30-Second Takeaway
- CT-based models and new MRI/CT techniques are refining oncologic staging and characterization in lung, rectal, hepatic, gastric, and breast disease.
- AI for diagnostic mammography and intracranial hemorrhage triage can match or exceed radiologists but has indication- and lesion-specific blind spots.
- Non-contrast CT–based net water uptake and MRI T1ρ mapping offer practical quantitative biomarkers for stroke edema and liver fibrosis.
- New contrast-enhanced signs and interpretable AI models can upgrade nodal and T staging beyond size-based or subjective assessment.
- Economic evidence for radiology AI remains entirely model-based, limiting claims about real-world value despite growing clinical deployment.
Week ending December 27, 2025
Actionable imaging advances and AI tools across thoracic, breast, abdominal, neuro, and GI radiology
CT-based ternary model sharply stratifies invasiveness of pure ground-glass nodules
In 1683 patients with 2125 nonsolid lung nodules, a CT-based ternary model differentiated preinvasive, MIA, and IAC with a C-index of 0.92. Key predictors of higher invasiveness included larger average diameter, higher mean attenuation, heterogeneous density, spiculation, lobulation, and pleural retraction. Vascular and internal features also mattered: increasing intranodular vessels, bubble-like lucencies, and air bronchograms all independently raised invasiveness odds. Adding mean attenuation and morphology clearly outperformed diameter alone, supporting structured assessment of pure GGNs beyond simple size thresholds.
Optimized AI-CAD boosts specificity and PPV in diagnostic mammography
Among 1534 diagnostic mammograms, AI-CAD at a 50% abnormality threshold achieved higher specificity, accuracy, and PPV than radiologists while maintaining similar AUC. At this threshold, specificity reached 95.0% vs 86.2%, with PPV 85.1% vs 69.5%, reducing false positives and potentially unnecessary biopsies. In symptomatic women, AI-CAD AUC exceeded radiologists, suggesting particular value when a lesion is clinically suspected. For BI-RADS 3 follow-up and imaging-detected abnormalities, higher specificity came with lower sensitivity, underscoring the need for indication-specific threshold tuning.
Hepatic T1ρ mapping outperforms T1 and ECV for fibrosis in chronic liver disease
In 112 chronic liver disease patients, hepatic T1ρ values were significantly higher in those with MR-elastography–defined significant fibrosis. T1ρ correlated more strongly with liver stiffness than native T1 and ECV, including in patients with steatosis where other markers often degrade. For significant fibrosis, T1ρ achieved AUC 0.90 in the full cohort, surpassing native T1 and numerically outperforming ECV. Diagnostic performance remained robust in steatotic livers, positioning T1ρ as a promising fibrosis biomarker where MR elastography is unavailable or unreliable.
References
Numbered in order of appearance. Click any reference to view details.
Additional Reads
Optional additional studies from this edition.