2024,10,15]]},"group-title":"Cardiovascular Medicine","reference-count":35,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2024,10,15]],"date-time":"2024-10-15T00:00:00Z","timestamp":1728950400000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by-nc-nd\/4.0\/"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"accepted":{"date-parts":[[2025,9,9]]},"abstract":"<jats:title>ABSTRACT<\/jats:title>\n                <jats:sec>\n                  <jats:title>Background<\/jats:title>\n                  <jats:p>Traditional ECG criteria for left ventricular hypertrophy (LVH) have modest diagnostic yield.<\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Objective<\/jats:title>\n                  <jats:p>Develop and validate machine learning models for LVH diagnosis from ECG.<\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Methods<\/jats:title>\n                  <jats:p>\n                    ECG summary features (rate, intervals, axis), R-wave, S-wave and overall-QRS amplitudes, and QRS voltage-time integrals (VTI\n                    <jats:sub>QRS<\/jats:sub>\n                    ) were extracted from 12-lead, vectorcardiographic X-Y-Z-lead, and 3D (L2 norm) representative-beat ECGs. Latent features (30 per ECG) were extracted using a variational autoencoder (trained on unselected &gt;1 million ECGs) from X-Y-Z-lead representative-beat ECG signals. Logistic regression, random forest, light gradient boosted machine (LGBM), residual network (ResNet) and multilayer perceptron network (MLP) models using ECG features and sex, and a convolutional neural network (CNN) using ECG signals alone, were trained to predict LVH (left ventricular mass indexed in women &gt;95 g\/m\n                    <jats:sup>2<\/jats:sup>\n                    , men &gt;115 g\/m\n                    <jats:sup>2<\/jats:sup>\n                    ) on 482,734 adult ECG-echocardiogram (within 45 days) pairs. ROC-AUCs for LVH classification are reported from a separate hold-out test set.\n                  <\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Results<\/jats:title>\n                  <jats:p>\n                    In the test set (n=54,984), AUC for LVH classification was higher for ML models using ECG features (LGBM 0.794, MLP 0.793, ResNet 0.795) compared with the best individual ECG variable (VTI\n                    <jats:sub>QRS-Z<\/jats:sub>\n                    0.707), the best traditional criterion (Cornell voltage-duration product 0.716), and the CNN using ECG signals (0.788). Among patients without LVH who had a follow-up echocardiogram &gt;1 (closest to 5) year later, LGBM false positives, compared to true negatives, had a 3.07 (95% CI 2.44, 3.86)-fold higher odds of developing future LVH (p&lt;0.0001).\n                  <\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Conclusions<\/jats:title>\n                  <jats:p>ML models are superior to traditional ECG criteria to classify LVH. Models trained on extracted ECG features, including latent variational autoencoder representations, can outperform CNN models directly trained on ECG signals.<\/jats:p>\n                <\/jats:sec>","DOI":"10.1101\/2024.10.14.24315460","type":"posted-content","created":{"date-parts":[[2024,10,15]],"date-time":"2024-10-15T13:30:22Z","timestamp":1728999022000},"source":"Crossref","is-referenced-by-count":3,"title":["Machine learning to classify left ventricular hypertrophy using ECG feature extraction by variational autoencoder"],"prefix":"10.64898","author":[{"ORCID":"https:\/\/orcid.org\/0000-0001-5317-6024","authenticated-orcid":false,"given":"Amulya","family":"Gupta","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Christopher J.","family":"Harvey","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ashley","family":"DeBauge","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sumaiya","family":"Shomaji","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zijun","family":"Yao","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yongkuk","family":"Lee","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-9947-3849","authenticated-orcid":false,"given":"Amit","family":"Noheria","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"54368","reference":[{"key":"2025091204400466000_2024.10.14.24315460v2.1","doi-asserted-by":"publisher","DOI":"10.1067\/mhj.2001.113218"},{"key":"2025091204400466000_2024.10.14.24315460v2.2","doi-asserted-by":"crossref","first-page":"203","DOI":"10.1007\/s40119-022-00260-y","article-title":"Left