A new algorithm for the quantitation of myocardial perfusion SPECT. II: Validation and diagnostic yield

This study validates a new quantitative perfusion SPECT algorithm for the a ssessment of myocardial perfusion. The algorithm is not based on slices and provides fully 3-dimensional sampling and analysis independent of assumpti ons about the geometric shape of the left ventricle, Methods: Radiopharmace utical- and sex-specific normal limits and thresholds for perfusion abnorma lity in 20 segments of the left ventricle were developed for separate, dual -isotope rest (TI)-T-201-exercise Tc-99m-sestamibi SPECT in 36 patients wit h <5% before-scanning likelihood of coronary artery disease (CAD) (group 1) and 159 patients with perfusion abnormalities (group 2). These thresholds were validated in 131 patients (group 3) by comparison with expert visual i nterpretation. Thresholds for automatic segmental scores were developed and validated for groups 2 and 3, respectively. The accuracy of CAD detection was assessed in 94 patients, who underwent coronary angiography (group 4). Results: Overall sensitivity for detection of stress and rest segmental per fusion abnormality was 91% and 96%, respectively, for men and 89% and 79%, respectively, for women. Overall specificity for stress and rest was 87% an d 90%, respectively, for men and 88% and 90%, respectively, for women. Agre ement between automatic and visual scores was good (weighted kappa of 0.71 and 0.60 for stress and rest images, respectively). Sensitivity and specifi city were 88% for the detection of greater than or equal to 70% stenosis. F or the detection of left anterior descending, left circumflex, and right co ronary artery stenosis, sensitivity was 84%, 86%, and 88%, respectively, an d specificity was 84%, 88%, and 81%, respectively. Conclusion: The new quan titative perfusion SPECT approach is highly sensitive and specific for the detection and localization of CAD, provides accurate automatic scores for t he assessment of regional perfusion, and overcomes the low-specificity limi tations of previous quantitative algorithms.