High-quality imaging matters: interscan reproducibility of CT derived fractional flow reserve

Abstract Background Coronary computed tomography derived fractional flow reserve (FFRCT) offers a comprehensive, non-invasive evaluation of the physiological significance of coronary lesions. However, the interscan reproducibility of FFRCT analysis has yet to be investigated. Objectives To evaluate the interscan reproducibility of FFRCT and the impact of clinical and technical factors on interscan reproducibility. Methods A total of 102 patients with coronary artery disease underwent two coronary computed tomography angiography (CCTA) scans within a one-hour interval using identical preparation and acquisition protocols. FFRCT computation was performed in a blinded manner at a core laboratory (HeartFlow, Inc., Mountain View, CA, USA). FFRCT recordings were conducted by an independent operator blinded to clinical information. The FFRCT value 2 cm distal to obstructive stenoses (FFRCT2cm) and the distal-vessel FFRCT (FFRCTdistal) value was recorded in all scans, and the trans-lesional gradient (ΔFFRCT) was calculated. Reproducibility was assessed using intraclass correlation coefficients, limits of agreement (LoA) and Bland Altman plots. Sub-analyses investigated the impact of image quality (adequate versus excellent), iterative reconstruction level (ADMIRE Siemens, 2 versus 2), Agatston score (≤400 versus 400), and heart rate (≤65 versus 65) on LoA and were compared using an F-test. A Bayesian prediction model was applied to predict the probability of patients’ estimated true (eTrue) FFRCT2cm ≤0.80 from a single FFRCT2cm value. Results Eighty-two patients had FFRCT analyzed in paired CCTA scans (mean age of 64 ±9 years, 79% male). No significant differences in median FFRCT values were observed between scans. Interscan correlation was high across all FFRCT analyses (ICC=0.81-0.92). The interscan agreement was highest for FFRCT2cm and ΔFFRCT, while the lowest agreement was observed for FFRCTdistal (Table 1). Excellent CCTA image quality, lower iterative reconstruction level, and lower calcific burden were associated with narrower limits of agreement for FFRCT2cm and ΔFFRCT, while only calcific burden affected LoA for FFRCTdistal (Figure 1A). The Bayesian prediction analysis demonstrated a 95% probability of an eTrue FFRCT2cm ≤0.80 when the calculated FFRCT2cm ≤0.70, and 5% if the calculated FFRCT2cm ≥0.85. In patients with paired scans of excellent image quality, there was a 95% probability of eTrue FFRCT ≤0.8 when FFRCT2cm ≤0.74 and 5% probability when FFRCT2cm ≥0.83 (Figure 1B). Conclusions Overall, this study demonstrated modest test-retest reproducibility of FFRCT in patients with CAD. However, in the event of excellent quality CCTA images reproducibility was high, in particularly for the FFRCT2cm value. These findings emphasize the importance of good CCTA image quality in obtaining high FFRCT interscan reproducibility.