Premium
Quantitative cardiac SPECT
Author(s) -
Peace Richard A.
Publication year - 2001
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.1398569
Subject(s) - voxel , receiver operating characteristic , cad , nuclear medicine , coronary artery disease , contrast (vision) , standard deviation , pattern recognition (psychology) , artificial intelligence , dipyridamole , mathematics , medicine , radiology , computer science , statistics , engineering drawing , engineering
This thesis studied automated statistical mapping in myocardial perfusion SPECT to detect coronary artery disease (CAD). Registering myocardial studies to a 3D template allows an analysis on a voxel by voxel basis. Normal mean and standard deviation templates were created for each sex by registering 25 male and 25 female studies to a standard shape and position. A test group of 104 patients undergoing dipyridamole technetium‐99m tetrofosmin SPECT and angiography were used to assess the automated method. Patients were divided into those with angiographic evidence of CAD ( n =56) and those without ( n =48). The test studies were registered to the templates and count normalized by minimizing the sum of absolute differences. A Z ‐score map of the statistical differences between registered study and template were calculated for all voxels within the myocardium. The contrast ( Z ‐score) and extent (number of voxels in a cluster exceeding the contrast threshold) thresholds for detection of CAD were optimized using receiver operating characteristic (ROC) analysis. The optimal thresholds resulted in a sensitivity of 73% and a specificity of 92% for automatic detection of CAD. The area under the fitted ROC curve (±1 SE) was 0.86±0.08 for a Z ‐score contrast threshold of 5. The performance of this method and that of three experienced observers was compared by continuous ROC analysis. There was no statistically significant difference between the performances of the three observers and that of automatic detection in terms of the area under the ROC curves ( p ⩾0.25). The use of this automated statistical mapping approach shows a performance comparable with experienced observers, but avoids observer variability.