AI-based tool for pancreatic cancer diagnostics
Researchers have successfully developed a deep learning model that uses histopathology images to classify pancreatic ductal adenocarcinoma (PDAC), the most common form of pancreatic cancer, into molecular subtypes. This method achieves high precision and provides a rapid, cost-effective alternative to current methods that rely on expensive molecular assays. new research in American Journal of PathologyPublished by Elsevier, it is expected to advance personalized treatment strategies and improve patient outcomes.
PDAC recently surpassed breast cancer to become the third leading cause of cancer death in Canada and the United States. If caught early, surgery can cure about one-fifth of PDAC cases. Despite surgical intervention for these patients, the five-year survival rate remains at 20%. Approximately 80% of patients already have metastatic disease at the time of diagnosis, and most of these patients die from the disease within a year.
The aggressive nature of PDAC poses significant challenges when sequencing technologies are used to determine patient care plans. The rapid clinical progression of the disease requires rapid action to identify eligible individuals for targeted therapies and inclusion in clinical trials. However, current turnaround times for molecular analysis range from 19 to 52 days after biopsy, which cannot meet these time-sensitive needs.
Co-principal investigator David Schaeffer, MD, of the Department of Pathology and Experimental Medicine at the University of British Columbia, Vancouver General Hospital, and the BC Pancreatic Centre, explained: “More and more potentially actionable subtypes are being discovered that can provide new treatment options for patients with pancreatic cancer. Provide individualized treatment. This is not always the case.
The study involved training a deep learning AI model on whole-slide pathology images using hematoxylin and eosin (H&E)-stained slides to identify molecular subtypes of PDAC (basal-like and classic subtypes). H&E staining is a cost-effective and widely used technique often used in pathology laboratories for diagnosis and prognosis with rapid turnaround time. The models were trained on 97 slides from The Cancer Genome Atlas (TCGA) and tested on 110 slides from 44 patients in a local cohort. The best-performing model achieved 96.19% accuracy in identifying classic and basic subtypes in the TCGA dataset and 83.03% accuracy in the local cohort, highlighting its robustness across different datasets. .
“The model’s sensitivity and specificity are 85% and 100% respectively, making this AI tool possible,” said Dr. Ali Bashashati, co-principal investigator of the School of Biomedical Engineering and Department of Pathology and Experimental Medicine at the University of British Columbia. The main achievement of the research is the ability of the artificial intelligence model to detect subtypes from biopsy images, making it a very useful tool that can be deployed at the time of diagnosis.
Dr. Bashashati concluded: “This artificial intelligence-based approach represents an exciting advance in pancreatic cancer diagnosis, allowing us to quickly and cost-effectively identify key molecular subtypes.”
2024-12-12 16:58:39