IntroductionThe COVID-19 pandemic imposed severe burdens on global health infrastructure, underscoring the urgent need for rapid, reliable, and scalable diagnostic technologies. Conventional radiological modalities such as chest X-rays and computed tomography (CT) face critical deployment barriers, including ionizing radiation exposure, elevated procedural costs, and substantial infection-control risks in resource-constrained settings. Lung ultrasound (LUS) presents a compelling alternative owing to its portability, absence of ionizing radiation, and suitability for point-of-care application; however, systematic analysis of deep learning architectures optimized for LUS-based COVID-19 classification remains limited.MethodsTo address this gap, this study presents a comprehensive benchmarking framework in which 13 pretrained convolutional neural network (CNN) architectures—AlexNet, VGG-16, VGG-19, ResNet-50, ResNet-101, ResNet-152, DenseNet-121, DenseNet-169, DenseNet-201, InceptionV3, Inception-ResNetV2, MobileNetV2, and Xception—were evaluated under seven gradient-based optimization strategies: SGD, Adagrad, RMSProp, Adadelta, Adam, AdaMax, and Nadam. Transfer learning was applied uniformly across all architectures to compensate for the limited scale of the publicly available POCOVID-Net dataset, which comprised 455 LUS frames from confirmed COVID-19 cases and 226 frames from non-COVID subjects. In addition, geometric transformation-based data augmentation was employed to mitigate the effects of class imbalance and improve inter-patient generalization.ResultsAmong the architecture–optimizer combinations evaluated under the hold-out scenario, DenseNet-121 with the Adam optimizer achieved perfect classification with 100% accuracy, surpassing all previously reported LUS-based COVID-19 detection methods evaluated under comparable single-dataset conditions. Given the risk that a single hold-out split may overstate performance, rigorous stratified 5-fold cross-validation was subsequently performed on this best-performing hold-out model to ensure robustness and address concerns regarding overfitting. This yielded a mean accuracy of 99.11% ± 0.35%, sensitivity of 99.41% ± 0.31%, specificity of 99.18% ± 0.38%, F1-score of 99.36% ± 0.29%, and AUC of 99.45% ± 0.25%.Discussion/ConclusionDespite consistently strong performance across both evaluation schemes, these results were derived from a single, limited-diversity public dataset and, therefore, represent a methodological benchmark rather than a validated estimate of real-world performance. External multicenter validation is necessary before clinical application of the model. Nonetheless, the evaluated DenseNet-121 framework holds practical utility as a rapid triage aid and complementary screening tool within clinical workflows, particularly in low-resource environments where CT infrastructure is unavailable.
BackgroundConventional visual magnetic resonance imaging (MRI) assessment of neurovascular compression may overlook subtle trigeminal nerve abnormalities. We evaluated the pooled discriminative performance of MRI-based artificial intelligence (AI) models for identifying pain-related trigeminal nerve phenotypes in non-secondary trigeminal neuralgia.MethodsPubMed, Embase, Web of Science, and the Cochrane Library were searched from inception to April 2026. Eligible studies developed or validated MRI-based radiomics or AI models and provided sufficient information to construct or approximate 2 × 2 contingency tables. Risk of bias was assessed using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool. Sensitivity and specificity were jointly pooled using a bivariate random-effects model, and a summary receiver operating characteristic curve was constructed.ResultsEight studies comprising 628 nerve-level observations (298 symptomatic or affected nerves and 330 comparator nerves) were included. Pooled sensitivity was 0.83 (95% confidence interval [CI], 0.78–0.87), pooled specificity was 0.81 (95% CI, 0.74–0.86), and the summary area under the curve was 0.86 (95% CI, 0.82–0.88). Results remained stable after leave-one-out analysis. Radiomics-only models and models combining radiomic and anatomical features showed broadly comparable performance.ConclusionsPreliminary evidence suggests that MRI-based AI models provide moderate discriminative ability for identifying the affected nerve side in trigeminal neuralgia, with relatively stable sensitivity but lower specificity. Given the limited number of studies, particularly in subgroup analyses, these findings should be interpreted cautiously. Such models may currently serve as adjunctive imaging tools. Larger multicenter studies with independent external validation and clinically relevant reference standards are needed.
