Delayed diagnosis and treatment resistance make pancreatic ductal adenocarcinoma (PDAC) mortality rates high. Identifying molecular subtypes can improve treatment, but current methods are costly and time-consuming. In this study, deep learning models were utilized to identify histological features that classify PDAC molecular subtypes based on routine hematoxylin-eosin (H&E)-stained histopathological slides. 97 histopathology slides associated with resectable PDAC from the Cancer Genome Atlas (TCGA) project were utilized to train a deep learning model and tested the performance on 44 needle biopsy material (110 slides) from a local annotated patient cohort. The model achieved balanced accuracy of 96.19% and 83.03% in identifying the classical and basal subtypes of PDAC in the TCGA and the local cohort, respectively. This study provides a promising method to cost-effectively and rapidly classifying PDAC molecular subtypes based on routine H&E slides, potentially leading to more effective clinical management of this disease.

Background: RNA‐sequencing‐based classifiers can stratify pancreatic ductal adenocarcinoma (PDAC) into prognostically significant subgroups but are not practical for use in clinical workflows. Here, we assess whether histomorphological features may be used as surrogate markers for predicting molecular subgroup and overall survival in PDAC.Methods: Ninety‐six tissue samples from 50 patients with non‐resectable PDAC were scored for gland formation, stromal maturity, mucin, necrosis, and neutrophil infiltration. Prognostic PDAC gene expression classifiers were run on all tumors using whole transcriptome sequencing data from the POG trial (NCT02155621). Findings were validated using digital TCGA slides (n = 50). Survival analysis used multivariate Cox proportional‐hazards tests and log‐rank tests.Results: The combination of low gland formation and low neutrophil infiltration was significantly associated with the poor prognosis PDAC molecular subgroup (basal‐like or squamous) and was an independent predictor of shorter overall survival, in both frozen section (n = 47) and formalin‐fixed paraffin‐embedded (n = 49) tissue samples from POG patients, and in the TCGA samples. This finding held true in the subgroup analysis of primary (n = 17) and metastatic samples (n = 79). The combination of high gland formation and high neutrophils had low sensitivity but high specificity for favorable prognosis subgroups.Conclusions: The assessment of gland formation and neutrophil infiltration on routine histological sections can aid in prognostication and allow inferences to be made about molecular subtype, which may help guide patient management decisions and contribute to our understanding of heterogeneity in treatment response.