Abstract Introduction: Pancreatic cancer (PC) is a highly lethal malignancy, largely due to its propensity for early dissemination and protracted subclinical course until it reaches advanced stages. Available diagnostic modalities are often unable to diagnose early events in the metastatic process, including pre-metastatic niches, micrometastases, or small macrometastases. We have previously shown that quantitative image features (QIFs) of the liver in pre-operative contrast-enhanced CT (CECT) scans can identify PC patients at risk of impending liver metastasis (PMID 35366706). Further, we found that perioperative liver biopsy in patients with untreated localized PC can capture cellular and molecular features of evolving pre-metastatic niches, which can predict timing and patterns of PC recurrence (doi.org/10.1158/1538-7445.PANCA22-PR012). In the present study, we sought to examine whether liver QIFs can capture specific biologic processes at the cellular level, which could in turn inform non-invasively on the status of the liver milieu that determine metastatic risk and potentially reveal opportunities for therapeutic intervention. Methods: Intraoperative liver biopsies were obtained from patients undergoing pancreatectomy for resectable PC (n=29) or non-cancerous pancreatic lesions (n=7) and were analyzed histologically for fibrosis, steatosis, portal and lobular inflammation. Immunostaining for immune cell populations and extracellular matrix proteins were performed. Liver biopsy histopathological features were then compared to 255 QIFs. Univariate analysis (UVA) was performed with logistic/linear regression, after stratification for biliary obstruction and presence of PC. For continuous variables, multivariate analysis (MVA) was performed using generalized linear model. For categorical variables, MVA was performed with Naïve-bayes classifier and ‘Leave-one image out’ method of cross validation after selecting the features with MRMR algorithm. Results: On UVA, multiple features correlated significantly with liver fibrosis, steatosis, and portal inflammation, as well fibronectin and collagen IV immunostaining. Further, significant correlations were found with CD11b+ cell density (cells/mm2), IBA1+ area, CD8+ staining intensity, CD3+ and CD8+ lobular infiltration, and total Ki67+ cell density. No significant associations were found with lobular inflammation, CD3+ intensity, or CD68+ cell density. On MVA, several QIFs remained significant predictors of the aforementioned 11 histopathologic variables. The best prediction was achieved for liver fibrosis (QIFs: RLM9, LBP56, LBP118, LBP3, LBP9, and LBP102; AUC 0.78) and steatosis (FD1_47; AUC 0.79). Conclusion: CECT liver texture analysis may predict biological processes affecting differential contrast perfusion, tissue distribution and parenchymal retention, enabling determination of liver histopathology non-invasively. Therefore, this approach may hold promise in identifying pre-metastatic niches or micrometastatic disease in evolution and risk-stratifying PC patients in a non-invasive fashion. Citation Format: Constantinos P. Zambririnis, Linda Bojmar, Jonathan Hernandez, Gokce Askan, Jian Zheng, Ahmad B. Barekzai, Natally Horvat, Vinod P. Balachandran, Jeffrey A. Drebin, Peter Kingham, Michael I. D’Angelica, Olca Basturk, Mithat Gönen, Alice Wei, Robert Schwartz, David C. Lyden, William R. Jarnagin, Jayasree Chakraborty. Radiomics of pre-operative CT scans capture biologic processes within the liver in patients undergoing pancreatectomy [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Pancreatic Cancer; 2023 Sep 27-30; Boston, Massachusetts. Philadelphia (PA): AACR; Cancer Res 2024;84(2 Suppl):Abstract nr C071.