Because the malignancy is relatively uncommon, it has received too little attention. Mayo Clinic researchers are doing their part to highlight its devastating effects and are using the latest deep learning models to detect it early on.
By John Halamka, M.D., M.S., Dwight and Dian Diercks President, Mayo Clinic Platform and Paul Cerrato, MA, senior research analyst and communications specialist, Mayo Clinic Platform
How do you detect a progressive disease that doesn’t announce itself until it’s almost impossible to stop? According to the National Cancer Institute, pancreatic cancer will likely take the lives of over 52,000 patients, with more than 67,000 new cases expected in 2026 and a 5-year survival of only 13.7%. Detecting the disease early, before it has advanced, offers the best hope of a cure.
Cornelius A. Thiels, D, MBA, with Mayo Clinic’s department of surgery, and his colleagues have been using the massive, de-identified data set from Mayo Clinic Platform to help identify those at high-risk of developing pancreatic cancer with the help of a transformer model they designed. The team obtained data from patient encounters that date back up to 20 years before patients were diagnosed with pancreatic cancer. They looked at the various diagnoses that everyone acquires across their lifetime, alongside the rich data within routine lab work, including platelet count, MCV, blood urea nitrogen, hematocrit, chloride, and red blood cell count. Feeding this data into an AI transformer model, they found subtle changes in patients’ medical histories and lab values over time. Several of these ICD-Diagnostic codes helped predict the eventual onset of the malignancy, including EZ8.5 (disorders of lipoprotein metabolism), E11 (type 2 diabetes), and Z000.x (general exam). The transformer was able to separate patients into high and lower risk groups, predicting those most likely to develop the cancer up to three years in advance.
Mayo Clinic investigators have also developed the Radiomics-based Early Detection MODel (REDMOD) to help detect visually occult pancreatic cancer in patients from several healthcare facilities using standard-of-care CT scans. Among nearly 500 scans, REDMOD identified occult cancer with a sensitivity of 73% about 475 days before it appeared in patient records—almost twice the sensitivity achieved by radiologists (38.9%). The research paper describing the new model explains that REDMOD is: “a next-generation AI framework featuring a robust ensemble architecture (logistic regression, random forest, XGBoost) and fully automated volumetric pancreas segmentations.” It was validated using a low prevalence patient population using an adaptable research design and a modifiable diagnostic threshold. Mukherjee et al go on the explain: “For the first time, this work establishes the longitudinal stability of a pre-clinical radiomic signal (90–92% test–retest concordance) and confirms the robust specificity of the model across independent multi-institutional (81.3%) and public (87.5%) datasets.”
Of course, Mayo Clinic is only one of several organizations working to improve early detection of the malignancy. As we point out in an earlier column, European researchers have also successfully developed an AI model based on a dataset called PANORAMA to help detect the cancer. Their noninferiority observational investigation used data from the Netherlands, USA, Sweden, and Norway concluding that “AI demonstrated substantially improved PDAC [pancreatic ductal adenocarcinoma] detection on routine CT scans compared to radiologists on average, showing potential to detect cancer earlier and improve patient outcomes.” While the study has limitations that go beyond the scope of our column and did not directly demonstrate that its model can improve early detection, their model improved sensitivity and reduced the number of unnecessary referrals by generating up to 38% fewer false positives when compared to the performance of 68 radiologists.
Early detection programs for colon and breast cancer have had a major impact on the incidence of these diseases and saved countless lives. We look forward to the day when the same thing can be said about the number of patients who fall victim to pancreatic cancer. The research done by Mayo Clinic and others are moving us in that direction.
Share this:
