A Deep Learning-Aided Automated Method for Calculating Metabolic Tumor Volume in Diffuse Large B-Cell Lymphoma

Area of Science:

• Oncology
• Radiology
• Medical Imaging

Background:

• Metabolic tumor volume (MTV) is a crucial prognostic biomarker in diffuse large B-cell lymphoma (DLBCL).
• Current methods for MTV calculation are semiautomatic, requiring manual input that hinders routine clinical research application.

Purpose of the Study:

• To develop and validate a fully automated method (AM) for calculating MTV in DLBCL.
• To compare the AM's MTV calculations against manual assessments by nuclear medicine (NM) readers.

Main Methods:

• A deep convolutional neural network was utilized to segment physiologic structures from CT scans with PET avidity.
• The AM was validated on a cohort of 100 newly diagnosed DLBCL patients from the Alliance/CALGB 50,303 trial.

Main Results:

• The AM demonstrated high concordance with NM readers for MTV calculations, with Pearson's correlation coefficients and interclass correlations exceeding 0.98 (p < 0.0001).
• Bland-Altman plots indicated minimal systematic errors between the AM and readers for both MTV and SUVmax.

Conclusions:

• The fully automated method for MTV calculation shows high accuracy and concordance with expert readers.
• This automated approach has the potential to facilitate the integration of PET-based biomarkers into clinical trials for DLBCL.