ObjectiveTo compare the clinical, perioperative, and radiological outcomes of endoscopic unilateral biportal endoscopic lumbar interbody fusion (ULIF) and conventional posterior lumbar interbody fusion (PLIF) for single-level degenerative lumbar disease.MethodsThis retrospective study included 67 patients undergoing lumbar interbody fusion (ULIF, n = 29; PLIF, n = 38). Perioperative outcomes included operative time, blood loss, and hospital stay. Clinical outcomes were assessed using the visual analogue scale (VAS) and Oswestry Disability Index (ODI). Radiological outcomes included changes in spinal canal cross-sectional area (CSA), canal diameter, and fusion status.ResultsOperative time was longer in the ULIF group than in the PLIF group (226.5 ± 50.6 vs. 184.1 ± 57.4 min, p = 0.002). However, ULIF resulted in less blood loss (87.6 ± 56.7 vs. 251.2 ± 198.0 mL, p < 0.001), shorter total hospital stay (11.7 ± 4.0 vs. 14.6 ± 4.6 days, p = 0.008), and shorter postoperative hospital stay (4.1 ± 1.0 vs. 8.5 ± 1.4 days, p < 0.001). VAS improvements were comparable between the two surgical approaches. Although PLIF showed a greater reduction in ODI from baseline, absolute ODI values at 12 months were similar.ConclusionsULIF was associated with less intraoperative blood loss and shorter postoperative hospitalization while achieving similar 12-month pain outcomes and fusion rates to PLIF in this cohort. No statistically significant between-group differences were detected in postoperative complications. The modest radiological differences observed with ULIF should be interpreted cautiously, and the small observational cohort was not powered to establish equivalence or comparable safety between the two procedures. Larger prospective studies with longer follow-up are warranted.
PurposeTo propose a novel stent implantation technique for chronic iliocaval occlusion—the One-piece Bifurcated Stent-Graft with Bridging Technique (OBSGBT)—which utilizes a one-piece bifurcated arterial stent-graft as the main body, bridged with venous stents.Case reportWe report a 76-year-old male patient who underwent retrievable inferior vena cava (IVC) filter placement for lower extremity deep venous thrombosis 4,913 days prior. Over time, he developed chronic complete occlusion of the left iliac vein and severe stenosis of the IVC below the filter. After successful filter retrieval, the iliocaval venous channel was reconstructed using the OBSGBT. At 1-year follow-up, all stent segments remained patent, with no stent migration, fracture, or thrombosis.ConclusionFor filter-related chronic iliocaval occlusion, the OBSGBT appears feasible, safe, and effective. Its preliminary outcomes appear promising compared with conventional kissing-stent techniques, warranting further investigation in larger cohorts.
Converging evidence supports dystonia as a disorder of large-scale brain networks arising from dysfunction within, or impaired communication between, nodes of the cortico-basal ganglia-thalamo-cortical and cerebello-thalamo-cortical circuits. Focal hand dystonia is associated with abnormalities across cortical and subcortical networks. High intensity focused ultrasound (HiFU) offers a novel therapeutic approach that provides an opportunity to examine treatment related structural and functional network changes. Eleven individuals with focal hand dystonia, of musician's dystonia and writer's cramp subtype, underwent multimodal MRI, including structural analysis, diffusion tensor imaging (DTI), and resting-state functional MRI (rs-fMRI) at baseline, with comparison made to ten age- and sex-matched healthy control participants. These imaging metrics were repeated at 1, 3 and 12 months following HiFU ventro-oral (Vo)-complex thalamotomy. This trial was registered with the Australia New Zealand Clinical Trials Registry (ACTRN12622000775718p). No differences in regional volumes were observed at baseline between the groups. Following thalamotomy, small uncorrected volumetric changes were noted in several regions of interest, and DTI demonstrated lower tract-wise fractional anisotropy (FA) and coefficient of linearity (CL) within the treated pallidothalamic tract, with node-wise analysis demonstrating focal microstructural alterations that were most apparent at 12 months following treatment. There were no corresponding changes in fronto-striatal or fronto-thalamic tracts. rs-fMRI revealed baseline alterations in intra- and inter-network connectivity involving sensorimotor, visual, attentional, default mode and executive networks. Following thalamotomy, changes in functional connectivity were modest and temporally variable, without evidence of large-scale network reorganisation. These findings suggest that focal hand dystonia is associated with distributed functional network dysfunction, in the context of preserved macroscopic structure. We propose that Vo-complex thalamotomy is associated with focal, localized microstructural alterations within the pallidothalamic tract, most notable at 12 months, which may contribute to modulation of pathological pallidothalamic drive rather than widespread restoration of functional network connectivity.
