30-Second Takeaway
- Routine CT, MRI, and radiography can support opportunistic systemic risk assessment well beyond the original imaging indication.
- Quantitative MRI and PET techniques are refining lesion characterization and surgical or radiation planning in oncologic and neuro-oncologic care.
- Institution-tuned AI and secure LLMs can match radiologists for specific tasks but require careful scoping and workflow integration.
- Foundation models and radiomics from standard studies are yielding strong prognostic biomarkers for cardiovascular and oncologic outcomes.
- Radiologists will increasingly gatekeep how, when, and for whom these imaging-derived biomarkers and AI tools are deployed.
Week ending April 11, 2026
Imaging as a Predictive Biomarker: Opportunistic Risk Stratification, Quantitative MRI, and Practice-Ready AI Tools
Deep learning on ED head CT opportunistically stratifies cardiovascular risk
A deep learning model applied to 27,990 emergency department head CTs predicted incident cardiovascular events better than the AHA PREVENT risk model (C-index 0.82 vs 0.75). The model also estimated CAC categories in 2,313 patients, with good discrimination for CAC >100 when combined with PREVENT (AUC 0.80). Integration of CT features reclassified 15.7% of patients, preferentially upshifting younger individuals with brain infarcts and vascular calcifications. These findings suggest routine head CT can opportunistically flag high cardiovascular risk without additional imaging, cost, or radiation.
Epicardial fat radiomics from CCTA predicts incident heart failure
In 72,751 adults undergoing CCTA, an automated epicardial adipose tissue radiomic profile (FRPHF) strongly stratified future heart failure risk. Each 25-percentile increase in FRPHF was associated with nearly a fourfold higher adjusted HF risk in both internal and external cohorts. Patients in the highest FRPHF decile had nearly 20-fold higher HF risk than those in the lowest decile. Adding FRPHF to conventional models, including CAD severity and EAT volume, improved 5-year discrimination and net reclassification for HF events.
Pediatric chest radiograph AI estimates DXA-equivalent BMD and detects low bone density
In 1,464 pediatric chest radiograph–DXA pairs, an AI model predicted lumbar spine BMD Z scores with strong correlation to DXA (r up to 0.85). For detecting low BMD (Z ≤ -2.0), AUCs were 0.92 internally and 0.90 externally, with high specificity in both cohorts. External testing showed sensitivity 82% and specificity 85% for low BMD identification. These data support opportunistic bone health screening from routine pediatric chest radiographs where DXA access is limited.
Whole-body DWI/MRI after NACT predicts cytoreduction and survival in advanced ovarian cancer
In 105 women with initially unresectable ovarian cancer, post-NACT WB-DWI/MRI predicted complete interval debulking with 97.3% sensitivity and 83.9% specificity. MRI-based prediction of complete resection was independently associated with both progression-free and overall survival on multivariate analysis. Complete resection at surgery and serum CA-125 also impacted progression-free survival, but MRI prediction remained the strongest imaging predictor. In 69 patients with paired imaging, WB-DWI/MRI prediction of complete resection was significantly more accurate than CT (91.3% vs 72.5%).
References
Numbered in order of appearance. Click any reference to view details.
Additional Reads
Optional additional studies from this edition.