Related Experiment Video
Updated: Jul 14, 2025
08:08
Detection and Isolation of Cancer in Prostate Biopsies Using Stimulated Raman Histology and Artificial Intelligence
Published on: June 10, 2025
69
Real-Time Identification of Pancreatic Cancer Cases Using Artificial Intelligence Developed on Danish Nationwide
Anders Bo Bojesen1, Frank Viborg Mortensen1,2, Jakob Kirkegård1,2
1Department of Surgery, HPB Section, Aarhus University Hospital, Aarhus, Denmark.
JCO Clinical Cancer Informatics
|October 9, 2023
Summary
A new dynamic algorithm can identify patients at high risk for undiagnosed pancreatic cancer. This machine learning approach accelerates diagnosis, potentially saving lives by improving early detection rates.
Area of Science:
- Oncology
- Medical Informatics
- Machine Learning
Background:
- Pancreatic cancer is a leading cause of cancer death, with late diagnosis limiting curative treatment options.
- Accelerating diagnosis is crucial for reducing pancreatic cancer mortality.
- A dynamic algorithm was developed to proactively identify high-risk individuals for undiagnosed pancreatic cancer.
Purpose of the Study:
- To develop and test a dynamic algorithm for proactive identification of patients at elevated risk for undiagnosed pancreatic cancer.
- To improve early detection rates and reduce pancreatic cancer mortality.
Main Methods:
- Machine learning applied to nationwide Danish registry data.
- Hybrid case-control and prospective cohort design with incidence density sampling.
- Performance evaluated using out-of-sample, out-of-time data in a monthly walk-forward strategy to prevent temporal bias.
Main Results:
- Identified subgroups with a 10.1% 1-year risk of pancreatic cancer diagnosis (number needed to screen: 9.9).
- Risk reduced to 5.7 when considering other gastrointestinal (GI) cancers.
- Accelerated diagnosis time by up to 142 days.
Conclusions:
- Nationwide live data and computational resources support real-time identification of high-risk individuals for pancreatic and GI cancers in Denmark.
- The area under the curve is not ideal for prospective high-risk patient identification.
- Demonstrated design solutions address limitations in existing cancer prediction efforts, with further efficacy evaluations needed.
