Deep-learning dose prediction generalized across six head and neck radiotherapy centres
A model trained at one centre closely predicted doses in 560 head and neck plans from six external institutions.
A model trained at one centre closely predicted doses in 560 head and neck plans from six external institutions.
A single knowledge-based model generated comparable ocular plans for 25 Gy once, 42 Gy in three fractions and 50 Gy in five.
Low Dose HyperSight protocols reduced exposure by 55%, while Slow and Large protocols increased imaging dose and accentuated secondary-risk estimates.
Two- to five-millimetre gastrointestinal margins usually limited delivered dose increases but frequently failed to encompass interfraction organ motion.
SFUD and robust IMPT maintained target coverage under uncertainty, while IMPT modestly reduced rectal and bladder dose.
A retrieval-augmented GPT-4.1 chatbot rapidly surfaced previous TrueBeam faults, but procedural, role-definition and safety errors remained.
An autoencoder identified anomalous IMRT plans with an area under the curve of 0.98 and explained which parameters drove each alert.
Radioactive carbon-11 beams enabled PET-guided range adjustment during irradiation, with biological outcomes matching insufficient, optimal and excessive beam penetration.
Brain metastasis contours showed generally high agreement, although lesion volume, oedema and magnetic resonance imaging field strength influenced variability.
A national end-to-end audit found acceptable spine stereotactic radiotherapy accuracy but systematic differences between dose-reporting conventions.
Thoracic AI contours showed strong geometric and dosimetric agreement, but physicians still rejected some heart and oesophageal contours.
Deep-learning reconstruction produced pancreatic gated cone-beam computed tomography comparable with conventional imaging while reducing estimated acquisition time and imaging dose.
A 4.5 mm isotropic margin covered most observed small-bowel occupancy despite craniocaudal displacements exceeding 20 mm in some patients.
Planning CT features added prognostic information for local and distant control but did not significantly improve overall or progression-free survival prediction.
Upright carbon ion treatment plans achieved target coverage, organ sparing and simulated robustness comparable with conventional supine plans.