PDAC Subtyping Tool

A recent study has introduced a deep learning-based approach to identify molecular subtypes of pancreatic ductal adenocarcinoma (PDAC) from routine hematoxylin-eosin-stained slides, offering a cost-effective alternative to genomic sequencing. Published in The American Journal of Pathology, the research addresses the challenges of delayed diagnosis and high treatment resistance in PDAC, which is one of the leading causes of cancer mortality.

Using two patient cohorts, researchers trained their model on histopathology slides from The Cancer Genome Atlas project and validated it on a separate dataset of needle biopsy samples. The study employed four deep learning architectures. The Vanilla model showed the highest accuracy, achieving 83 percent on external biopsy data and 96 percent during internal validation.

The models classified PDAC into two molecular subtypes: classical and basal-like, based on the PurIST molecular subtyping algorithm. These subtypes are clinically significant, as basal-like tumors often show poor response to standard chemotherapy regimens like FOLFIRINOX. The Vanilla model achieved a positive predictive value of 1.0 for basal-like tumors, suggesting high reliability in identifying this critical subtype.

A key innovation was the use of a two-stage classification process. First, the model identified tumor regions within the slides, and then it predicted molecular subtypes based on these areas. The use of Grad-CAM activation maps enabled visualization of morphological features, such as gland-forming and non-gland-forming regions, which contributed to subtype differentiation.

This approach bypasses the need for molecular profiling, making subtype classification accessible in standard pathology labs. Corresponding author David Schaeffer said, “Our study provides a promising method to cost-effectively and rapidly classify PDAC molecular subtypes based on routine hematoxylin-eosin–stained slides, potentially leading to more effective clinical management of this disease.”