Abstract
Objective:
Patients with dilated cardiomyopathy (DCM) and severely reduced left ventricular ejection fractions (LVEFs) are at very high risks of experiencing adverse cardiac events. A machine learning (ML) method could enable more effective risk stratification for these high-risk patients by incorporating various types of data. The aim of this study was to build an ML model to predict adverse events including all-cause deaths and heart transplantation in DCM patients with severely impaired LV systolic function.
Methods:
One hundred and eighteen patients with DCM and severely reduced LVEFs (<35%) were included. The baseline clinical characteristics, laboratory data, electrocardiographic, and cardiac magnetic resonance (CMR) features were collected. Various feature selection processes and classifiers were performed to select an ML model with the best performance. The predictive performance of tested ML models was evaluated using the area under the curve (AUC) of the receiver operating characteristic curve using 10-fold cross-validation.
Results:
Twelve patients died, and 17 patients underwent heart transplantation during the median follow-up of 508 days. The ML model included systolic blood pressure, left ventricular end-systolic and end-diastolic volume indices, and late gadolinium enhancement (LGE) extents on CMR imaging, and a support vector machine was selected as a classifier. The model showed excellent performance in predicting adverse events in DCM patients with severely reduced LVEF (the AUC and accuracy values were 0.873 and 0.763, respectively).
Conclusions:
This ML technique could effectively predict adverse events in DCM patients with severely reduced LVEF.
Advances in knowledge:
The ML method has superior ability in risk stratification in severe DCM patients.
Introduction
Dilated cardiomyopathy (DCM) is one of the most common non-ischemic heart diseases. Although diagnostic techniques and medical treatments have advanced over the years, the prognoses of patients with DCM remain poor due to sudden cardiac death and irreversible functional deterioration of the heart.1,2 Current guidelines emphasize the role of the left ventricular ejection fraction (LVEF), which is the classical indicator of heart function, in the therapeutic guidance of DCM patients with systolic function impairments.3,4 However, recent cohort study showed that value of LVEF in predicting adverse cardiac events may be imperfect.5 Moreover, the ability to provide accurate prognoses in DCM patients with LVEFs and severely impaired systolic function is additionally hampered. Many models have been developed to predict the outcome of DCM patients with heart failure; however, the performance of these models needs improvements.6,7
Rapid improvements in computer science have led a wide array of machine learning (ML) applications using computational algorithms to learn complex rules and identify patterns in multidimensional datasets.8–10 In recent years, the efficacy of ML has been proven in many areas of cardiology, including diagnostics, precision phenotyping, and prognostication.8,11
Although DCM patients with severely reduced LVEFs represent a high-risk population facing adverse cardiac events,12 few studies have applied an ML method to tackle this issue. Therefore, the aim of our study was to evaluate the feasibility and accuracy of an ML-based risk stratification system to predict all-cause death and heart transplantation in DCM patients with severely reduced LVEFs.
Methods and materials
Study population
Patients referred to our institute for heart failure were retrospectively collected since April 2014. Patients that fulfilled the following criteria were included: (1) increased left ventricular end-diastolic volume indices defined as volumes > 80 ml/m2; and (2) severely impaired systolic function assessed with cardiac magnetic resonance (CMR) imaging and defined by LVEFs < 35%. The exclusion criteria included: (1) patients with ischemic cardiomyopathy characterized by significant stenosis (>50%) of the main coronary artery, a previous myocardial infarction, or coronary revascularization; (2) those with severe valvular heart disease; and (3) those with hypertrophic cardiomyopathy, arrhythmogenic right ventricular cardiomyopathy, or infiltrative disease. Moreover, patients with ischemic patterns found during the CMR examinations were also excluded. For all included patients, baseline clinical characteristics at the time of the CMR study, including demographics, physical statuses, medical histories, laboratory data, electrocardiogram data, and currently applied medical therapies, were retrospectively collected from medical records (Table 1).
Table 1.
