Added value of quantitative [¹⁸F]FDG-PET analysis in MRI-negative epilepsy: A simulation-based study using realistic...

Quantitative analysis of [¹⁸F]FDG-PET images is expected to improve the localization of foci in non-lesional epilepsy. However, the lack of reliable gold standards has prevented a comprehensive evaluation of the potential improvements derived from this approach. Here, we aimed at evaluating these improvements using a novel dataset of realistic simulated studies.

125 realistic simulated [¹⁸F]FDG-PET studies were generated (100 with synthetic hypometabolic foci (HF) with different levels of identification complexity and 25 controls). Eight nuclear physicians performed visual rating (VR) and were given the chance to modify their assessment after reviewing quantitative results (QR). Physicians reported the presence/absence of HF, HF location, and diagnostic confidence (DC) before/after QR. Success Rate (SR) of physician's assessments was analyzed, as well as inter-rater agreement and changes in DC.

In 31.3% of the assessments, physicians changed their interpretation after QR, with SR increasing from 16.3% to 61.0% in these cases. Overall SR improved from 49.5% in VR to 63.5% in QR, mostly on pathologic cases (relative improvement: +34.0%). Improvement was found at each level of HF identification complexity and was higher for challenging cases (relative improvement: +71.8%). Inter-rater agreement also improved significantly (0.273 vs. 0.475, p < 0.001). QR also significantly increased DC ("High" confidence of 8.1% on VR vs. 38.5% on QR, p < 0.001).

Quantitative analysis significantly improved diagnostic accuracy, confidence and inter-rater agreement, especially in challenging cases. Furthermore, this work introduces a novel methodological approach using simulated MRI-negative epilepsy [¹⁸F]FDG-PET images for realistic quantification research studies.