Toward reproducible metabolic tumor volume quantification in endometrial cancer: optimizing [¹⁸F]FDG PET/CT tumor segmentation methods.

BACKGROUND

Metabolic tumor volume (MTV) derived from [18F]FDG PET/CT has shown promise in improving preoperative staging and prognostication in endometrial cancer, both as a standalone biomarker and as a foundation for PET-based tumor radiomics. Despite its potential, standardized segmentation methods are lacking. This study aimed to (1) evaluate interobserver agreement in MTV delineation, (2) compare MTVs obtained through various thresholding approaches by examining their correlation with anatomical tumor volume (ATV) from MRI, and (3) assess their value for predicting aggressive disease.

RESULTS

This study included 146 patients with histologically confirmed endometrial cancer who underwent preoperative PET/CT imaging on a Siemens TruePoint (n = 67) or a Siemens Biograph Vision scanner (n = 79). MTV2.5 defined by a standardized uptake value (SUV) threshold larger than 2.5, demonstrated excellent reproducibility between readers, with an intraclass correlation coefficient (95% confidence interval (CI)) of 0.97 (0.96, 0.98). MTV40%, defined by SUV > 40% of SUVmax, yielded an ICC (95% CI) of 0.91 (0.87, 0.93). Both the fixed 2.5 threshold and all relative SUVmax-based thresholds (20-60% SUVmax) produced MTVs with strong correlation to ATV, with Spearman's rank correlation coefficients ranging from 0.806 to 0.874 (p < 0.001). The closest volumetric match to ATV was observed for MTV30%, with a median difference (95% CI) of -4% (-11%, 2%). In terms of predictive performance, MTV2.5 and MTV20-50% yielded comparable results for predicting lymph node metastases, advanced FIGO stage (III-IV), and 3-year disease specific survival, with area under receiver operating characteristic curve (AUC) values (95% CI) ranging from 0.68 (0.53, 0.82) to 0.77 (0.65, 0.88). MTV60% showed slightly lower prediction values across all outcomes, with AUCs (95% CI) between 0.64 (0.49, 0.80) and 0.72 (0.63, 0.82).

CONCLUSION

MTV2.5 has excellent reproducibility and correlates well with anatomical volume. Both MTV2.5 and MTV20-50% thresholds yield similar performances for predicting aggressive disease features, supporting their potential for clinical implementation in preoperative assessment of endometrial cancer.