Baseline clinical characteristics and cardiac magnetic resonance imaging data
| All patients n = 118 | Patients without events, n = 89 | Patients with events, n = 29 | P-value | |
|---|---|---|---|---|
| Female, n (%) | 39 (33.1%) | 32 (36.0%) | 7 (24.1%) | 0.240 |
| Age (years) | 47.5 (35.8, 57.0) | 47.0 (36.0, 56.0) | 56.0 (37.0, 59.0) | 0.266 |
| BMI (kg/m2) | 24.15 (21.15, 26.40) | 24.55 (21.45, 26.70) | 22.50 (19.10, 25.50) | 0.023 |
| Family history, n (%) | 14 (11.9%) | 12 (13.5%) | 2 (6.9%) | 0.343 |
| Smoking, n (%) | 40 (33.9%) | 31 (34.8%) | 9 (31.0%) | 0.708 |
| Alcohol, n (%) | 22 (18.6%) | 17 (19.1%) | 5 (17.2%) | 0.823 |
| Hypertension, n (%) | 24 (20.3%) | 20 (22.5%) | 4 (13.8%) | 0.313 |
| Diabetes, n (%) | 17 (14.4%) | 11 (12.4%) | 6 (20.7%) | 0.360 |
| SBP (mm Hg) | 115.0 (102.5, 130.5) | 121.5 (109.0, 135.0) | 100.0 (94.0, 114.0) | 0.000 |
| DBP (mm Hg) | 72.0 (63.5, 88.0) | 76.5 (66.8, 89.3) | 65.0 (60.0, 74.0) | 0.004 |
| NYHA functional class | 0.003 | |||
| I | 1 (0.8%) | 1 (1.1%) | 0 (0.0%) | |
| II | 31 (26.3%) | 24 (27.0%) | 7 (24.1%) | |
| III | 49 (41.5%) | 42 (47.2%) | 7 (24.1%) | |
| IV | 30 (25.4%) | 15 (16.9%) | 15 (51.7%) | |
| HR (bpm) | 86.7 (19.6) | 87.8 (19.9) | 83.1 (18.5) | 0.267 |
| QRS (ms) | 100.0 (91.5, 116.5) | 100.0 (90.0, 110.0) | 107.0 (100.0, 135.0) | 0.022 |
| QTc (ms) | 438.0 (415.5, 465.3) | 436.0 (413.8, 463.3) | 447.5 (423.0, 481.3) | 0.148 |
| AVB, n (%) | 3 (2.6%) | 2 (2.3%) | 1 (3.6%) | >0.99 |
| LBBB, n (%) | 15 (13.2%) | 11 (12.8%) | 4 (3.7%) | >0.99 |
| RBBB, n (%) | 4 (3.5%) | 3 (3.5%) | 1 (3.6%) | >0.99 |
| AST (U/L) | 26.0 (20.0, 37.8) | 26.0 (20.0, 38.0) | 25.0 (20.0, 40.5) | 0.806 |
| ALT (U/L) | 29.5 (17.3, 55.8) | 32.0 (19.0, 60.0) | 23.0 (16.0, 42.0) | 0.119 |
| BUN (mmol/L) | 6.6 (5.5, 8.0) | 6.4 (5.4, 7.9) | 7.4 (5.8, 10.5) | 0.064 |
| Cr (mmol/L) | 80.8 (66.1, 100.9) | 83.3 (65.9, 100.1) | 74.7 (66.0, 116.0) | 0.923 |
| EGFR (ml/min) | 85.4 (72.6, 103.1) | 83.8 (73.2, 100.1) | 89.5 (54.3, 110.2) | 0.866 |
| CMR parameters | ||||
| LVEF (%) | 14.1 (10.8, 20.0) | 15.5 (11.1, 22.1) | 11.9 (8.5, 17.2) | 0.016 |
| LVEDVI (ml/m2) | 173.4 (141.7, 213.8) | 165.4 (135.8, 201.7) | 225.6 (181.8, 295.0) | 0.000 |
| LVESVI (ml/m2) | 143.7 (118.8, 190.0) | 135.2 (112.2, 175.6) | 195.4 (149.4, 256.4) | 0.000 |
| LV MASSI (mg/m2) | 87.8 (73.9, 104.7) | 84.9 (70.9, 96.2) | 92.1 (82.9, 117.7) | 0.053 |
| RVEF (%) | 19.8 (9.5, 33.5) | 21.9 (12.0, 35.4) | 11.7 (5.4, 22.7) | 0.014 |
| RVEDVI (ml/m2) | 85.3 (64.7, 113.0) | 80.3 (63.3, 104.1) | 114.4 (71.9, 133.2) | 0.005 |
| RVESVI (ml/m2) | 65.5 (45.5, 93.8) | 63.3 (43.7, 84.9) | 96.0 (55.2, 123.3) | 0.002 |
| LGE extent (%) | 2.2 (0.7, 4.9) | 1.6 (0.5, 3.3) | 3.7 (1.3, 11.1) | 0.001 |
| Cardiac medication | ||||
| ACEI/ARB, n (%) | 36 (30.5%) | 25 (28.1%) | 11 (37.9%) | 0.318 |
| ARNi, n (%) | 81 (68.6%) | 62 (69.7%) | 19 (65.5%) | 0.676 |
| Beta-blocker, n (%) | 90 (76.3%) | 72 (80.9%) | 18 (62.1%) | 0.038 |
