Improved accuracy of myocardial perfusion SPECT (MPS) for detection of coronary artery disease (CAD) by utilizing a support vector machine (SVM) algorithm

Objectives We aimed to improve the diagnostic accuracy of automatic MPS interpretation analysis by integrating several quantitative perfusion and functional variables for attenuation-corrected (AC) and non-corrected (NC) data by a SVM statistical learning algorithm (probabilistic binary linear classifier), because algorithms for integrating these variables have not been well developed. SVM takes a set of input data with multiple variables and for each given input pattern, predicts two possible outcomes.

Methods 957 rest/stress 99m-technetium gated MPS AC and NC studies from 623 consecutive patients (pts) with correlating angiography and 334 with likelihood of CAD <5% (LLK) were assessed. Pts with stenosis <70% and LLK pts were considered normal. Total perfusion deficit (TPD) was computed automatically for AC (TPD-AC) and NC (TPD-NC) data sets. In addition, ischemic (ISCH) (JNM.2004;45(2):183-91) and ejection fraction changes (EFC) between stress and rest were derived by quantitative software. TPD ≧ 3% and EFC ≧ 5% were considered abnormal. The SVM was trained using a group of 125 pts (25 LLK, 25 0-, 25 1-, 25 2- and 25 3-vessel CAD) using above quantitative variables and polynomial fitting. The remaining pts (N = 832) were divided by SVM into 2 categories (CAD, NO-CAD) in the testing phase.

Results The Receiver-Operator-Characteristics areas-under-curve (ROC-AUC), sensitivities, specificities, and accuracies are shown in the table. The AUC for the SVM algorithm combining perfusion and functional MPS was significantly higher than TPD NC, EFC, and TPD NC+AC+ISCH (P <0.02). In addition, the sensitivity, specificity, and accuracy of the combined method were higher than TPD NC or AC, or EFC method (P < 0.04).

Conclusions A SVM approach allows improvement of diagnostic accuracy, sensitivity and specificity of MPS by computational integration of several quantitative perfusion and functional variables.

Research Support This research was supported in part by grant R0HL089765-01 from the National Heart, Lung, and Blood Institute/National Institutes of Health (NHLBI/NIH)

• Download figure
• Open in new tab
• Download powerpoint