PurposeLesions in the spinal cord (SC) can be found in up to 83% of patients with multiple sclerosis (MS). As they are mainly located in the cervical segment, many centers exclude the thoracic part from SC imaging. The aim of our study was to quantify the decrease in sensitivity resulting from this approach.MethodsMR images (3T) of 543 consecutive patients with clinically isolated syndrome (CIS) (n: 37) and MS (n: 506) were analyzed retrospectively. Lesions were segmented semi-automatically on axial T2-weighted images of the whole SC using BrainSeg3D. The volume of lesions was related to vertebral levels.ResultsAltogether 1782 lesions (CIS: 19; MS: 1763) were found in 409 patients. 70% of the lesion volume was located in the SC above the 3rd thoracic vertebral body, in a segment that is commonly covered by an isolated examination of the cervical SC. However, 26 patients (6%) showed lesions exclusively below the 3rd thoracic vertebral body, thus 94% of all patients with SC lesions could be detected with isolated MR imaging of the cervical SC.ConclusionThough the majority of lesions can be detected in an isolated examination of the upper part of the SC, some patients showed lesions exclusively below the 3rd thoracic vertebral body. We recommend routine scanning of the whole SC in suspected MS.

The Panoptic Quality (PQ) metric is the standard for jointly evaluating instance and semantic segmentation. However, its original definition relies on a One-to-One matching between predicted and ground truth segments, which is only straightforward when the IoU threshold exceeds 0.5. Below 0.5, multiple matching strategies emerge in a poorly explored problem space. We systematically elucidate this space by recasting segment matching as a constrained bipartite assignment problem. Independently bounding the prediction- and ground-truth-side degrees yields four matching strategies: One-to-One, Many-to-One, One-to-Many, and Many-to-Many. We show that the first three are well-defined within the PQ framework, while Many-to-Many falls outside it. These strategies become relevant when instances are fragmented, adjacent objects are difficult to delineate, or annotations are noisy. Central to our framework is a vertex-based accounting of TP, FN, and FP, anchored to ground truth and predicted segments rather than to matching edges. We further show that the framework extends naturally to part-aware panoptic segmentation, and we explore part-aware evaluation on biomedical data. Across configurable case studies we report how different combinations of thresholds and matching strategies behave in practice. We release a unified open-source package built on Panoptica. It exposes Voronoi-based region-wise analysis, part-aware evaluation, and Area Under Threshold Curve computations as configurable options.

Background:Precise glioma segmentation in magnetic resonance imaging (MRI) is essential for accurate diagnosis, optimal treatment planning, and advancing clinical research. However, most deep learning approaches require complete, standardized MRI protocols that are frequently unavailable in routine clinical practice. This study presents and evaluates GlioMODA, a robust deep learning framework designed for automated glioma segmentation that delivers consistent high performance across varied and incomplete MRI protocols. Methods:GlioMODA was trained and validated on the BraTS 2021 dataset (1251 training, 219 testing cases), systematically assessing performance across 11 clinically relevant MRI protocol combinations. Segmentation accuracy was evaluated using Dice similarity coefficients (DSC) and panoptic quality metrics. Volumetric accuracy was benchmarked against manual ground truth, and statistical significance was established via Wilcoxon signed‑rank tests with Benjamini-Yekutieli correction. Results:GlioMODA demonstrated state-of-the-art segmentation accuracy across tumor subregions, maintaining robust performance with incomplete or heterogeneous MRI protocols. Protocols including both T1-weighted contrast-enhanced and T2-FLAIR sequences yielded volumetric differences vs manual ground truth that were not statistically significant for enhancing tumor (median difference 55 mm³, P = .157) and whole tumor (median difference -7 mm³, P = 1.0), and exhibited median DSC differences close to zero relative to the 4‑sequence reference protocol. Omitting either sequence led to substantial and significant volumetric errors. Conclusions:GlioMODA facilitates reliable, automated glioma segmentation using a streamlined 2‑sequence protocol (T1‑contrast + T2‑FLAIR), supporting clinical workflow optimization and broader implementation of quantitative volumetry compatible with RANO 2.0 criteria. GlioMODA is published as an open-source, easy-to-use Python package at https://github.com/BrainLesion/GlioMODA/.

