OGC O08 Quantification of fluorescence angiography for visceral perfusion assessment in gastric conduit reconstruction: measuring agreement between two software algorithms

Abstract

Abstract Background Building evidence suggests that fluorescence angiography is capable of reducing anastomotic leakage rates. Many software-based quantification algorithms have been developed aiming to obtain an objective measure for intraoperative visceral perfusion at the anastomotic site. Validation and comparison of these models in the current literature is scarce. The aim of this study was to measure agreement between two independently developed software algorithms for the quantitative assessment of visceral perfusion using indocyanine green fluorescence angiography. Methods 81 video recordings from patients who underwent fluorescence angiography-guided Ivor Lewis esophagectomy with gastric conduit reconstruction between August 2020 and February 2022 were available for analysis. In this dataset, we standardized patient, theatre, camera and imaging display variables. Recordings were analyzed retrospectively based on intraoperatively selected regions of interest using two software based-quantification algorithms (AMS and CPH). Quantitative parameters derived from the fluorescence time curve (FTC) were recorded. The comparison focused on the predefined primary outcome variable for quantitative perfusion assessment "normalized maximum slope". Agreement between the two software algorithms was evaluated in a Bland-Altman analysis. Quantitative parameters in patients with anastomotic leakage were compared to those without. Results Out of 81 video recordings, 70 could be processed using both algorithms. The output values for the normalized maximum slope from the AMS and the CPH software held a moderate-to-strong correlation in a linear regression model (R-squared: 0.544, p<0.001). The Bland-Altman plot indicated a mean±SD difference of 0.037±0.018 arbitrary units for the normalized maximum slope, with significant differences between the two software measurements (p<0.001). Proportional bias was observed using linear regression in the Bland-Altman plot (R-squared=0.715, p<0.001). Disparities between the two measurement methods became proportionally larger among higher mean values. Neither of the two software methods showed a significant difference in perfusion measurement between the anastomotic leakage and no leakage groups (p=0.32 and p=0.51). Differences in signal processing, such as smoothing techniques and window selection, were identified as potential factors contributing to the disparity in outcomes. Conclusions The two software-based quantification methods for visceral perfusion demonstrated significant differences. Varying agreement among quantification methods should be considered while interpreting studies that report quantitative parameters and derived thresholds for clinical endpoints, as there may be a limited external validity.