R Kuker¹*, J Alderuccio², A Rincon³, W Zhao³, C Moskowitz¹, F Yang⁴, (1) Department Of Radiology, University Of Miami, Miami, FL, (2) Division of Hematology, Department of Medicine, University Of Miami, Miami, FL,(3) Department Of Biomedical Engineering, University Of Miami, Coral Gables, FL, (4) Department Of Radiation Oncology, University Of Miami, Miami, FL

• 
  R Kuker

PO-GePV-I-73 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: The maximum standardized uptake value (SUV Max) of the dominating lesion plays a relevant role in the management of patients with diffuse large B-cell lymphoma (DLBCL). Determination of SUV Max, however, remains largely a manual process, not suitable for large-scale production and with potential vulnerability to bias. The current study aimed to develop a fully automated method to determine SUV Max on FDG-PET for DLBCL.

Methods: The study cohort included 40 patients with newly diagnosed DLBCL who were randomly selected from the Alliance/CALGB 50303 (NCT00118209) trial. The automated SUV Max determination started with deploying a deep convolutional neural network on the CT to segment anatomic structures with avid physiologic FDG uptakes, such as the brain, heart, kidneys, and bladder. Contours obtained for these structures were then automatically adjusted to account for their respective PET presentations using an array of ad-hoc algorithms including active contour and fast matching. The final SUV Max determination was made within the imaging body volume excluding the aforementioned anatomic structures. The automated SUV Max readings were compared with the accompanying manual readings from the trial for agreement in terms of Pearson’s correlation and the limit of agreement by Bland-Altman.

Results: The SUV Max manual readings for the study cohort ranged from 6.1 to 53.5 with mean±SD of 24.0±11.2 while ranging from 9.3 to 54.6 with mean±SD of 24.6±10.9 for the automated readings. The Pearson’s correlation coefficient between the manual and the automated readings was statistically significant at 0.98 (p < 0.0001). Bland-Altman analysis of the agreement between the manual and the automated readings yielded a bias of 0.62 with lower/upper limit of agreement of -3.3 and 4.5.

Conclusion: Promising preliminary results obtained have demonstrated that the proposed method may serve as a robust alternative to manual annotation for determining SUV Max on FDG-PET for DLBCL.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email