Deep learning models analyzing chest x-rays can predict cardiovascular risk on par with the PREVENT clinical score.
Key Details
- 1Study used nearly 40,000 patient chest x-rays from two large datasets.
- 2Convolutional neural networks extracted coronary artery calcium (CAC) and radiomics features from chest x-rays.
- 3CXR-based models achieved AUROC up to 0.80 for detecting any CAC.
- 4Composite CXR radiomics and CXR-predicted CAC had concordances (0.65, 0.64) nearly matching PREVENT (0.66) for 10-year MACE prediction; combined models reached 0.71.
- 5CXR-based methods stratified risk among patients PREVENT classified as low-risk, with fourfold differences between high and low radiomics quartiles.
- 6CXR AI models require no lab results, offering potential for broader opportunistic screening.
Why It Matters
This study suggests chest x-ray-based AI could expand cardiovascular risk prediction to settings where lab data are unavailable, potentially enabling population-level screening for heart disease using existing imaging.
Source
AuntMinnie
Related News
•AuntMinnie
Deep Learning on PSMA-PET Outperforms Tools in Prostate Cancer Survival Prediction
Deep learning models using PSMA-PET scans predict survival in metastatic prostate cancer more accurately than established clinical tools.
•AuntMinnie
AI Boosts Sensitivity but Raises False Alarms in Microbleed Detection
AI significantly increases sensitivity for detecting cerebral microbleeds on brain MRI but at the cost of more false positives, highlighting the complementary strengths of AI and radiologists.
•AuntMinnie
AI Algorithm Enhances Detection of Incidental Intracranial Aneurysms on Neck CT
An AI algorithm helped detect previously missed intracranial aneurysms in routine contrast-enhanced neck CT scans.