Radiology

Photon-counting CT spectral radiomics separates benign from malignant liver lesions with robust performance at higher energies

In 378 patients from two centers, PCD-CT–based spectral radiomics with machine learning differentiated benign from malignant liver lesions with good accuracy. A random forest model trained on virtual non-contrast images achieved AUC 0.899 and accuracy 83.5% for malignancy classification. Across reconstructions, test performance was most stable on 110–180 keV and VNC images, with median AUCs at or above 0.83. Low-energy images and iodine density maps yielded lower and more variable AUCs, underscoring reconstruction choice as critical for radiomics workflows. 1

AI reconstruction allows up to sixfold faster prostate MRI without significant loss in cancer detection

In this multicenter paired study, AI reconstruction enabled T2-weighted prostate MRI acceleration up to R=6 without statistically significant loss in prostate cancer detection. Eight experienced radiologists achieved AUROCs of 0.86, 0.82, and 0.80 for R=1, R=3, and R=6 scans, respectively. R=3 images were rated sharper and less noisy than standard scans, and overall visual quality at R=6 was comparable to R=1. T2 acquisition times dropped to 0:33–1:27 minutes, suggesting potential gains in throughput and patient comfort despite modest AUROC decline. 2

Pretreatment prostate MRI provides independent prognostic information after radical prostatectomy

This meta-analysis of 40 studies including 24,941 men showed pretreatment prostate MRI features independently predict outcomes after radical prostatectomy. MRI-defined extraprostatic extension (mrT3a) was associated with higher biochemical recurrence, metastatic failure, and prostate cancer-specific mortality. Seminal vesicle invasion (mrT3b) similarly predicted increased biochemical recurrence and metastatic failure risks. PI-RADS 4–5 category, large tumor size, and low ADC values were prognostic for biochemical recurrence beyond standard clinical factors. 3

Conditional transitional-phase washout improves LI-RADS LR-5 performance on gadoxetate MRI

In 438 patients with 558 hepatic observations, transitional-phase liver parenchymal signal substantially influenced LI-RADS washout behavior on gadoxetate MRI. Using portal venous phase–only washout yielded LR-5 sensitivity 70.7% and specificity 96.9% for hepatocellular carcinoma. Extending washout into the transitional phase only when liver parenchyma was iso- or hypointense increased sensitivity to 77.8% without reducing specificity. About one-quarter of patients showed hyperintense transitional-phase parenchyma, supporting a conditional rather than universal TP washout criterion. 4