| Diuretics, n (%) | 45 (38.1%) | 30 (33.7%) | 15 (51.7%) | 0.083 |
| Digoxin, n (%) | 38 (32.2%) | 28 (31.5%) | 10 (34.5%) | 0.762 |
| Aldosterone antagonist, n (%) | 76 (64.4%) | 55 (61.8%) | 21 (72.4%) | 0.300 |
ACEI, Angiotensin-Converting Enzyme Inhibitors; ALT, Alanine transaminase; ARB, Angiotensin Receptor Blockers; ARNi, Angiotensin Receptor-Neprilysin Inhibitors; AST, Aspartate aminotransferase; AVB, Atrioventricular Block; BMI, Body Mass Index; BUN, Blood Urea Nitrogen; CMR, Cardic Magnetic Resonance; Cr, Creatinine; DBP, Diastolic Blood Pressure; EGFR, Effective Glomerular Filtration Rate; HR, Heart Rate; LBBB, Left Bundle Branch Block; LGE, Late Gadolinium Enhancement; LVEDVI, Left Ventricular End-Diastolic Volume Index; LVEF, Left Ventricular Ejection Fraction; LVESVI, Left Ventricular End-Systolic Volume Index; MASSI, MASS Index; NYHA, New York Heart Association; RBBB, Right Bundle Branch Block; RVEDVI, Right Ventricular End-Diastolic Volume Index; RVEF, Right Ventricular Ejection Fraction; RVESVI, Right Ventricular End-Systolic Volume Index; SBP, Systolic Blood Pressure.
MR imaging and image post-processing
CMR examinations were performed on 1.5 T or 3 T scanners (MAGNETOM Aera or MAGNETOM, Skyra, Siemens, Erlangen, Germany) with surface phased array coils and four-lead vector electrocardiogram-gating. The standard study protocol consisted of localizer sequences, cine sequences to identify ventricular structure and function, perfusion imaging, and late-enhanced scanning. Cine images were acquired on the LV long-axis and continuous LV short-axis views. Late-enhanced sequences were scanned for 10–15 min after cumulative i.v. administration of 0.2 mmol/kg gadopentetate dimeglumine (Magnevist, Bayer, Germany) or 0.1 mmol/kg gadobutrol (since 2018) (Gadovist, Bayer, Germany) using the same views as used for the cine-CMR imaging. Typical imaging parameters for cine and late enhanced imaging were as follows: field of view 415 × 340 mm2, matrix 256 × 256, slice thickness 8 mm, and slice gap 1.6 mm. All CMR scanning were performed by two technicians with more than 10 years’ experiences. Image post-processing was performed using dedicated commercially available software (CMR42, Circle Cardiovascular Imaging Inc., Calgary, Canada). Volumes and ejection fractions of bilateral ventricles were acquired using standardized recommendations.13 The presence and extent (%) of LGE in the LV myocardium were judged and quantitatively calculated.
Follow-up
All patients were followed regularly via clinical visits or telephone calls. The endpoint events, including all-cause deaths and heart transplantations, were examined. The duration of follow-up was defined as the time from the start of the CMR study to the endpoint events or last contact. Our study was approved by the local ethics committee, and written informed consent was waived due to the retrospective nature of the study.
