ASFNR clinical state-of-practice: For posterior circulation large vessel occlusion stroke, use noncontrast CT to exclude hemorrhage, CT angiography to identify occlusion, and MRI with DWI as reference standard. Imaging scores like PC-ASPECTS help select patients for thrombectomy.
The ACR Appropriateness Criteria 2026 update provides imaging guidelines for suspected or confirmed diffuse lung disease, covering initial evaluation, acute exacerbation, and surveillance.
New expert consensus framework for symptomatic B3 breast lesions: assess concordance, lower excision threshold for discordant/palpable lesions, conservative management for concordant non-atypical lesions, and risk-based surveillance for atypia.
ACR Appropriateness Criteria for indolent lung cancer: For patients with slow-growing subsolid nodules or atypical cysts where surveillance is chosen over intervention, imaging follow-up aims to detect changes that alter management, with modality selection guided by accuracy and…
A meta-analysis of 14 studies (n=34,322) shows ML models predict extubation success with a pooled AUC of 0.90 (95% CI:0.82-0.95). Only 2 studies reported external validation, and high heterogeneity limits routine clinical use.
A correction has been published for the study 'Deep learning-based motion correction: cardiac motion artifact and image quality improvements on chest CT.' No new findings; the notice addresses errors in the original article.
A review outlines an AI framework spanning breast imaging, digital pathology, multi-omics, and treatment decisions, but cautions that bias, limited datasets, and poor interpretability impede clinical adoption.
A fine-tuned RadImageNet ResNet-50 classified adult scoliosis as none, mild, or severe on AP full-body X-rays with 76.2% accuracy (95% CI 69.2–82.1%) and AUROCs 0.77–0.85. Misclassification mainly occurred between adjacent severity levels; the model does not measure Cobb angle d…
A CNN for breast density assessment in postmenopausal women with Gilbert's Syndrome showed only 55.6% sensitivity for dense breasts, though specificity reached 100.0%. Findings are preliminary due to a very small sample (n=21) and warrant cautious interpretation in this specific…
AI tools could narrow the worldwide radiotherapy gap through workflow automation, treatment planning, and quality assurance, per a JCO Global Oncology study. Quantitative impact is not yet quantified.
Retrospective study: CLIPS classification predicts cavernous sinus invasion in pituitary neuroendocrine tumors with sensitivity 90.8%, specificity 98.4%, and better interobserver agreement than Knosp.
Percutaneous cholangioscopic laser lithotripsy safely removed an intraductal suture acting as a nidus for recurrent intrahepatic cholelithiasis. The minimally invasive approach avoided surgical revision. This was a single case report.
In a retrospective study of 30 patients with refractory locally advanced pancreatic cancer, irreversible electroporation (IRE) achieved median overall survival 11.4 months (IQR 6-28) and 12.1% serious adverse events. Subsequent chemotherapy extended overall survival to 20.4 mont…
In stage IB3/IIA2 cervical cancer, TACE pre-treatment achieved 77.9% resectability and significantly improved 5-year survival (69.7% vs 41.0% with surgery alone, P<0.05). No survival benefit was seen when adding TACE before chemoradiation for stage IIB/IIIA disease.
A letter in European Radiology outlines challenges, priorities, and strategies for implementing AI in medical imaging in LMICs, emphasizing infrastructure gaps, workforce training, and data governance.
Authors reply to a letter discussing challenges and strategies for implementing artificial intelligence in medical imaging in low- and middle-income countries (LMICs).
A new study shows that in radiomics, undertraining acts as a performance floor rather than a ceiling when statistical and validation failures combine, emphasizing the need for robust model evaluation. (Exact figures not reported in source.)
Radiomics model validation is in crisis: the field’s overemphasis on external validation sets overlooks undertraining, which is equally critical. A balanced approach to both is needed.
In a retrospective study of 33 patients with systemic sclerosis ILD, automated HRCT analysis identified early ground-glass opacity increases that visual scoring missed. These early quantitative changes predicted subsequent pulmonary function decline (p=0.04), with a >50 mL GGO i…
MULLET AI improved radiologists' AUC for focal liver lesion detection from 0.82 to 0.93 (p<0.0001), sensitivity from 74% to 90%, and cut reading time by 22% in a 375-patient, 10-radiologist multicenter trial.
A deep learning model integrating abdominal radiographs and early clinical text predicted hospital admission with an AUC of 0.86 after data augmentation, outperforming single-modality approaches. Retrospective study; no external validation reported.
A deep learning model (ResNet-50) predicted objective response to TACE+TKI therapy in HBV-related unresectable HCC with AUC 0.85 (95% CI 0.75-0.95), outperforming clinical (0.59) and radiomics (0.71) models, and significantly stratified progression-free survival (P=0.011). Retro…
A physics-inspired scatter removal framework improved contrast-to-noise ratio to 2.94 and achieved lower image quality scores (NIQE 4.67, BRISQUE 22.37) on 60 weld radiographs, outperforming conventional enhancement methods.
ML model using spontaneous brain activity from fMRI achieved AUC 0.894 for distinguishing bipolar vs. unipolar depression. Adding disease duration in a nomogram raised C-index to 0.926. Preliminary, single-center study requiring external validation.
GeneLLM, a Transformer that analyzes cell-free RNA nucleotide sequences directly, detected multiple cancers with ROC-AUCs of 0.925–0.996 in a multi-centre cohort—without relying on gene annotations, potentially enabling lower-depth, more scalable liquid biopsy screening.
A commentary in the Journal of the Formosan Medical Association addresses the recent study on deep learning for detecting peri-implant marginal bone loss on periapical radiographs.
A combined model integrating 3D volumetric segmentation and an LLM-based classifier achieved AUROC 0.900 (95% CI 0.882-0.918) for csPCa on mpMRI, significantly outperforming single-modality metrics in multi-institutional external validation.
In a retrospective multicenter study of 1621 patients aged 18-49 undergoing clinically indicated CCTA with AI plaque analysis, 62.8% had coronary plaque; prevalence rose from 34.7% (18-29y) to 68.8% (40-49y). Plaque burden and FFR-CT abnormalities also increased with age.
SVM integrating weekly CBCT delta-radiomic kinetics predicted late xerostomia at week 4 with AUC 0.81 (p=0.002), sensitivity 100%, specificity 86% (retrospective, n=131).
A nomogram integrating infarct location, radiomics, and clinical factors predicted 90-day unfavorable outcome after acute ischemic stroke with AUC 0.894 in training and 0.874 in external validation. Infarct location added value beyond radiomics alone (0.793 vs 0.767, p<0.01).
In a retrospective 40-patient trial, higher baseline tumor burden on craniospinal MRI trends toward predicting progression (OR 1.18, p=0.08) and volume ≥5.5 mL tied to shorter overall survival (p=0.003).
Generative AI improves diagnostic accuracy and workflow efficiency but risks automation bias (reported in 10/29 studies) and clinician deskilling (9/29), urging balanced integration with human oversight and explainable AI.
In a multicenter study of 603 patients, a radiomics model using super-resolution T2-weighted MRI did not outperform standard-resolution images for predicting placenta accreta spectrum disorders. The 4× model achieved an AUC of 0.874 in training, but no statistically significant…
FDA grants 510(k) clearance to Provect.AI's 3D-Anywhere for radiology image processing.