Researchers have successfully created a deep learning model that categorizes pancreatic ductal adenocarcinoma (PDAC), the most prevalent type of pancreatic cancer, into various molecular subtypes by analyzing histopathology images. This innovative method not only achieves high levels of accuracy but also serves as a quick and economical alternative to existing techniques that depend on costly molecular tests. The findings from this study, published in The American Journal of Pathology by Elsevier, have the potential to enhance personalized treatment approaches and significantly improve patient outcomes.
Researchers have successfully created a deep learning model that categorizes pancreatic ductal adenocarcinoma (PDAC), the most prevalent type of pancreatic cancer, into various molecular subtypes by analyzing histopathology images. This innovative method not only achieves high levels of accuracy but also serves as a quick and economical alternative to existing techniques that depend on costly molecular tests. The findings from this study, published in The American Journal of Pathology by Elsevier, have the potential to enhance personalized treatment approaches and significantly improve patient outcomes.
PDACs have recently overtaken breast cancer as the third leading cause of cancer-related deaths in both Canada and the United States. Early detection can lead to surgical intervention that has a chance of curing about 20% of PDAC cases. However, even with surgery, the five-year survival rate still rests at 20%. Unfortunately, around 80% of patients are diagnosed with metastatic disease, and the majority tend to pass away within a year.
The aggressive nature of PDAC makes it particularly challenging to utilize sequencing technologies effectively in developing patient care plans. The disease progresses quickly, necessitating rapid identification of patients who are suitable for targeted treatments and clinical trials. However, the current timelines required for molecular profiling, which can extend from 19 to 52 days post-biopsy, do not meet these urgent needs.
Dr. David Schaeffer, one of the lead researchers from the University of British Columbia’s Department of Pathology and Laboratory Medicine and Vancouver General Hospital, suggests, “There are increasingly more potential subtypes we can leverage to customize treatment for pancreatic cancer patients. However, current classification methods solely depend on genomic techniques using DNA and RNA sourced from tissue samples. While this approach is effective when adequate tissue is available, which is often not the case for PDAC due to its challenging anatomical position, our study introduces a promising way to accurately and affordably classify PDAC subtypes using standard hematoxylin-eosin-stained slides and could lead to improved clinical management.”
The research involved training deep learning AI models on complete slide images to determine PDAC’s molecular subtypes—specifically basal-like and classical—by utilizing hematoxylin and eosin (H&E) stained slides. H&E staining is a widely accessible and cost-effective method routinely used in pathology for diagnostics and prognostics, allowing for rapid results. The training involved 97 slides from The Cancer Genome Atlas (TCGA), with testing conducted on 110 slides collected from 44 patients within a local group. The model that performed best showed an accuracy rate of 96.19% for identifying classical and basal subtypes in the TCGA dataset and 83.03% in the local patient cohort, showcasing its reliability across diverse datasets.
Co-lead investigator Ali Bashashati, from the School of Biomedical Engineering and the Department of Pathology and Laboratory Medicine at the University of British Columbia, highlights, “The model demonstrated a sensitivity of 85% and specificity of 100%, making it an excellent tool for determining which patients should undergo further molecular testing. Importantly, this study’s key achievement is that the AI model successfully identified subtypes directly from biopsy images, presenting a highly useful resource that can be implemented at the time of diagnosis.”
Dr. Bashashati concludes, “This AI-driven technique marks a significant leap forward in the diagnosis of pancreatic cancer, allowing for rapid and cost-effective identification of critical molecular subtypes.”
