The 2026 Chinese lung cancer guideline adds AI and opportunistic screening in low-dose CT, expands molecular testing, and updates treatment regimens for NSCLC, including perioperative, consolidation immunotherapy, and targeted therapy for EGFR, ALK, KRAS, HER2, and MET alteratio…
On 76 test cases, the AI framework achieved Cobb angle MAE 2.4°, Lenke curve type accuracy 0.895, and reduced review time from 8.2 to 1.5 min/case.
An AI model (EAGLE) analyzing noncontrast chest CT achieved 90.0% sensitivity for esophageal cancer and 98.5% specificity in external multicenter testing (11,466 patients), and 99.94% specificity in low-dose screening. Could enable opportunistic esophageal cancer screening.
PSAD remained below 0.11 ng/mL² across all ages in 1037 healthy German men, contrasting with continuously rising PSA. 95th percentile reference: PV 51.39 mL, PSA 2.79 ng/mL, PSAD 0.071 ng/mL².
For breast MRI, diffusion-weighted imaging (DWI) boosts diagnostic AUC to 0.960 when combined with other sequences, and non-contrast DWI screening yields a 99.5% negative predictive value. Malignant lesions show restricted diffusion (low ADC).
A 3D vision transformer foundation model achieved high sex classification (AUC 0.979-0.988) and age estimation mean absolute error of ~8 years on femoral CT from 1,489 Japanese individuals, with attention maps highlighting distal femur for sex and shaft for age.
Letter to the Editor: Caution in interpreting early performance gains in AI detection of inflicted fractures in children.
Letter to the Editor: Clinical implications of automated breast ultrasound in preoperative breast cancer staging.
EUCAIM's FAIR hyperontology semantically integrates heterogeneous cancer imaging data, validated with prostate and breast cancer cases.
In a formative usability study of 46 clinicians, a sentence-level 'click to inspect' interface for AI-generated medical record summaries achieved a mean System Usability Scale score of 86.25 (95% CI 83.96–88.54) and a net promoter score of 35. Rapid access to supporting evidence…
A semisupervised deep learning framework classified postop recurrence and metastasis from CT reports with up to 99.33% accuracy (binary), matching clinician consistency, while explicitly modeling diagnostic uncertainty (1.4% recurrence, 6.9% metastasis).
For intermediate-likelihood choledocholithiasis, EUS generates 4.53 kg CO2e per procedure vs 18.90 kg for MRCP—a 4.2-fold difference. Choosing EUS for 108 patients saved 1,552 kg CO2e (74 trees' annual carbon uptake).
In 4214 UK Biobank adults, higher obesity (body roundness index) was associated with reduced renal corticomedullary T1 differentiation (ΔT1; β = -8.507, 95% CI -11.893 to -5.120), mainly from lower medullary T1, while age correlated with higher cortical T1.
Longitudinal dark-field CT in mice revealed microstructural lung changes earlier than attenuation, with distinct trajectories in COPD and injury models, while attenuation remained stable. May serve as early biomarker.
A deep learning model (nnU-Net) automatically segmented peripheral nerves with Dice 0.97 and lesions with surface Dice 0.64–0.77 on MR neurography; automated total lesion load correlated with nerve conduction velocity (r=-0.62, p<0.001).
ESMRMB 2026 focus topic: future of MR machines.
Higher basal ganglia perivascular space burden on MRI was associated with sustained amyloid-β accumulation on follow-up PET in 524 ADNI participants (β=8.350, P=0.001). White matter PVS showed no link. Findings suggest BG-EPVS as a biomarker for amyloid trajectory.
A stacking fusion model combining clinical, radiomics, habitat, and deep learning features from DCE-MRI achieved AUC 0.854 on external test for classifying luminal vs non-luminal breast cancer, with sensitivity 0.839. Deep learning contributed most (48.6% SHAP).
In 160 depressed adolescents vs 90 controls, gut microbiota dysbiosis and disrupted glycerophospholipid metabolism were linked to fMRI brain alterations in precuneus and precentral gyrus. A multi-omics biomarker panel strongly predicted depression.
Letter to the Editor: Dual imaging modality approach for acute stroke: a primary short brain MRI protocol for selected patients.
Artificial intelligence for thyroid nodule imaging has evolved from simple classifiers to comprehensive systems handling detection, segmentation, and risk stratification, now aligning with biopsy decision pathways and clinical workflows.
A multitask deep learning model classified hypertrophic cardiomyopathy vs hypertensive heart disease on cardiac MRI native T1 maps with AUC 0.941 (95% CI 0.903-0.979), outperforming a clinical baseline AUC 0.757; external 20-patient test AUC 0.885.
In 149 patients, an Attention U-Net using contrast-enhanced T1-weighted, T1-weighted, and FLAIR MRI detected meningiomas with detection rate 0.98 (SD 0.03) and Dice 0.82 (SD 0.04), compared with contrast-enhanced T1 alone at 0.79 and Dice 0.67.
A 2-month rounds-integrated point-of-care ultrasound (POCUS) curriculum for eight medical students at a Guatemalan public hospital was feasible; knowledge and confidence largely persisted at 2 months, with strongest hands-on skills in pelvic/bladder view acquisition and image op…
In 539 triple-negative breast cancers, HER2-low vs HER2-0 tumors showed different ultrasound vascularity (avascular: 6.3% vs 12.0%) and radiomics features; overall survival was similar, but HER2-low was associated with improved event-free survival in non-pCR patients.
A stochastic orthonormal projective NMF method analyzed 10,000 UK Biobank brain MRIs, identifying 20 structural covariance patterns; 13 in females and 11 in males showed significant links to visceral adipose tissue, mostly negative.
Longitudinal hippocampal and cognitive trajectories in 415 older adults revealed resilience, stability, vulnerability subgroups. Leading risk indicators: lower plasma Aβ1-42 for resilience, lower education for vulnerability.
At-rest primary motor cortex GABA was higher in mild-to-moderate Parkinson’s disease (32 patients) vs 28 controls on 7T MRS, but the difference did not survive FDR correction (pFDR=0.06). Inhibitory/excitatory ratio was also higher. Glutamate did not differ. Findings support 7T…
A new k-space-based AI model (LAPANet) accurately estimates cardiac MRI motion from just 2 Cartesian lines/frame or 3 radial spokes/frame, achieving sub-5 ms temporal resolution. This bypasses image reconstruction and avoids undersampling artifacts, enabling real-time motion tra…
A prostate cancer DLA tested on DLR-accelerated bpMRI showed similar per-patient performance. However, with C-T2WI+DLR-DWI, detected lesions fell (99 vs 131), FP/case dropped (0.478 vs 0.722, P=0.003), and IoU improved (0.194 vs 0.153, P=0.014).
FDA clears DRTECH Corporation's EXSYS DEXi digital X-ray detectors (models S40S, S40P, S40T, S50H) under 510(k) for general radiography.
A decision tree model using multi-phase CT radiomic features achieved AUC 0.977 (95% CI 0.934–1.000) in lesions with washout >60% and 0.983 (95% CI 0.962–1.000) in those without, distinguishing adrenal lipid-poor adenomas from pheochromocytomas.
Radiology Partners Employer Profile - American College of Radiology Career Center
<![CDATA[Mushing Stuff Around a Dinner Plate: The Modern Approach to Health-Care Spending]]>