OBJECTIVES: To introduce an algorithmic approach to improve the interpretation of myocardial perfusion images in women with suspected myocardial ischemia. BACKGROUND: Gated single photon emission computed tomography (SPECT) and magnetic resonance (MR) myocardial perfusion imaging (MPI) approaches have relatively poor diagnostic and prognostic value in women with suspected myocardial ischemia. Here we introduce an approach: Decisions Informed by Combining Entities (DICE) that forms a mathematical model utilizing MPI and cardiac dimensions generated by one modality to predict the perfusion status of another modality. The effect of the model is to systematically incorporate cardiac metrics that influence the interpretation of perfusion images, leading to greater consistency in designation of myocardial perfusion status between studies. METHODS: Women (n=213), with suspected myocardial ischemia, underwent MPI assessment for regional perfusion defects using two modalities: gated SPECT (n=207) and MR imaging (n=203). To determine perfusion status, MR data were evaluated qualitatively and semi-quantitatively while SPECT data were evaluated using conventional clinical criteria. These perfusion status readings were designated "Original". Four regression models were generated to model perfusion status obtained with one modality [e.g., semi-quantitative magnetic resonance imaging (MRI)] against another modality (e.g., SPECT) and a threshold applied (DICE modeling) to designate perfusion status as normal or low. The DICE models included perfusion status, left ventricular (LV) chamber volumes and myocardial wall thickness. Women were followed for 40±16 months for the development of first major adverse cardiovascular event (MACE: CV death, nonfatal myocardial infarction (MI) or hospitalization for congestive heart failure). Original and DICE perfusion status were compared in their ability to detect high-grade coronary artery disease (CAD) and for prediction of MACE. RESULTS: Adverse events occurred in 25 (12%) women and CAD was present in 34 (16%). In receiver-operator characteristic (ROC) analysis for CAD detection, the average area under the curve (AUC) for DICE vs. Original status was 0.77±0.03 vs. 0.70±0.03, P<0.01. Similarly, in Kaplan-Meier survival analysis the average log-rank statistic was higher for DICE vs. the Original readings (10.6±5.2 vs. 3.0±0.6, P<0.05). CONCLUSIONS: While two data sets are required to generate the DICE models no knowledge of follow-up results is needed. DICE modeling improved diagnostic and prognostic value vs. the Original interpretation of the myocardial perfusion status.