Characteristics of supine myocardial perfusion SPECT (MPS) which predict normalization on prone imaging could simplify acquisition and aid image interpretation

Abstract

Background: Imaging in the prone position is known to differentiate true perfusion defects from abnormalities created by artifacts seen on supine MPS images. If normalization on prone imaging could be predicted, additional prone images could be avoided while maintaining diagnostic accuracy. Methods: Two blinded observers identified 236 coronary vascular areas in 176 patients, 124 men and 52 women, with abnormal or possibly abnormal supine MPS who had prone images after stress performed at UCSF between 9 and 11/03. Nineteen image features based on gender, defect location, intensity, variation at rest MPS (when acquired), evidence of attenuation or motion, cavitary dilation, lung uptake, background activity, counts density and wall motion (WM) were related by multiple logistic regression (LR) and a C4.5 decision tree (DT) analyses to the likelihood of defect normalization on prone imaging. The overall population was divided into test and validation groups. Coronary angiograms within 2 years of MPS were available in 48 patients. Results: Prone images agreed with angiographic findings in 41 of 48 (86%) patients. LR predicted prone image findings in all cases but required input from all 19 factors. In both men and women 10 fold stratified cross validation predicted prone image normalization (p < 0.05) by non-segmental, low intensity, shifting defects, patient motion and low count images, and persistent prone defects by dense, segmental, stress induced defects, cavitary dilation, lung uptake and WM abnormalities. DT analysis simplified the findings and identified image findings and their combination with high predictive accuracies in the validation group, where a dense defect, a low intensity defect with either abnormal WM, or with an abnormal polar map (PM) and segmental defect were predictive of persistence on prone, while a normal PM, or a low intensity defect with either normal WM, or with a normal PM with a non-segmental defect predicted prone normalization with accuracies over 93%. Conclusion: A few critical findings on supine MPS predict the findings on prone images. These findings aid image interpretation and reduce the need for added prone images.