Wednesday, December 3 | 1:50 p.m.-2:00 p.m. | W6-SSNMMI06-3 | S405
Automated, AI-based tumor segmentation using standardized uptake value (SUV)-based thresholds derived from F-18 FDG-PET/CT scans shows very high correlation with manual whole-body tumor volume assessments, according to a study in this session.
Presenter Ricarda Ebner, MD, of Ludwig Maximilian University Hospital in Munich, and colleagues collected imaging from 123 patients with non-small cell lung cancer (NSCLC) who underwent F-18 FDG-PET/CT. Patients each had a baseline scan and up to five follow-up scans. Manual volumetric segmentation of all FDG-avid tumor lesions was performed and compared with automated segmentation using a CE-marked, AI-driven model (Pionus v1.2.2).
The researchers calculated automated metabolic tumor volume (TMTV) metrics -- TMTV41% and TLG41% (41% SUVmax threshold) and TMTVSUV4 (fixed threshold SUV > 4) -- separately for each time point and used Pearson correlation coefficients (r) to assess the linear relationship between manually defined PET tumor volume and AI-derived metrics.
At baseline, PET tumor volume showed very high correlations with automated metrics (TMTV41%: r = 0.892; TMTVSUV4: r = 0.983; TLG41%: r = 0.92; all p < 0.001). Across all time points, correlations remained robust but were moderately lower (TMTV41%: r = 0.723; TMTVSUV4: r = 0.867; TLG41%: r = 0.821; all p < 0.001).
Ultimately, validating automated volumetric analysis is clinically relevant, as it may simplify workflows, reduce interobserver variability, and support consistent assessment over time and across institutions, the researchers noted.
“Automated PET volumetry enables high-throughput, operator-independent tumor burden quantification, offering a scalable tool for standardized response assessment and improved clinical decision-making in NSCLC,” they concluded.
Check out this session to learn the details.