Data pre-processing and machine learning model development
Baseline variables, including the electrocardiographic and cardiac magnetic resonance imaging parameters with features missing <40% of cases, were included. Missing values were completed with the mean imputation method. Specifically, for continuous variables, missing values were replaced with the mean value of the available cases. For the categorical variables, missing values were replaced with the mode of the available cases. Fifty-nine baseline variables were finally used as an input for the ML algorithm (Supplementary Material 1 ). FAE, a Python toolbox, was used to develop an ML classification model.14 Figure 1 shows the model development pipeline. ML model generation included the following steps: feature normalization, class balance, dimension reduction, feature selection and classifier training, and different data processing techniques could be used at each step. Before training the classifier, three preprocessing steps including feature normalize, class balance and feature selection, were performed to preprocess the features. After feature selection, we trained a classifier based on data in the training set. When training completed, the performance of the classifier model was tested on the validation set. . The whole dataset was split into 10-fold validations with roughly the same ratio of positive and negative samples. In each validation, a one-fold validation was chosen as the validation set, and other validations were used as the training set. The performances of each cross-validation in the ML model were merged into a final cross-validation result. By comparing all the randomly combined ML models, the model with the best performance was chosen. As LVEF and LGE extent have been proven as effective markers for risk stratification in patients with heart failure,7 their value was also evaluated as independent marker for patient classification.
Figure 1.
Workflow showing the building and validation of the machine-learning model with the FAE software.
Statistical analysis
Continuous variables are presented as medians and interquartile ranges (IQRs), and differences between the two groups were compared using the Mann-Whitney U-test. Categorical variables are expressed as counts and percentages, and inter-group differences were calculated using the χ2 test or Fisher’s exact test. To test the classification effectiveness of ML, the areas under the receiver operating characteristic (ROC) curve (AUCs) of the 10-fold cross-validations were calculated. All statistical analyses were performed using SPSS software (version 22.0, IBM, Armonk, USA).
Results
A total of 118 patients were finally included in this retrospective study. Thirty-nine (33.1%) patients were females, and the median age was 47.5 years. Of these patients, 12 died, and 17 underwent heart transplantations during a median follow-up of 508 days (IQR, 307 to 700 days). In this study cohort, left ventricular systolic function was severely impaired due to very low LVEFs (median LVEF, 14.12%; range, 10.82 to 20.02%). The baseline characteristics (demographic, ECG, laboratory, medication, and CMR data) are shown in Table 1.
Finally, an ML model was selected that adopted a Z-score for feature normalizations, recursive feature eliminations to select features, and a support vector machine to classify the data. This model showed excellent predictive power for adverse events in patients with DCM and severely reduced LVEF. The AUC and accuracy values for this model were 0.873 and 0.763, respectively (Figure 2). The model emphasized four baseline variables, systolic blood pressure (SBP), the LGE extent, the left ventricular end-diastolic volume index (LVEDVI), and the left ventricular end-systolic volume index (LVESVI). Comparatively, the ability of the LGE extent and LVEF to predict adverse events was significantly lower than the ML model values (AUCs = 0.698 and 0.350, respectively, p < 0.01, Figure 2).
Figure 2.
Receiver operating characteristic curves for the predictive performance of the selected machine-learning model, showing the left ventricular ejection fraction (LVEF), and late gadolinium enhancement (LGE) extent.
Discussion
In this preliminary study, we demonstrated that an ML method could integrate clinical and imaging variables at baseline and effectively predict adverse events in patients with DCM and severely reduced LVEFs. The combined parameters of SBP, LGE extent, and left ventricular volume indices showed great prognostic value. Results from our study serve as a proof-of-concept study that artificial intelligence could provide a meaningful risk stratification method for patients with severe DCM.