Ventricular Hypertrophy: Etiology-Based Therapeutic Options","volume":"11","year":"2022","journal-title":"Cardiol Ther"},{"key":"2025091204400466000_2024.10.14.24315460v2.3","doi-asserted-by":"crossref","unstructured":"Cuspidi C , Meani S , Valerio C , Fusi V , Sala C , Zanchetti A . Left ventricular hypertrophy and cardiovascular risk stratification: impact and cost-effectiveness of echocardiography in recently diagnosed essential hypertensives. Journal of Hypertension 2006;24.","DOI":"10.1097\/01.hjh.0000239305.01496.ca"},{"key":"2025091204400466000_2024.10.14.24315460v2.4","doi-asserted-by":"publisher","DOI":"10.1161\/HYP.0000000000000065"},{"key":"2025091204400466000_2024.10.14.24315460v2.5","first-page":"e005289","article-title":"Automated and Interpretable Patient ECG Profiles for Disease Detection","volume":"12","year":"2019","journal-title":"Tracking, and Discovery. Circulation: Cardiovascular Quality and Outcomes"},{"key":"2025091204400466000_2024.10.14.24315460v2.6","doi-asserted-by":"crossref","unstructured":"Ricciardi D , Vetta G , Nenna A et al. Current diagnostic ECG criteria for left ventricular hypertrophy: is it time to change paradigm in the analysis of data? Journal of Cardiovascular Medicine 2020;21.","DOI":"10.2459\/JCM.0000000000000907"},{"key":"2025091204400466000_2024.10.14.24315460v2.7","doi-asserted-by":"publisher","DOI":"10.1038\/ajh.2010.42"},{"key":"2025091204400466000_2024.10.14.24315460v2.8","doi-asserted-by":"publisher","DOI":"10.1161\/CIRCULATIONAHA.108.191097"},{"key":"2025091204400466000_2024.10.14.24315460v2.9","doi-asserted-by":"publisher","DOI":"10.1007\/s11886-024-02062-1"},{"key":"2025091204400466000_2024.10.14.24315460v2.10","doi-asserted-by":"publisher","DOI":"10.1097\/hco.0000000000000812"},{"key":"2025091204400466000_2024.10.14.24315460v2.11","doi-asserted-by":"crossref","first-page":"15882","DOI":"10.1038\/s41598-024-66247-y","article-title":"Diagnostic accuracy of artificial intelligence in detecting left ventricular hypertrophy by electrocardiograph: a systematic review and meta-analysis","volume":"14","year":"2024","journal-title":"Scientific Reports"},{"key":"2025091204400466000_2024.10.14.24315460v2.12","first-page":"022022","article-title":"An Overview of Overfitting and its Solutions","volume":"1168","year":"2019","journal-title":"Journal of Physics: Conference Series"},{"key":"2025091204400466000_2024.10.14.24315460v2.13","doi-asserted-by":"publisher","DOI":"10.1016\/j.jacc.2007.01.024"},{"key":"2025091204400466000_2024.10.14.24315460v2.14","unstructured":"Harvey CJ , Shomaji S , Yao Z , Noheria A . Comparison of Autoencoder Encodings for ECG Representation in Downstream Prediction Tasks. arXiv preprint 2024:2410.02937."},{"key":"2025091204400466000_2024.10.14.24315460v2.15","doi-asserted-by":"crossref","first-page":"172","DOI":"10.1016\/S0735-1097(24)02162-4","article-title":"DEEP LEARNING ENCODED ECG \u2013 AVOIDING OVERFITTING IN ECG MACHINE LEARNING","volume":"83","year":"2024","journal-title":"Journal of the American College of Cardiology"},{"key":"2025091204400466000_2024.10.14.24315460v2.16","doi-asserted-by":"crossref","first-page":"26","DOI":"10.1016\/S0735-1097(24)02016-3","article-title":"REDUCING DATA DIMENSIONALITY OF ECG SIGNAL USING DEEP LEARNING","volume":"83","year":"2024","journal-title":"Journal of the American College of Cardiology"},{"key":"2025091204400466000_2024.10.14.24315460v2.17","doi-asserted-by":"publisher","DOI":"10.1136\/jamia.2009.000893"},{"key":"2025091204400466000_2024.10.14.24315460v2.18","unstructured":"Waitman LR , Warren JJ , Manos EL , Connolly DW . Expressing observations from electronic medical record flowsheets in an i2b2 based clinical data repository to support research and quality improvement. AMIA Annu Symp Proc 2011;2011:1454\u201363."