A new algorithm for the quantitation of myocardial perfusion SPECT. I: technical principles and reproducibility.

UNLABELLED We have developed a new, completely automatic 3-dimensional software approach to quantitative perfusion SPECT. The main features of the software are myocardial sampling based on an ellipsoid model; use of the entire count profile between the endocardial and epicardial surfaces; independence of the algorithm from myocardial shape, size, and orientation and establishment of a standard 3-dimensional point-to-point correspondence among all sampled myocardial regions; automatic generation of quantitative measurements and 5-point semiquantitative scores for each of 20 myocardial segments and automatic derivation of summed perfusion scores; and automatic generation of normal limits for any given patient population on the basis of data fractionally normalized to minimize hot spot artifacts. METHODS The new algorithm was tested on the tomographic images of 420 patients studied with a rest 201TI (111-167 MBq, 35 s/projection)-stress 99mTc-sestamibi (925-1480 MBq, 25 s/projection) separate dual-isotope protocol on a single-detector camera, a dual-detector 90 degrees camera, and a triple-detector camera. RESULTS The algorithm was successful in 397 of 420 patients (94.5%) and 816 of 840 image datasets (97.1%), with a statistically significant difference between the success rates of the 201TI images (399/ 420, or 95.0%) and the 99mTc images (417/420, or 99.3%; P < 0.001). Algorithm failure was caused by extracardiac uptake (10/24, or 41.7%) or inaccurate identification of the valve plane because of low count statistics (14/24, or 58.3%) and was obviated by simply limiting the image volume in which the software operates. Reproducibility of measurements of summed perfusion scores (r = 0.999 and 1 for stress and rest, respectively), global defect extent (r = 0.999 and 1 for stress and rest, respectively), and segmental perfusion scores (exact agreement = 99.9%, kappa = 0.998 for stress and 0.997 for rest) was extremely high. CONCLUSION Automatic 3-dimensional quantitation of perfusion from 201Tl and 99mTc-sestamibi images is feasible and reproducible. The described software, because it is based on the same sampling scheme used for gated SPECT analysis, ensures intrinsically perfect registration of quantitative perfusion with quantitative regional wall motion and thickening information, if gated SPECT is used.