The ejection fraction is a comprehensive index reflecting the function of ventricular motion. It has been established as a parameter for disease grading and treatment guidance in patients with cardiomyopathy and heart failure.3,4 Although it has been used to predict adverse events in patients with chronic heart failure, its value remains to be optimized.15,16 Specifically, the prognostic value of LVEF was extremely limited in our study as all included patients had severely impaired LV systolic function. Conversely, LGEs in CMR has been shown to be a sensitive marker of myocardial fibrosis.17,18 In addition, its presence and volume fractions were shown to be independent risk factors of adverse cardiac events in patients with ischemic and non-ischemic cardiomyopathies.19–22 Due to the high prevalence of LGEs in our study cohort, the presence of LGEs showed no difference between patients with adverse events and those who survived. For the same reason, although the LGE extents were significantly greater in the event-group patients, the performance of risk stratification was not so satisfied. Thus, a more effective model for risk stratification in patients with severe DCM is needed to guide treatments and predict prognoses.
ML techniques, such as the one used in our study, offer the advantage of exploiting expansive amounts of data to design comprehensive models to provide powerful risk stratification systems and outcome predictions.23 ML methods can integrate many features regardless of the type, including demographic features, laboratory data, and imaging parameters.24 These methods have shown immense potential in medical decisions-making, especially in cardiovascular disease management. In a retrospective study, Chen et al used a Bayes classifier for risk prediction in DCM patients with SBP, showing that the duration of the QRS interval was the most important feature.25 In another study, Cikes et al successfully identified responders to a cardiac resynchronization therapy by phenogrouping heart failure patients in an ML system.26 Similarly, in this study, clinical data, laboratory data, and electrographic and CMR parameters were included for the ML analyses. After multiple feature selection processes, we examined the performance of different ML models and selected the model with the best classification power. In addition to the two left ventricular volume-related indices (LVESVI and LVEDVI), SBP and LGEs were also included in the ML model. Specifically, the extent of LGEs, which was found to be a robust risk factor in previous studies played an important role in the ML model.
Several limitations existed in our study. The sample was relatively small as this study was conducted at a single institution and could have decreased the power of the ML model. The event rate could have been increased by the easily accessible transplantation treatments at the tertiary center; however, the number of patients with adverse events was small. Finally, how the ML model performs in larger populations with severe DCM remains to be studied.
Conclusions
In the retrospective study with a relatively small patient population, ML proved to be an interpretable and clinically meaningful technique for risk stratification in patients with DCM and severely reduced LVEFs. This novel and feasible method could perform better than conventional risk stratification techniques and should be verified in multi center studies with a larger number of patients.
Footnotes
Funding: This study was supported by the National Natural Science Foundation of China (Grant No. 81701653).
Patient consent: Written informed consent was waived due to the retrospective nature of the study.
Ethics approval: The study was approved by ethical committee of Tongji Medical College of Huazhong University of Science and Technology (No. 2020-S301).
The authors Shenglei Shu and Ziming Hong contributed equally to the work.
Contributors: Chuansheng Zheng and JW contributed to the conception of the study; Shenglei Shu and Ziming Hong analyzed the clinical data and performed machine learning models building and selection and they were major contributors in writing the manuscript; Qinmu Peng helped perform the analysis with constructive discussions. Xiaoyue Zhou and Tianjing Zhang contributed greatly in writing and editing the manuscript.
Contributor Information
Shenglei Shu, Email: shuslwuhan@gmail.com.
Ziming Hong, Email: hongzm@hust.edu.cn.
Qinmu Peng, Email: pengqinmu@hust.edu.cn.
Xiaoyue Zhou, Email: xiaoyue.zhou@siemens-healthineers.com.
Jing Wang, Email: jjwinflower@126.com.
Chuansheng Zheng, Email: cszheng@hust.edu.cn.
