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…
ML-assisted label-free SERS achieved 100% accuracy in discriminating senescent vs normal ovarian granulosa cells at the single-cell level, detecting DNA damage, protein changes, and lipid peroxidation in mitochondria. (No external validation reported.)
Multi-cohort brain MRI: a simple 3D CNN (SFCN) achieved AUC 1.00 for sex, MAE 2.66y for age (internal); external AUC 0.85-0.91, MAE 4.98-5.81y. Simpler architectures generalized better than Swin Transformer.
CT radiomics SVM model predicted 5-year recurrence after BCG for high-grade NMIBC (AUC 0.82, 95%CI 0.77-0.86) in internal validation, with external validation confirming. Retrospective.
Deep learning chemical shift correction (DLCSC) reconstruction of zero echo time (ZTE) MRI significantly improved sharpness, SNR, and bone depiction quality in lumbar spine imaging compared to standard DL and non-DL reconstruction in a retrospective study of 38 patients.
A narrative review finds AI can cut report turnaround time (e.g., from 11.2 to 6.9 min with a templated EHR-PACS interface, 38% reduction), but links AI use to emotional exhaustion and notes outdated reimbursement limits financial benefits for radiologists.
In patients with large body habitus (mean 117 kg), DLIR-H improved hypodense liver and kidney lesion conspicuity vs ASIR-V (renal score 4.25 vs 3.75, p<0.05) and CNR (liver 8.62 vs 4.98), in a single-center retrospective study.
Super-resolution DL reconstruction improved brain FLAIR quality (noise, SNR, sharpness; p<0.001) in 36 volunteers, preserving white matter hyperintensity volumes matching high-res standard, without distorting lesion measures.
Markedly sharpened 60%- and 40%-dose cephalometric radiographs showed significantly higher reference-point visibility than standard 100%-dose images in a phantom study, suggesting post-acquisition sharpening may enable dose reduction.
Meta-analysis of 10 studies: AI for proximal caries detection on radiographs had pooled sensitivity 76% (95% CI 70–80%), specificity 94% (90–96%), AUC 0.90. Bitewing outperformed panoramic. High heterogeneity and limited external validation remain barriers to clinical use.
Trend reversals in cognitive test scores appeared first among high-income men, with lower-income groups later peaks and smaller declines, in 579,379 Norwegian conscripts. Educational expansion temporarily offset a broader decline, but after plateau, scores fell across all groups.
A review in JNCI examines AI models for breast cancer risk prediction, focusing on calibration, performance variability, and the steps needed before routine clinical use (exact figures not reported in source).
Random forest model using dual-phase 99mTc-MIBI SPECT/CT semiquantitative parameters achieved AUC 0.979 (95% CI 0.951-0.995) for differentiating malignant from benign pulmonary nodules in a retrospective study of 132 patients.
Older age and larger tumor size independently predicted malignancy in 104 women with BI-RADS 4/5 breast lesions (aOR 1.13/yr, 1.10/mm). BI-RADS 5 was not significant after adjustment, but the small subgroup (n=11) limited power.
Prospective, single-center study of 240 mammograms found AI-assisted interpretation impacted reader performance metrics across training levels (general radiologist to technologist).
[18F]FDG PET is shifting from regional interpretation to network-based and computational analyses for individualized brain dysfunction assessment. Dynamic and total-body PET add temporal physiology and brain-body interaction data. Clinical impact hinges on developing standardize…
In a retrospective study of 254 patients, an integrated model combining intratumoral and peritumoral delta radiomics from multiparametric MRI with clinical data outperformed radiomics or clinical models alone for predicting pathologic complete response to neoadjuvant breast canc…
Deep learning tools improved dental students' diagnostic accuracy (sensitivity, F1) in most studies, but evidence for reasoning or long-term retention is limited. One study found LLM assistance lowered test scores. Systematic review of 16 studies. Adjunctive role only.
New Korean tinnitus guideline: imaging for pulsatile tinnitus or asymmetric hearing loss (Grade B); CBT and TRT conditionally recommended; zinc/vitamins strongly advised against (Grade D†); Ginkgo biloba may be considered (Grade B). Most evidence low certainty.
Cryoablation of endophytic renal tumors caused an early 7.1 mL/min/1.73m² eGFR drop vs 2.4 for exophytic (p=0.003). CKD stage progression was higher at 1 year (p=0.01), but 3-year decline and CKD progression were comparable between groups.
Deterministic tract planning for CT-guided lung biopsy produced skin entry points within a median of 15.2mm of operator-chosen sites (κ=0.53, 95%CI 0.43-0.63) in a retrospective 199-patient study. Supports use for planning review and training.
In a randomized trial, the Instylla Embrace Hydrogel Embolic System (HES) achieved 99% freedom from major adverse events and 88.6% technical success vs. 77.9% for standard embolics (p=0.001) in hypervascular tumors.
A nomogram combining DBT-based intralesional and perilesional radiomics with patient age and BI-RADS 4 subclassification achieved an AUC of 0.942 in classifying benign vs malignant breast masses in the validation set, with SHAP analysis showing feature contributions.
In a feasibility study, medical students using a handheld surface scanner created digital 3D models of dissected donor anatomy with a mean structure-presence score of 0.892 (most expected structures identifiable). Lower accuracy for brachial plexus (score 0.773).
A YOLO-based AI model detected lumbar disc herniation with an overall precision of 0.74 and recall of 0.70 in a small single-center retrospective study. The model remains hypothesis-generating and is not ready for clinical use due to the tiny test set and lack of external valida…
In a retrospective study of 255 early Alzheimer’s patients, brain volumetry software-based models predicted MMSE score with r=0.56, but domain-specific cognitive scores were modestly correlated (r=0.21-0.40) and visuospatial function not significantly predicted.
Multimodal MRI/US fusion model achieved AUC 0.877 for classifying HER2-low vs HER2-zero in TNBC; hybrid model adding postop clinical variable reached AUC 0.884 in this small retrospective development study (n=122).
COVID-19 diagnosis increased in-hospital mortality after ischemic stroke 2.74-fold (95% CI 2.42–3.11), while broad pandemic timing showed no independent association; machine-learning models had modest prediction (AUC 0.62–0.64).
Hypoplastic/absent anterior cerebral artery A1 segment independently predicted larger basal ganglia infarct volume (β=5.71 cm3, 95% CI 2.86–8.57) and increased hemorrhagic transformation (aOR=3.06, 1.28–7.29) after thrombectomy for MCA occlusion. Retrospective cohort.
A chest radiograph deep learning model detected preserved ratio impaired spirometry (PRISm) with AUROC 0.892 (95% CI 0.875-0.906), sensitivity 71.6%, and specificity 87.9% in a retrospective internal test set (n=10,917 paired exams).
A mixture of ophthalmic foundation models enables mitigation of altitude-associated domain shift.