Researchers developed machine learning models using routinely collected health data to predict individual risk for 20 common cancers, outperforming traditional methods for several cancer types.
Key Details
- 1Study used health data—including blood tests, medical history, and lifestyle—from 463,620 UK Biobank participants followed for 18 years.
- 2About 500 machine learning models were developed for 20 common cancers; 57,055 participants developed cancer.
- 3Performance was highest for lung, oesophageal, and liver cancers, with AUROC of 0.83 for combined cancer outcome at 15 years.
- 4Predictive value came mostly from routine blood tests and medical history, with genetic and metabolomic data adding little improvement.
- 5Models were tested and performed well across populations in the UK, US, and Japan, though predictive performance varied by cancer type.
- 6A software toolbox (PRE-Screen) enabled large language models to apply the risk models, outperforming unaided LLMs.
Why It Matters
This approach could provide scalable, low-cost cancer pre-screening by leveraging existing clinical data, potentially influencing screening protocols and early detection. For radiology and imaging AI professionals, such tools may change referral patterns, resource allocation, and integration with imaging diagnostics in cancer pathways.
Source
EurekAlert
Related News
•EurekAlert
AI Deep Learning Model Enhances Echocardiogram RV Assessment and Prognosis
Researchers developed DROID-RV, a deep learning tool that automates right ventricular assessment from echocardiograms, showing strong prognostic links with cardiovascular outcomes.
•EurekAlert
Duke Receives $25M to Lead National Trustworthy AI Hub for Dementia and Aging
Duke University has launched a $25 million national center to make AI for aging and dementia clinically robust and trustworthy.
•EurekAlert
Penn Medicine's Percival AI Detects Hidden Health Risks in CT Scans
Penn Medicine's Percival AI leverages CT scans to uncover unseen health risks and disease patterns, expanding the utility of radiological imaging.