PocketNet matched nnU-Net segmentation with 40-fold lower model storage
PocketNet matched nnU-Net accuracy for cervical MRI segmentation while using 40-fold less storage and approximately half the training time.
Tag
Technical relevance
CyberKnife dose models found size-dependent but non-monotonic brain metastasis control
Tumour-size strata showed non-monotonic BED–control relationships, while machine-learning performance remained unvalidated outside a single retrospective centre.
Exit-dose gamma alerts identified head-and-neck replanning needs on Halcyon-E
Strict exit-dose gamma thresholds correlated with anatomical change and identified all clinically replanned patients in a retrospective validation cohort.
Reinforcement learning adapted proton spot delivery across respiratory phases
Patient-specific reinforcement learning improved moving-target coverage over static GTV plans while reducing some normal-tissue exposure compared with ITV planning.
Probabilistic coronary habitats enabled artery-focused cardiac sparing without angiography
Automatically predicted coronary habitats captured most arteries and enabled substantial dose reductions in three retrospectively replanned thoracic cases.
GPU dose engine calculated MRI-guided proton plans within 14 seconds
gPRIDE closely matched Monte Carlo calculations for most anatomies and completed MRI-guided proton dose calculations within 13.7 seconds.
First on-treatment CBCT best predicted lung radiotherapy response in a small cohort
Adding the first treatment CBCT improved response and survival prediction, while additional scans progressively reduced classification performance.
Surface guidance matched skin marks for breast setup and reduced positioning variability
Single-camera surface guidance provided comparable mean setup accuracy with less variability, while intrafraction breast motion remained below 3 mm in most observations.
Simple phantoms increased students’ reported understanding of radiation medicine
Low-cost diagnostic and treatment phantoms improved students’ self-reported understanding of radiation medicine, although formal learning outcomes were not assessed.
Acuros XB increased synthetic CT dose differences but remained clinically acceptable
Acuros XB was more sensitive than AAA to synthetic CT discrepancies, but target-dose differences remained small in prostate and glioma plans.
Hybrid radiomics model improved tomotherapy PSQA prediction but transferred poorly across centres
Combining plan complexity with three-dimensional dose radiomics improved tomotherapy PSQA prediction, although performance deteriorated substantially during cross-institution testing.
Biopsy localization uncertainty substantially altered HDR prostate brachytherapy dose estimates
Millimeter-scale localization uncertainty shifted assigned biopsy doses by several Gy and made nominal DVH threshold classification frequently unreliable.
Bragg peak proton FLASH matched CyberKnife plans and cut beam delivery to seconds
In 10 recurrent glioblastoma replans, Bragg peak proton FLASH improved conformity and reduced estimated beam delivery from approximately 17 minutes to under two seconds.
Additional PMMA changed mammography multimeter calibration coefficients by up to 42%
Spectral hardening altered multimeter calibration by up to 11% for kerma, 37% for HVL and 42% for voltage.
Early CBCT shifts failed to predict dosimetric compromise in head and neck VMAT
Early three-dimensional CBCT shifts performed no better than chance for identifying patients with at least 3% PTV coverage loss.