Individual intervertebral disc (IVD) characteristics contribute to spinal mechanics and pain, but individualized biomechanical modeling is insufficiently examined, especially regarding the validity of using a single material model across different degeneration grades and disc geometries. Based on MRI images, we generated and simulated 241 patient-specific lumbar finite element method (FEM) IVD models using a single material model calibrated to a healthy L4-L5 disc. This study is twofold: In one respect, it establishes and evaluates the automated morphing algorithm femReg, which is used to generate the models. In another respect, it evaluates the capacity of the geometry-only individualized IVD models to represent the range of motion (ROM) of variously degenerated discs by simulating them in four load cases under 1, 2.5, 5 and 7.5 Nm load. To do so, we grouped numerical models by height loss and compared absolute and normalized ROM results with in vitro data across different degeneration grades. 241 high-quality hexahedral IVD models (mean aspect ratio: 1.65, mean Hausdorff distance: 0.037 mm) were created and simulated within an average runtime of 6 minutes per IVD. Healthy models (n = 149) reproduced experimental ROM with root mean square errors (RMSEs) of 0.96∘-2.05∘ across bending load cases. The closest agreement was observed in lateral bending (RMSE = 0.68∘), where facet joint contributions are minimal. In contrast, degenerated models exhibited reduced height-normalized ROM compared with in vitro data (RMSE = 0.05-0.24∘/mm), indicating an effectively stiffer response and suggesting that degeneration-specific material parameters, such as fiber loosening or reduced hydration, should be incorporated to better capture pathological behavior.

Precise molecular characterization of glioblastoma (GB) is fundamental for accurate risk stratification and therapeutic planning. DNA methylation profiling reliably identifies key molecular features, including O(6)-methylguanine-DNA methyltransferase (MGMT) promoter methylation status and specific molecular subtypes, such as receptor tyrosine kinase (RTK) I and II, and the mesenchymal (MES) subtype. In this study, we investigated the hypothesized correlation between these molecular profiles and preferential tumor locations, which could reveal a link to underlying tumor biology. We analyzed 227 GB patients characterized by DNA methylation profiling. To map significant clusters of tumor occurrence across subtypes and subcomponents, we performed voxel-wise analysis of differential involvement, utilizing 500 permutations to correct for multiple comparisons. While uncorrected frequency differential maps suggested localization tendencies for the RTK I, RTK II, and MES subtypes, stringent statistical correction revealed only one robust association: the non-enhancing component of MES tumors showed significant clustering in the left frontal lobe, the insula, and the temporal lobe. Contrary to prior literature, we observed no significant hemispheric preference regarding MGMT promoter methylation status. Our findings challenge prior assumptions regarding the spatial distinctiveness of GB subtypes and highlight the need to further elucidate the mechanisms governing tumorigenesis and spatial growth patterns.

Interpreting quantitative CT biomarkers, such as organ volume and tissue attenuation, requires large-scale healthy reference distributions. However, creating these is challenging because clinical datasets are often heavily enriched with pathology. Here, we develop an evidence-grounded, cross-verified large language model (LLM) ensemble to filter pathological findings from radiology reports, enabling the construction of pathology-reduced cohorts from over 350,000 CT examinations. Five LLMs, first, flag structure-level abnormality candidates grounded in verbatim report evidence and, second, resolve disagreements via cross-verification. Using distribution-aware generalized additive models for location, scale, and shape, we establish comprehensive whole-body reference charts for 106 anatomical structures (volumes and attenuation) across adulthood, accounting for age, sex, contrast enhancement, and acquisition parameters. Longitudinal analyses reveal structure- and contrast-dependent changes distinct from cross-sectional trends. These resources facilitate covariate-adjusted centile scoring from routine CT, supporting standardized quantitative phenotyping, multi-site imaging studies, and scalable opportunistic screening research.

