2290 - Integrative Radiogenomic Analysis Identifies Imaging-Linked Molecular Subtypes and Drivers of Heterogeneity in Pancreatic Adenocarcinoma

03:00pm - 04:00pm ET

Poster Hall - Exhibit Hall A

Screen: 33

POSTER

Presenter(s)

Dandan Zheng, PhD - University of Rochester, Rochester, NY

D. Zheng¹, Z. Qu¹, M. Baine², P. Grandgenett³, M. A. Hollingsworth³, and C. Zhang⁴; ¹University of Rochester, Rochester, NY, ²Department of Radiation Oncology, University of Nebraska Medical Center, Omaha, NE, ³University of Nebraska Medical Center, Omaha, NE, ⁴University of Nebraska Lincoln, Lincoln, NE

Purpose/Objective(s): Pancreatic ductal adenocarcinoma (PDAC) is characterized by profound molecular and clinical heterogeneity, contributing to its poor 5-year survival rate (<12%). While molecular subtypes influence prognosis, their relationship with clinical imaging remains poorly defined. We developed an integrative radiogenomic framework to identify imaging-linked molecular signatures that may serve as non-invasive biomarkers for treatment stratification.

Materials/Methods: We utilized a multi-omics cohort of 15 patients from the University of Nebraska Medical Center Pancreatic Rapid Autopsy Program (RAPID), featuring matched primary tumor RNA sequencing, somatic single-nucleotide variant (SNV) profiles, and longitudinal contrast-enhanced CT imaging. First, unsupervised clustering was performed on TCGA PDAC transcriptomic data to identify molecular subtypes. A random forest classifier was then trained on these patterns and applied to the RAPID cohort. Radiomic features were extracted from CT scans and integrated with transcriptomic/genomic data via a mediation network model to link somatic variants and gene expression to imaging phenotypes and overall survival (OS).

Results: The classifier successfully identified five transcriptional subtypes within the RAPID cohort, enriched in pathways involving inflammation, extracellular matrix remodeling, and tumor microenvironment (TME) organization. These subtypes exhibited distinct survival trends. Integration of transcriptomic profiles with CT-derived radiomics identified cross-modal signatures of tumor heterogeneity. The mediation network analysis revealed that specific somatic variants drive subtype-specific expression patterns, which in turn manifest as distinct radiomic phenotypes. These radiogenomic signatures were associated with OS, providing a mechanistic link between genomic instability and macro-scale imaging traits.

Conclusion: This study establishes a multi-omics framework for PDAC biomarker discovery, demonstrating that CT-based radiomics can reflect underlying molecular and TME heterogeneity. These findings suggest that non-invasive radiomic profiling could potentially augment biopsy-based subtyping to personalize treatment strategies and improve prognostic accuracy in PDAC.