Radiology

Netherlands survey: strong public support for pre-procedural disclosure of diagnostic error risk

A population-based Dutch survey (n = 1524) found 88.4% agreed or strongly agreed patients should be informed about diagnostic error risk before radiologic exams. Agreement was highest for younger patients, with an OR 4.023 for a 9-year-old versus a 71-year-old (p = 0.010). Specifying very low risks (<0.1% or <1%) reduced willingness to be informed, whereas a <5% probability did not. These data reflect public preferences in the Netherlands and may inform consent communication policies locally. 1

Multimodal deep learning predicts radiation dermatitis after hypofractionated breast RT

In a single-center cohort of 160 early-stage breast cancer patients, a DS-LightGBM model predicted multiclass radiation dermatitis with test accuracy 0.87 and ROC-AUC 0.95. Binary Grade ≥2 dermatitis occurred in 40% of patients and the model achieved binary accuracy 0.89 with ROC-AUC ~0.88–0.90. High-dose dosiomic subvolumes and deep-learning features were dominant predictors by SHAP analysis. Authors note fibrosis modeling was exploratory due to few events and external validation is required before clinical deployment. 2

PROBASE substudy: expert reference MRI reading and high-quality scans improve csPCa detection

Among 516 men in PSA-based screening, expert reference reads raised NPV to 92% versus 86% for local reads and increased sensitivity from 69% to 83%. Reference reading reduced missed csPCa in PI-RADS 1–2 (2.9% vs 17.1%) and lowered false positives in PI-RADS 4–5. Lower MRI quality (PI-QUAL 1–3) correlated with reduced csPCa detection (46% vs 62%) and lower true negative rate (84% vs 95%). The findings support standardized acquisition, QA, and expert secondary reading for MRI-based prostate screening programs. 3