REFERENCES
- 1.Schultheiss H-P, Fairweather D, Caforio ALP, Escher F, Hershberger RE, Lipshultz SE, et al. Dilated cardiomyopathy. Nat Rev Dis Primers 2019; 5: 32. doi: 10.1038/s41572-019-0084-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Jefferies JL, Towbin JA. Dilated cardiomyopathy. Lancet 2010; 375: 752–62. doi: 10.1016/S0140-6736(09)62023-7 [DOI] [PubMed] [Google Scholar]
- 3.Ponikowski P, Voors AA, Anker SD, Bueno H, Cleland JGF, Coats AJS, et al. 2016 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure: The Task Force for the diagnosis and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC)Developed with the special contribution of the Heart Failure Association (HFA) of the ESC.. . Eur Heart J 2016; 37: 2129–2002016 Dec 30;:]. doi: 10.1093/eurheartj/ehw128 [DOI] [PubMed] [Google Scholar]
- 4.Yancy CW, Jessup M, Bozkurt B, Butler J, Casey DE, Colvin MM, et al. 2017 ACC/AHA/HFSA focused update of the 2013 ACCF/AHA guideline for the management of heart failure: a report of the American College of Cardiology/American heart association Task force on clinical practice guidelines and the heart failure Society of America. Circulation 2017; 136: e137–61. doi: 10.1161/CIR.0000000000000509 [DOI] [PubMed] [Google Scholar]
- 5.Køber L, Thune JJ, Nielsen JC, Haarbo J, Videbæk L, Korup E, et al. Defibrillator implantation in patients with nonischemic systolic heart failure. N Engl J Med 2016; 375: 1221–30. doi: 10.1056/NEJMoa1608029 [DOI] [PubMed] [Google Scholar]
- 6.Li X, Fan X, Li S, Sun W, Shivkumar K, Zhao S, et al. A novel risk stratification score for sudden cardiac death prediction in middle-aged, nonischemic dilated cardiomyopathy patients: the estimated score. Can J Cardiol 2020; 36: 1121–9. doi: 10.1016/j.cjca.2019.11.009 [DOI] [PubMed] [Google Scholar]
- 7.Marrow BA, Cook SA, Prasad SK, McCann GP. Emerging Techniques for Risk Stratification in Nonischemic Dilated Cardiomyopathy: JACC Review Topic of the Week. J Am Coll Cardiol 2020; 75: 1196–207. doi: 10.1016/j.jacc.2019.12.058 [DOI] [PubMed] [Google Scholar]
- 8.Litjens G, Ciompi F, Wolterink JM, de Vos BD, Leiner T, Teuwen J, et al. State-Of-The-Art deep learning in cardiovascular image analysis. JACC Cardiovasc Imaging 2019; 12(8 Pt 1): 1549–65. doi: 10.1016/j.jcmg.2019.06.009 [DOI] [PubMed] [Google Scholar]
- 9.Dey D, Slomka PJ, Leeson P, Comaniciu D, Shrestha S, Sengupta PP, et al. Artificial intelligence in cardiovascular imaging: JACC state-of-the-art review. J Am Coll Cardiol 2019; 73: 1317–35. doi: 10.1016/j.jacc.2018.12.054 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Bi WL, Hosny A, Schabath MB, Giger ML, Birkbak NJ, Mehrtash A, et al. Artificial intelligence in cancer imaging: clinical challenges and applications. CA Cancer J Clin 2019; 69: 127–57. doi: 10.3322/caac.21552 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Leiner T, Rueckert D, Suinesiaputra A, Baeßler B, Nezafat R, Išgum I, et al. Machine learning in cardiovascular magnetic resonance: basic concepts and applications. J Cardiovasc Magn Reson. 2019; 21: 61.2019;21(1):61. Published. doi: 10.1186/s12968-019-0575-y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Merlo M, Gentile P, Naso P, Sinagra G. The natural history of dilated cardiomyopathy: how has it changed? J Cardiovasc Med 2017; 18 Suppl 1(Suppl 1): e161–5. doi: 10.2459/JCM.0000000000000459 [DOI] [PubMed] [Google Scholar]
