Deep Learning of Histopathology Predicts Outcomes After Surgery for Pancreatic Cancer

Area of Science:

• Computational pathology
• Digital pathology
• Oncology

Background:

• Pancreatic cancer recurrence after surgery poses challenges for treatment planning.
• Accurate prediction of recurrence is crucial for guiding adjuvant therapies and patient follow-up.
• Histopathological features in whole-slide images (WSI) hold potential prognostic information.

Purpose of the Study:

• To develop and validate a deep learning model for predicting pancreatic cancer recurrence risk using digitized histopathology WSI.
• To correlate model-derived risk classifications with known histopathologic and genomic features.
• To assess the model's prognostic performance in external validation cohorts.

Main Methods:

• A deep learning model was developed using a pan-cancer foundation model for embeddings and a fully connected neural network.
• The model was trained on WSI from pancreatic ductal adenocarcinoma resections across multiple cohorts.
• External validation was performed using a meta-analysis of three independent cohorts, including the PRODIGE 24 trial.

Main Results:

• The deep learning model identified high-risk classifications associated with specific histopathologic features (e.g., squamous morphology, necrosis) and gene expression profiles.
• High-risk pancreatic cancers showed enrichment for basal-like gene expression and distinct oncogenic pathways.
• External validation demonstrated that high-risk classifications significantly correlated with increased risk of death (HR=1.49) and recurrence or death (HR=1.41).

Conclusions:

• An open-source deep learning model effectively classifies pancreatic cancer recurrence risk from WSI.
• The model's classifications correlate with histopathologic and genomic characteristics, providing prognostic value.
• This WSI-based deep learning tool can offer individualized prognostic information to aid clinical decision-making in pancreatic cancer management.