Automated quantitation of 18F-flurpiridaz PET MPI and its diagnostic performance of coronary artery disease

Abstract Background/Introduction 18F-flurpiridaz, a novel positron emission tomography (PET) myocardial perfusion (MPI) imaging tracer, has shown efficacy in the detection of obstructive coronary artery disease (CAD) and is anticipated to be adopted in clinical practice soon. This post-hoc analysis of the 18F-flurpiridaz phase III clinical trial (NCT01347710) investigates the diagnostic performance of fully automated quantitative analysis by QPET software (Cedars-Sinai, Los Angeles, California, United States) for stress total perfusion deficit (S_TPD) and ischemic total perfusion deficit (I_TPD), compared to expert reader assessments in the detection of obstructive coronary artery disease (CAD) defined as stenosis ≥70%. Purpose The current standard method for interpreting PET MPI images involves visual assessment by a physician, which is inherently subjective and prone to high inter-reader variability. Quantification improves reproducibility of PET MPI evaluation. Methods This study retrospectively analyzes 753 cases from the original trial data, calculating area under the receiver operating characteristic curve (AUC) for S_TPD and I_TPD, and compares these values with summed stress scores (SSS) and summed difference scores (SDS) provided by three expert readers (R1, R2, R3), accounting for repeated readings as in the original trial, to evaluate diagnostic accuracy. Normal databases for this analysis were developed using 15 patients from the Phase II trial. Contours were verified by an experienced technologist. All other processing was fully automated. Results The contours were corrected in 2.4% stress and 4.4% rest cases. The AUC values for the readers’ SSS for stenosis ≥70% were found to be 0.812 (95% confidence interval [CI] 0.778 – 0.846) for R1, 0.801 (95% CI 0.766 – 0.836) for R2, and 0.828 (95% CI 0.795 – 0.860) for R3, compared to 0.812 (95% CI 0.778 – 0.846) for automatically quantified S_TPD (Figure 1A). The AUC values for the readers’ SDS for stenosis ≥70% were found to be 0.789 (95% CI 0.753 – 0.826) for R1, 0.780 (95% CI 0.744 – 0.816) for R2, and 0.798 (95% CI 0.761 – 0.834) for R3, compared to 0.813 (95% CI 0.778 – 0.848) for I_TPD (Figure 1B). Sensitivity of automated S_TPD quantification was 87% (95% CI 81% - 91%) and specificity was 57% (95% CI 52% - 61%) (Figure 2A). Sensitivity of I_TPD quantification was 83% (95% CI 77% - 87%) and specificity was 62% (95% CI 58% - 66%) (Figure 2B). Almost all quantitative results were similar to the readers’ results (p = NS), except for the comparison of AUC for I_TPD to SDS for R2, where I_TPD was higher (p = 0.04) (Figure 1B). Conclusion Quantitative analysis is comparable to expert reader evaluations in detecting ≥70% stenosis in CAD but can be derived consistently and automatically. The incorporation of TPD quantitative measures could standardize and potentially enhance the diagnostic process for obstructive CAD.