Automated analysis of cardiac SPECT on par with experts

Fully automated quantitative review of attenuation-corrected and noncorrected myocardial perfusion SPECT data provided performance at least equivalent to expert analysis for detection of stenosis of 70 percent or greater, according to a study published online Jan. 11 in The Journal of Nuclear Medicine.

Quantitative analysis of MPS, which the American Society of Nuclear Cardiology currently recommends as an adjunct to visual review, can supplement visual analysis and provide greater reproducibility. In addition, attenuation-corrected (AC) MPS has been shown to improve diagnostic accuracy compared with noncorrected (NC) MPS.

Reza Arsanjani, MD, from the departments of imaging and medicine and Cedars-Sinai Heart Institute in Los Angeles, and colleagues devised a study to compare automated analysis of NC and AC MPS for detection of constructive coronary artery disease (CAD) with visual scoring of experienced readers.

The researchers focused on patients referred for rest and stress MPS from March 1, 2003, to Dec. 31, 2006, excluding those with a history of CAD or significant valve disease. The cohort included 345 low likelihood patients who performed an adequate stress test with a less than 5 percent likelihood of CAD. The remaining 650 individuals underwent correlative coronary angiography exams.

NC and AC images were analyzed via an automated process, and images were independently interpreted based on a 17-segment model by two experienced cardiologists using a 5-point scoring system.

The readers employed a step-wise approach to interpretation, starting with NC images, then AC images, results of computer analysis and clinical information, with the ability to rescore the rest and stress studies using AC data.

For the coronary angiography group, a stenosis of 50 percent or more for the left main or 70 percent or more for other coronary arteries served as the gold standard for detection of CAD.

Arsanjani and colleagues found high agreement between the two readers and between each reader and the automated results. However, one reader consistently had higher sensitivity and the other consistently higher specificity, which underscores the challenge of providing a definitive diagnosis based on visual interpretation of MPS.

Visual analysis improved by at least 3 percent for readers and the software with the addition of AC data. Based on these findings, the researchers recommended that AC information should be interpreted if it is available.

The automated analysis was comparable to visual analysis of one reader on a per-vessel basis and superior to the other reader. Neither physician achieved diagnostic accuracy greater than the computer for any territory.

“The reading experts in our study were attending physicians from premier imaging centers with extensive experience in MPS interpretation. It is therefore likely that a fully automatic analysis could play an integral role for the less experienced reader who may be less certain about normal variation in uptake,” wrote Arsanjani et al.

The researchers noted several shortcomings to the study, including the use of visual coronary angiography interpretation as the gold standard. They suggested further prognostic studies to demonstrate the superiority of the automated method and a multicenter evaluation to confirm these results with other SPECT cameras and attenuation-correction systems.