Prediction of cardiac death after adenosine myocardial perfusion SPECT based on machine learning
- 15 Downloads
We developed machine-learning (ML) models to estimate a patient’s risk of cardiac death based on adenosine myocardial perfusion SPECT (MPS) and associated clinical data, and compared their performance to baseline logistic regression (LR). We demonstrated an approach to visually convey the reasoning behind a patient’s risk to provide insight to clinicians beyond that of a “black box.”
We trained multiple models using 122 potential clinical predictors (features) for 8321 patients, including 551 cases of subsequent cardiac death. Accuracy was measured by area under the ROC curve (AUC), computed within a cross-validation framework. We developed a method to display the model’s rationale to facilitate clinical interpretation.
The baseline LR (AUC = 0.76; 14 features) was outperformed by all other methods. A least absolute shrinkage and selection operator (LASSO) model (AUC = 0.77; p = .045; 6 features) required the fewest features. A support vector machine (SVM) model (AUC = 0.83; p < .0001; 49 features) provided the highest accuracy.
LASSO outperformed LR in both accuracy and simplicity (number of features), with SVM yielding best AUC for prediction of cardiac death in patients undergoing MPS. Combined with presenting the reasoning behind the risk scores, our results suggest that ML can be more effective than LR for this application.
KeywordsCardiac death risk model machine learning feature selection data visualization
Single-photon emission computed tomography
Myocardial perfusion SPECT
Summed stress score
Summed rest score
Receiver operating characteristic
Area under the curve
Support vector machine
Least absolute shrinkage and selection operator
Research reported in this article was supported by the National Heart, Lung, and Blood Institute of the National Institutes of Health under Award Numbers R01HL122484 and R01HL089765. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
DH.A., Y.Y., and M.N.W. from the Illinois Institute of Technology have nothing to disclose. G.G., D.S.B., and P.S. from the Cedars-Sinai Medical Center have nothing to disclose.
- 1.Wernick MN, Aarsvold JN. Emission tomography: The engineering and physics of PET and SPECT. San Diego: Academic Press; 2004. p. 596.Google Scholar
- 2.Hachamovitch R, Berman DS, Kiat H, Cohen I, Lewin H, Amanullah A, et al. Incremental prognostic value of adenosine stress myocardial perfusion single-photon emission computed tomography and impact on subsequent management in patients with or suspected of having myocardial ischemia. Am J Cardiol 1997;80:426-33.CrossRefPubMedGoogle Scholar
- 5.Sharir T, Ben-Haim S, Merzon K, Prochorov V, Dickman D, Ben-Haim S, et al. High-speed myocardial perfusion imaging: initial clinical comparison with conventional dual detector anger camera imaging. JACC: Cardiovasc Imaging 2008;1:156-63.Google Scholar
- 6.Heller GV, Herman SD, Travin MI, Baron JI, Santos-Ocampo C, Mcclellan JR. Independent prognostic value of intravenous dipyridamole with technetium-99 m sestamibi tomographic imaging in predicting cardiac events and cardiac-related hospital admissions. J Am Coll Cardiol 1995;26:1202-8.CrossRefPubMedGoogle Scholar
- 8.Bishop CM. Pattern recognition. Mach Learn 2006;128:1-58.Google Scholar
- 9.Duda RO, Hart PE, Stork DG. Pattern classification. New York: Wiley; 2012.Google Scholar
- 12.Hachamovitch R, Hayes SW, Friedman JD, Cohen I, Berman DS. Stress myocardial perfusion single-photon emission computed tomography is clinically effective and cost effective in risk stratification of patients with a high likelihood of coronary artery disease (CAD) but no known CAD. J Am Coll Cardiol 2004;43:200-8.CrossRefPubMedGoogle Scholar
- 14.Hachamovitch R, Hayes SW, Friedman JD, Cohen I, Berman DS. Comparison of the short-term survival benefit associated with revascularization compared with medical therapy in patients with no prior coronary artery disease undergoing stress myocardial perfusion single photon emission computed tomography. Circulation 2003;107:2900-7.CrossRefPubMedGoogle Scholar
- 15.Berman DS, Kang X, Hayes SW, Friedman JD, Cohen I, Abidov A, et al. Adenosine myocardial perfusion single-photon emission computed tomography in women compared with men: Impact of diabetes mellitus on incremental prognostic value and effect on patient management. J Am Coll Cardiol 2003;41:1125-33.CrossRefPubMedGoogle Scholar
- 16.Berman DS, Kiat H, Friedman JD, Wang FP, Van Train K, Matzer L, et al. Separate acquisition rest thallium-201/stress technetium-99 m sestamibi dual-isotope myocardial perfusion single-photon emission computed tomography: A clinical validation study. J Am Coll Cardiol 1993;22:1455-64.CrossRefPubMedGoogle Scholar
- 17.Raschka S. Python machine learning. Birmingham: Packt Publishing Ltd; 2015.Google Scholar
- 19.Tibshirani R. Regression shrinkage and selection via the lasso. J R Stat Soc Ser B (Methodol) 1996;58:267-88.Google Scholar
- 20.Vapnik VN, Vapnik V. Statistical learning theory. New York: Wiley; 1998.Google Scholar
- 24.Cox DR. Regression models and life-tables. Breakthr Stat 1992;527-541.Google Scholar
- 27.Schapire RE. The boosting approach to machine learning: An overview. Nonlinear Estim Classif 2003;149-71.Google Scholar
- 29.Scholkopf B, Smola AJ. Learning with kernels: Support vector machines, regularization, optimization, and beyond. Cambridge: MIT Press; 2001.Google Scholar
- 31.Niculescu-Mizil A, Caruana R. Predicting good probabilities with supervised learning. In: Proceedings of the 22nd international conference on machine learning; 2005. p. 625–632.Google Scholar
- 32.Platt J. Probabilistic outputs for support vector machines and comparisons to regularized likelihood methods. Adv Large Margin Classif 1999;10:61-74.Google Scholar