AI-assisted [<sup>18</sup>F]FDG PET/CT quantification: a time-saving approach in large B-cell lymphoma?
Authors
Affiliations (5)
Affiliations (5)
- Department of Nuclear Medicine, Amiens-Picardy University Hospital, 1 Rue du Professeur Christian Cabrol, Amiens, France.
- Department of Radiology, New York Presbyterian Hospital - Columbia University Irving Medical Center, New York, NY, 10039, USA.
- Department of Nuclear Medicine, Eugène Marquis Center, Rennes, France.
- INSERM, LTSI-UMR 1099, University of Rennes, Rennes, France.
- Department of Nuclear Medicine, Amiens-Picardy University Hospital, 1 Rue du Professeur Christian Cabrol, Amiens, France. [email protected].
Abstract
To assess if AI-assisted segmentation on [<sup>18</sup>F]FDG PET/CT significantly reduces time in quantifying Total Metabolic Tumor Volume (TMTV) and the greatest distance between two lesions (Dmax) in Large B-cell lymphoma (LBCL) patients. A retrospective study at two institutions included 56 LBCL patients (Ann-Arbor staging: 14% with no disease, 25% stage I-II, 61% stage III-IV). Two methods were compared: AI-assisted segmentation (Pionus<sup>®</sup> v1.1.0, PAIRE, Paris) automatically identified lesions with expert correction of AI errors, while expert-led segmentation was fully manual. Both methods used a 41% SUVmax threshold. AI-assisted segmentation was 41% faster, saving an average of 104 s per patient (152 ± 134 s vs. 255 ± 270 s, p < 0.001), with time saved correlating with TMTV (r = 0.67, p < 0.001). No significant differences in TMTV or Dmax were found between methods. Intraclass correlation coefficients were high for both TMTV (0.91 and 0.95) and Dmax (0.97 and 0.95), with strong inter-reader agreement. AI-assisted segmentation significantly reduces time for TMTV and Dmax measurement in LBCL patients, especially in patients with high tumor burden, while maintaining high reliability and inter-reader agreement. The impact on the integrated clinical workflow has yet to be assessed.