Machine Learning: A Clinical Perspective on CV Disease

Bharath Ambale Venkatesh¹

¹Johns Hopkins University, United States

Synopsis

Machine learning methods are better suited for meaningful risk prediction in extensively phenotyped large-scale epidemiological studies than traditional methods or risk scores. This strategy could yield insights about specific use of variables for specific event prediction and guiding strategies to prevent cardiovascular disease outcomes. Potentially, these techniques could be applied retrospectively to analyze large data sets for identifying disease mechanisms, and as a means of hypothesis generation, without prior assumptions.

Target Audience

Researchers and clinicians interested in the maximization of existing datasets and in understanding clinical applications of machine learning.

Outcome/Objectives

Learners will be able to understand ways to look 'under the hood' instead of thinking of machine learning as a 'black box' approach. This course will present ways to extract meaningful clinical information beyond improvement in overall system performance from machine learning algorithms.

Purpose

To introduce machine learning techniques, such as LASSO-Cox and random survival techniques, that may be an effective statistical methodology for handling biomedical data of increased volume, velocity, and variety, under the curse of dimensionality. This is particularly true of imaging data where diagnosis is often complicated as there are several layers to obtained data that are often highly correlated -- regional variations in data, structural and functional measures. Some machine learning techniques do not require a priori assumptions on causality and may thus be suitable for defining the role of novel biomarkers in cardiovascular disease prediction.

Methods

There is a lack of studies using machine learning techniques with deep phenotyping (multiple evaluations of different aspects of a specific disease process) for cardiovascular event prediction. This course will examine the ability of combining deep phenotyping with machine learning for cardiovascular event prediction, with data from MESA (Multi-Ethnic Study of Atherosclerosis) as an example. This course will introduce random survival forests–based method of risk prediction as a means to superior prediction and improved accuracy than established risk scores.

This course will also introduce an overall paradigm for the use of deep learning for automated image analysis in conjunction with statistical machine learning for event prediction and risk stratification.

Results

The results of the algorithms and how to interpret them will be explained. The course will introduce methods to determine the value of including subclinical disease markers obtained from imaging, electrocardiography, and blood tests, using variable selection techniques. The course will introduce dimension-reduction techniques for the purpose of event prediction.

Discussion

This strategy could yield insights about specific use of variables for specific event prediction and guiding strategies to prevent cardiovascular disease outcomes. Potentially, these techniques could be applied retrospectively to analyze large phenotyping data sets for identifying disease mechanisms, and as a means of hypothesis generation, without prior assumptions. This course will introduce means to maximize the utility of imaging.

Acknowledgements

I thank Joao A Lima, David A Bluemke, Colin O Wu and Xiaoying Yang for their help in this work related to cardiovascular event prediction. I also would like to thank Alistair Young and Pau-Medrano-Gracia for their collaboration in work related to dimension-reduction techniques.

References

1. Ambale-Venkatesh, Bharath, et al. "Cardiovascular event prediction by machine learning: the Multi-Ethnic Study of Atherosclerosis." Circulation research (2017): CIRCRESAHA-117.

2. Ishwaran, H., Kogalur, U. B., Blackstone, E. H., & Lauer, M. S. (2008). Random survival forests. The annals of applied statistics, 841-860.

3. Tibshirani, R. (1997). The lasso method for variable selection in the Cox model. Statistics in medicine, 16(4), 385-395.

3. Medrano-Gracia, P., Cowan, B. R., Ambale-Venkatesh, B., Bluemke, D. A., Eng, J., Finn, J. P., ... & Young, A. A. (2014). Left ventricular shape variation in asymptomatic populations: the multi-ethnic study of atherosclerosis. Journal of Cardiovascular Magnetic Resonance, 16(1), 56.

4. Zhang, X., Ambale-Venkatesh, B., Bluemke, D. A., Cowan, B. R., Finn, J. P., Kadish, A. H., ... & Young, A. A. (2015). Information maximizing component analysis of left ventricular remodeling due to myocardial infarction. Journal of translational medicine, 13(1), 343.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)