},{"key":"2025091204400466000_2024.10.14.24315460v2.19","doi-asserted-by":"publisher","DOI":"10.1016\/j.echo.2014.10.003"},{"key":"2025091204400466000_2024.10.14.24315460v2.20","doi-asserted-by":"crossref","first-page":"73","DOI":"10.1016\/j.jelectrocard.2023.11.004","article-title":"Electrocardiographic Z-axis QRS-T voltage-time-integral in patients with typical right bundle branch block - Correlation with echocardiographic right ventricular size and function","volume":"82","year":"2024","journal-title":"J Electrocardiol"},{"key":"2025091204400466000_2024.10.14.24315460v2.21","doi-asserted-by":"publisher","DOI":"10.1016\/j.jelectrocard.2023.03.004"},{"key":"2025091204400466000_2024.10.14.24315460v2.22","doi-asserted-by":"publisher","DOI":"10.1093\/oxfordjournals.eurheartj.a059647"},{"key":"2025091204400466000_2024.10.14.24315460v2.23","doi-asserted-by":"publisher","DOI":"10.1016\/j.jelectrocard.2024.153787"},{"key":"2025091204400466000_2024.10.14.24315460v2.24","unstructured":"Harvey CJ , Shomaji S , Yao Z , Noheria A. Comparison of Autoencoder Encodings for ECG Representation in Downstream Prediction Tasks: arXiv."},{"key":"2025091204400466000_2024.10.14.24315460v2.25","doi-asserted-by":"publisher","DOI":"10.1111\/j.0954-6820.1920.tb18266.x"},{"key":"2025091204400466000_2024.10.14.24315460v2.26","doi-asserted-by":"publisher","DOI":"10.1161\/CIRCULATIONAHA.108.191095"},{"key":"2025091204400466000_2024.10.14.24315460v2.27","doi-asserted-by":"crossref","first-page":"251","DOI":"10.1161\/01.HYP.27.2.251","article-title":"Time-Voltage Area of the QRS for the Identification of Left Ventricular Hypertrophy","volume":"27","year":"1996","journal-title":"Hypertension"},{"key":"2025091204400466000_2024.10.14.24315460v2.28","doi-asserted-by":"publisher","DOI":"10.1016\/j.jacc.2017.01.037"},{"key":"2025091204400466000_2024.10.14.24315460v2.29","doi-asserted-by":"publisher","DOI":"10.1016\/0167-5273(90)90063-B"},{"key":"2025091204400466000_2024.10.14.24315460v2.30","doi-asserted-by":"publisher","DOI":"10.1016\/S0022-0736(87)80020-1"},{"key":"2025091204400466000_2024.10.14.24315460v2.31","doi-asserted-by":"publisher","DOI":"10.1016\/0735-1097(92)90376-X"},{"key":"2025091204400466000_2024.10.14.24315460v2.32","first-page":"AIoa2300297","article-title":"Four-Channel ECG as a Single Source for Early Diagnosis of Cardiac Hypertrophy and Dilation \u2014 A Deep Learning Approach","volume":"1","year":"2024","journal-title":"NEJM AI"},{"key":"2025091204400466000_2024.10.14.24315460v2.33","first-page":"e008360","article-title":"Artificial Intelligence\u2013Enabled Model for Early Detection of Left Ventricular Hypertrophy and Mortality Prediction in Young to Middle-Aged Adults","volume":"15","year":"2022","journal-title":"Circulation: Cardiovascular Quality and Outcomes"},{"key":"2025091204400466000_2024.10.14.24315460v2.34","doi-asserted-by":"crossref","first-page":"412","DOI":"10.1093\/europace\/euz324","article-title":"Comparing the performance of artificial intelligence and conventional diagnosis criteria for detecting left ventricular hypertrophy using electrocardiography","volume":"22","year":"2020","journal-title":"EP Europace"},{"key":"2025091204400466000_2024.10.14.24315460v2.35","doi-asserted-by":"crossref","first-page":"48","DOI":"10.1016\/j.cvdhj.2023.03.001","article-title":"Artificial intelligence\u2013enabled classification of hypertrophic heart diseases using electrocardiograms","volume":"4","year":"2023","journal-title":"Cardiovascular Digital Health Journal"}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.1101\/2024.10.14.24315460","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,1,15]],"date-time":"2026-01-15T20:14:49Z","timestamp":1768508089000},"score":1,"resource":{"primary":{"URL":"http:\/\/medrxiv.org\/lookup\/doi\/10.1101\/2024.10.14.24315460"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2024,10,15]]},"references-count":35,"URL":"https:\/\/doi.org\/10.1101\/2024.10.14.24315460","relation":{},"subject":[],"published":{"date-parts":[[2024,10,15]]},"subtype":"preprint"}}