Objectives: To compare PET/CT-derived quantitative metrics obtained using conventional Syngo Via software and the artificial intelligence (AI)-based Research Consortium for Medical Image Analysis (RECOMIA) platform in colorectal cancer (CRC), and to evaluate measurement differences in the absence of a gold standard for diagnosis. Material and Methods: We conducted a retrospective analysis of 18 F-Fluorodeoxyglucose PET/CT scans from 15 patients with CRC. Both platforms were used to gather the primary metrics that were measured, such as maximum standardised uptake value (SUVmax), mean SUV (SUVmean), MTV, and TLG. The agreement between Syngo Via and RECOMIA was assessed using Bland–Altman analysis. Results: While the differences in SUVmax were not deemed statistically significant (P = 0.2058), RECOMIA demonstrated lower values for SUVmean and higher values for MTV and TLG (P values of 0.0001, 0.0003, and 0.0312, respectively). Significant variability was found in the confidence intervals, showing platform-dependent measurement errors. Conclusion: When using the RECOMIA platform for AI-based segmentation, the SUVmean values were lower, whereas MTV and TLG values were higher than when using Syngo Via for traditional segmentation. These differences are attributed to the method of measurement rather than being influenced by pathological volume or clinical outcomes, indicating that they do not necessarily reflect the accuracy of the measurements. Further research is required to determine the clinical relevance and impact of these observed differences.
Kheruka et al. (Thu,) studied this question.