Automatic detection of coronary artery disease in myocardial perfusion SPECT using image registration and voxel to voxel statistical comparisons.

The purpose of this study was to compare the performance of automatic detection of coronary artery disease (CAD) with that of expert observers. A male and female normal image template was constructed from normal stress technetium-99m single photon emission computed tomography (SPECT) studies. Mean and standard deviation images for each sex were created by registering normal studies to a standard shape and position. The test group consisted of 104 patients who had been routinely referred for SPECT and angiography. The gold standard for CAD was defined by angiography. The test group studies were registered to the respective templates and the Z-score was calculated for each voxel. Voxels with a Z-score greater than 5 indicated the presence of CAD. The performance of this method and that of three observers were compared by continuous receiver operating characteristic (CROC) analysis. The overall sensitivity and specificity for automatic detection were 73% and 92%, respectively. The area (Az) under the CROC curve (+/-1 SE) for automatic detection of CAD was 0.88+/-0.06. There was no statistically significant difference between the performances of the three observers in terms of Az and that of automatic detection (P> or =0.25, univariate Z-score test). The use of this automated statistical mapping approach shows a performance comparable with experienced observers, but avoids inter-observer and intra-observer variability.

Comparison of methods for quantification of transient ischaemic dilation in myocardial perfusion SPET.

The purpose of this study was to compare six methods of measuring the left ventricular (LV) transient ischaemic dilation (TID) ratio during stress-rest myocardial perfusion single-photon emission tomography (SPET). The TID ratio was defined as the mean LV short-axis area at stress divided by the mean LV area of similar slices at rest. The centre of the LV wall was defined as either the maximum, mean or median of the radial short-axis count profiles. The area within the endocardial wall was also calculated for each definition of the LV wall centre. We identified 50 consecutive patients undergoing dipyridamole technetium-99m-tetrofosmin SPET imaging and angiography. Continuous receiver operating characteristic (CROC) analysis showed no significant difference between the six methods in terms of identifying severe coronary artery disease (P >0.47). Algorithms using the mean or the median value in the profile were significantly more robust than those using the maximum (P <0.0005). TID measured by all the algorithms is an indicator of severe coronary disease (P < 0.05). The algorithms compared provide a repeatable, quantitative and specific measure of the TID ratio.