Author + information
- Reza Arsanjani,
- Yuan Xu,
- Damini Dey,
- Aryeh Shalev,
- Rine Nakanishi,
- Sean Hayes,
- Mathews Fish,
- Daniel Berman,
- Guido Germano and
- Piotr Slomka
We aimed to improve the diagnostic accuracy of myocardial perfusion SPECT (MPS) by integrating clinical data and quantitative image features by machine learning (ML).
1183 rest/stress 99mTechnetium MPS studies (715 consecutive cases with correlating invasive angiography and 468 with low likelihood of coronary artery disease (CAD) < 5% were considered. Cases with stenosis < 70% and low likelihood were considered normal. Total stress perfusion deficit (TPD) for supine/prone (S/P) data, stress/rest perfusion change and transient ischemic dilatation were derived by automated perfusion quantification software and were combined with age, sex, and post-electrocardiogram CAD probability by boosted ensemble ML algorithm (LogitBoost). The diagnostic accuracy of the model for prediction of obstructive CAD was compared to standard quantification and to visual analysis by two experienced readers utilizing all imaging, quantitative, and clinical data. 10-fold stratified cross-validation was performed.
The Receiver-Operator-Characteristics areas-under-curve, sensitivities, specificities, and accuracies are shown in the Table. Diagnostic performance of ML was similar to Expert 1, significantly better than Expert 2 and better than quantitative S/P TPD analysis.
ML significantly improves diagnostic performance of MPS by computational integration of quantitative perfusion and functional data to the level rivaling expert analysis.
|Classifier||Sensitivity %||Specificity %||Accuracy %||Areas-under-curve|
|Machine Learning (ML)||78.9 (±3.9)||91.9 (±1.9)||87.2 (±1.9)||0.94 (±0.01)|
|Expert 1||76.3 (±4.1)||91.2 (±2.0)||85.9 (±2.0)||0.88 (±0.01)*|
|Expert 2||71.1 (±4.3)*||88.0 (±2.3)||82.0 (±2.2)*||0.84 (±0.01)*|
|S/P TPD||77.0 (±4.0)||84.8 (±2.6)*||82.8 (±2.2)*||0.87 (±0.01)*|
↵* worse than ML P < 0.01 (95% Confidence interval)
West, Room 3001
Sunday, March 10, 2013, 11:45 a.m.-Noon
Session Title: Nuclear Cardiology and PET: Pushing the Boundaries
Abstract Category: 21. Imaging: Nuclear
Presentation Number: 926-7
- 2013 American College of Cardiology Foundation