Objectives

Developed an automated approach for evaluating subendocardial analysis for stress total perfusion deficit (TPD) and ischemic TPD using 18F-flurpiridaz PET myocardial perfusion imaging (MPI).
Demonstrated that subendocardial analysis improves the detection of coronary artery disease (CAD) compared to transmural quantitative analysis or expert visual reading, especially for moderate (≥50%) stenosis.
Showed that subendocardial and transmural TPD achieved diagnostic performance comparable to or greater than that of the readers' assessments across different subgroups.

Methodology

Retrospective analysis of data from the 18F-flurpiridaz phase III clinical trial (flurpiridaz-301, NCT01347710).
Automated definition of the subendocardial layer on left ventricular contours.
Derivation of polar maps for both subendocardial and transmural analysis.
Quantification of stress and rest TPD using QPET software, with ischemic TPD calculated as the difference between stress and rest TPD.
Development of normal databases using 15 patients with visually normal studies from the phase II 18F-flurpiridaz trial.

In 753 cases, the area under the receiver operating characteristic curve (AUC) for detecting ≥50% stenosis was significantly higher for subendocardial than transmural analysis for stress (0.795 vs. 0.762, p=0.013) and ischemic (0.795 vs. 0.767, p=0.049) TPD.
Subendocardial TPD achieved diagnostic performance greater than or comparable to readers' assessments in the total population and across subgroups (sex, stress type, BMI).
For ≥70% stenosis, AUC results for subendocardial and transmural quantitation were similar (p=0.339 for stress TPD, p=0.978 for ischemic TPD).

The study demonstrates a novel and potentially clinically relevant application of subendocardial analysis in 18F-flurpiridaz PET MPI. However, the reliance on invasive coronary angiography (ICA) as the reference standard, rather than a functional measure like fractional flow reserve (FFR), is a limitation.
While the automated quantification is promising, the need for manual correction of contours in a small percentage of cases (2.4% stress, 4.3% rest) suggests further refinement is needed for full automation.
The study lacks test-retest variability data, which is crucial for assessing the reliability of the new technique.
The normal database was developed using only 15 patients. A larger and more diverse normal database would strengthen the generalizability of the findings.