An artificial intelligence coronary artery surrogate strongly correlated with manual dose metrics in simulated lung stereotactic body radiotherapy.
98.98% of virtual contrast-enhanced MRI scans were rated suitable for diagnosis, and 92.33% for tumor delineation.
The neural model reduced phase-space storage from 3 gigabytes to 600 kilobytes while maintaining at least 89.9% gamma passing at 1%/1 mm.
The framework achieved 85.5% external-validation accuracy and reduced estimated review time from 822 hours to 2.3–4.8 hours.
Computed tomography-derived muscle measures ranked highly for toxicity and quality of life, but added little predictive value beyond established clinical factors.
A hybrid foundation-model framework achieved 94% lesion-wise sensitivity and was preferred over physician-generated brain metastasis contours in blinded, bias-adjusted comparisons.