A prognostic index integrating deep learning baseline PET/CT biomarkers and multi-omics profiling in diffuse large B cell lymphoma

[ 18 F]-Fluorodeoxyglucose (FDG)-positron emission tomography (PET)/computed tomography (CT) is essential for disease staging and treatment response evaluation in diffuse large B cell lymphoma (DLBCL). In this study, we analyze 18 F-FDG-PET scans from 1,024 newly diagnosed DLBCL patients, integrating with DNA and RNA sequencing data. Using the nnUNet deep learning framework and training on both AutoPET public and in-house datasets, we identify key baseline biomarkers—including total metabolic tumor volume (TMTV), Max MTV, and the standardized tumor dissemination biomarker—that demonstrate significant prognostic value. Further integrating PET biomarkers with clinical factors and LymphPlex genetic subtypes, we develop high TMTV, elevated lactate dehydrogenase (LDH), and EZB-like MYC+, MCD-like, and TP53 Mut subtypes as risk factors to form the ClinicalPET LymphPlex model, efficiently distinguishing patient outcomes across different treatments. Notably, high TMTV correlates with an immunosuppressive tumor microenvironment, while elevated LDH is linked to increased metabolic activity and tumor proliferation. Collectively, our findings necessitate multimodal integration to enhance prognostic precision and advance personalized therapy in DLBCL. • Deep learning automates baseline PET biomarker extraction in 1,024 patients with DLBCL • A clinical-imaging-multi-omics prognostic index is developed and cross-validated • High TMTV and LDH correlate immunosuppressive microenvironment and tumor proliferation Wang et al. develop a clinical-imaging-multi-omics prognostic index for diffuse large B cell lymphoma by integrating quantitative PET/CT biomarkers, genetic subtypes, and clinical risk factors. By analyzing large cohorts of patients, key biomarkers—total metabolic tumor volume and lactate dehydrogenase—that reflect immunosuppressive microenvironments and tumor proliferation were identified.