Author + information
- Robi Goswamia,b,
- Thomas Stuckeya,b,
- Frederick Meinea,b,
- Narendra Singha,b,
- Jeremiah Deptaa,b,
- Sunny Guptaa,b,
- Shyam Ramchandania,b,
- R. Susan Crawforda,b,
- Tim Burtona,b and
- William E. Sanders Jr.a,b
Background: Heart failure is a progressive disease affecting approximately 6 million people in the United States. Left ventricular ejection fraction (LVEF) is used to guide therapy and determine cardiac risks. Methods used to determine LVEF include ventriculography, multigated acquisition scan, echocardiogram, and cardiac magnetic resonance (cMR) imaging. Cardiac phase space tomography, a novel technology based on computer machine learning, analyzes intrinsic thoracic signals and assesses ejection fraction within minutes. Ventriculograms obtained during the Coronary Artery Disease Learning and Algorithm Development (CADLAD) Study were used to determine diagnostic performance for computing LVEF.
Methods: Computation of LVEF was an additional endpoint of the CADLAD prospective multicenter study to develop a machine-learned algorithm to detect significant CAD. Noninvasive, resting phase signal data was acquired from subjects with symptoms suggestive of CAD prior to ventriculography using a cardiac Phase Space Tomography Analysis (cPSTA) System. The features used for computing LVEF are topographical representations of phase space which are reconstructed into a three-dimensional image of the heart. A machine learned algorithm assessed LVEF from the acquired phase space signals. An LVEF threshold of <50% was considered abnormal.
Results: Ventriculography results plus phase space signals from 96 subjects were used to generate a machine learned algorithm that computes LVEF. It was tested on 29 naïve signals. A total of 102 subjects (81.6%) had LVEF of >50% and 23 subjects (18.4%) had LVEF of <50%. Phase space tomography analysis distinguished LVEF's of <50% with a sensitivity of 86% (95% CI: 72-94%), specificity of 65% (95% CI: 58-70%), and NPV 96% (95% CI: 91-98%).
Conclusions: cPSTA System is a noninvasive tomographic imaging method which can be employed to assess LVEF. This technique readily distinguishes normal from abnormal EF (<50%) in this small cohort of subjects. Further machine learning and enrollment are required to increase the accuracy of the assessment. In addition, future studies will include pairing phase signal data with the gold standard (cMR) measurement of LVEF.
Poster Hall, Hall C
Sunday, March 19, 2017, 9:45 a.m.-10:30 a.m.
Session Title: The Evolving World of LVADs, Transplant and Other Novel Discoveries
Abstract Category: 13. Heart Failure and Cardiomyopathies: Clinical
Presentation Number: 1294-288
- 2017 American College of Cardiology Foundation