- 13.Schulz-Menger J, Bluemke DA, Bremerich J, Flamm SD, Fogel MA, Friedrich MG, et al. Standardized image interpretation and post-processing in cardiovascular magnetic resonance - 2020 update : Society for Cardiovascular Magnetic Resonance (SCMR): Board of Trustees Task Force on Standardized Post-Processing. J Cardiovasc Magn Reson 2020; 22: 19.(1):19. Published. doi: 10.1186/s12968-020-00610-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Song Y, Zhang J, Zhang Y-D, Hou Y, Yan X, Wang Y, et al. Feature explorer (Fae): a tool for developing and comparing radiomics models. PLoS One. 2020; 15():: e0237587Published8. doi: 10.1371/journal.pone.0237587 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Halliday BP, Cleland JGF, Goldberger JJ, Prasad SK. Personalizing risk stratification for sudden death in dilated cardiomyopathy: the past, present, and future. Circulation 2017; 136: 215–31. doi: 10.1161/CIRCULATIONAHA.116.027134 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Brown PF, Miller C, Di Marco A, Schmitt M. Towards cardiac MRI based risk stratification in idiopathic dilated cardiomyopathy. Heart 2019; 105: 270–5. doi: 10.1136/heartjnl-2018-313767 [DOI] [PubMed] [Google Scholar]
- 17.Iles LM, Ellims AH, Llewellyn H, Hare JL, Kaye DM, McLean CA, et al. Histological validation of cardiac magnetic resonance analysis of regional and diffuse interstitial myocardial fibrosis. Eur Heart J Cardiovasc Imaging 2015; 16: 14–22. doi: 10.1093/ehjci/jeu182 [DOI] [PubMed] [Google Scholar]
- 18.Galati G, Leone O, Pasquale F, Olivotto I, Biagini E, Grigioni F, et al. Histological and Histometric characterization of myocardial fibrosis in end-stage hypertrophic cardiomyopathy: a clinical-pathological study of 30 explanted hearts. Circ Heart Fail 2016; 9: e003090. doi: 10.1161/CIRCHEARTFAILURE.116.003090 [DOI] [PubMed] [Google Scholar]
- 19.Becker MAJ, Cornel JH, van de Ven PM, van Rossum AC, Allaart CP, Germans T. The Prognostic Value of Late Gadolinium-Enhanced Cardiac Magnetic Resonance Imaging in Nonischemic Dilated Cardiomyopathy: A Review and Meta-Analysis. JACC Cardiovasc Imaging 2018; 11: 1274–84. doi: 10.1016/j.jcmg.2018.03.006 [DOI] [PubMed] [Google Scholar]
- 20.Pi S-H, Kim SM, Choi J-O, Kim EK, Chang S-A, Choe YH, et al. Prognostic value of myocardial strain and late gadolinium enhancement on cardiovascular magnetic resonance imaging in patients with idiopathic dilated cardiomyopathy with moderate to severely reduced ejection fraction. J Cardiovasc Magn Reson; 2018; 20: 36.20(1):36. Published. doi: 10.1186/s12968-018-0466-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Halliday BP, Baksi AJ, Gulati A, Ali A, Newsome S, Izgi C, et al. Outcome in Dilated Cardiomyopathy Related to the Extent, Location, and Pattern of Late Gadolinium Enhancement. JACC Cardiovasc Imaging 2019; 12(8 Pt 2): 1645–55. doi: 10.1016/j.jcmg.2018.07.015 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Di Marco A, Anguera I, Schmitt M, Klem I, Neilan TG, White JA. Late Gadolinium Enhancement and the Risk for Ventricular Arrhythmias or Sudden Death in Dilated Cardiomyopathy: Systematic Review and Meta-Analysis [published correction appears in JACC Heart Fail.;. JACC Heart Fail 2017;.):: 28–385(4):316]2017;5(1. [DOI] [PubMed] [Google Scholar]
- 23.Johnson KW, Torres Soto J, Glicksberg BS, Shameer K, Miotto R, Ali M, et al. Artificial intelligence in cardiology. J Am Coll Cardiol 2018; 71: 2668–79. doi: 10.1016/j.jacc.2018.03.521 [DOI] [PubMed] [Google Scholar]
- 24.Fralick M, Colak E, Mamdani M. Machine learning in medicine. N Engl J Med 2019; 380: 2588–9. doi: 10.1056/NEJMc1906060 [DOI] [PubMed] [Google Scholar]
- 25.Chen R, Lu A, Wang J, Ma X, Zhao L, Wu W, et al. Using machine learning to predict one-year cardiovascular events in patients with severe dilated cardiomyopathy. Eur J Radiol 2019; 117: 178–83. doi: 10.1016/j.ejrad.2019.06.004 [DOI] [PubMed] [Google Scholar]
- 26.Cikes M, Sanchez-Martinez S, Claggett B, Duchateau N, Piella G, Butakoff C, et al. Machine learning-based phenogrouping in heart failure to identify responders to cardiac resynchronization therapy. Eur J Heart Fail 2019; 21: 74–85. doi: 10.1002/ejhf.1333 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.