Background Studies have demonstrated that large language models (LLMs) can perform differential diagnosis based on textual radiologic findings; however, it is unclear how variations in reader-generated inputs affect LLM performance and clinical utility. Purpose To evaluate how reader experience influences the diagnostic benefit of LLM assistance in brain MRI differential diagnosis. Materials and Methods In this retrospective multireader study, neuroradiologists (n = 4), radiology residents (n = 4), and neurology/neurosurgery residents (n = 4) provided textual radiologic findings and their top three differential diagnoses for brain MRI scans with confirmed diagnoses obtained between January 2009 and April 2024 from a single academic center. Confirmed diagnoses were established histopathologically or through consensus of at least two neuroradiologists. Three LLMs (GPT-4.1 [OpenAI], Gemini 2.5 Pro [Google DeepMind], and DeepSeek-R1 [Hangzhou DeepSeek Artificial Intelligence Basic Technology Research]) generated differential diagnoses based on reader-provided findings. Readers revised their diagnoses after reviewing the suggestions of GPT-4.1. A cumulative link mixed model was fitted to evaluate the association between reader experience and diagnostic benefit, with change in diagnostic result as an ordinal outcome, reader experience as a predictor, and random intercepts for rater and patient. Results Forty brain MRI scans (mean patient age, 50 years ± 18 [SD]; 23 female) were included. LLM-generated diagnoses achieved the highest top-three accuracy based on imaging findings from neuroradiologists (78.8%-83.8% across LLMs), followed by radiology residents (71.8%-77.6%) and neurology/neurosurgery residents (63.2%-67.1%). Mean absolute gains in top-three accuracy with LLM assistance diminished with increasing experience: +19.4% for neurology/neurosurgery residents (from 43.2% to 62.6%), +14.7% for radiology residents (from 59.6% to 74.4%), and +4.4% for neuroradiologists (from 83.1% to 87.5%). Models demonstrated a negative association between reader experience and diagnostic benefit from LLM assistance (β = -0.10; P = .005) and a positive association of reader experience with correctness (β = 0.11; P < .001) and completeness (β = 0.18; P = .002) of imaging findings. Conclusion With increasing reader experience, LLM accuracy with reader-generated input improved, whereas accuracy gains from LLM assistance diminished. © RSNA, 2026 Supplemental material is available for this article. See also the editorial by McMillan in this issue.

Thoracolumbar stump ribs are one of the essential indicators of thoracolumbar transitional vertebrae or enumeration anomalies. While some studies manually assess these anomalies and describe the ribs qualitatively, this study aims to automate thoracolumbar stump rib detection and analyze their morphology quantitatively. To this end, we train a high-resolution deep learning model for rib segmentation using nnUNet and achieve significant improvements over existing models (Dice score 0.997 vs. 0.779, p-value < 0.01). In addition, we employ a novel iterative algorithm and piecewise linear interpolation to estimate rib length, achieving a success rate of 98.2%. When analyzing morphological features, we show that stump ribs articulate more posteriorly at the vertebrae (-19.2 +/- 3.8 vs. -13.8 +/- 2.5 mm, p-value < 0.01), are thinner (260.6 +/- 103.4 vs. 563.6 +/- 127.1 mm2, p-value < 0.01), and are oriented more downwards and sideways within the first centimeters in contrast to full-length ribs. We show that with partially visible ribs, these features can achieve an F1-score of 0.84 and an AUC of 0.98 in differentiating stump ribs from regular ones. We publish the model weights and masks for public use.

Routine laboratory panels drawn during cancer treatment constitute longitudinal physiological recordings of organ function, yet their temporal structure is discarded by single-timepoint prognostic tools. A transformer trained on 2,777,595 laboratory measurements from 3,905 patients with multiple myeloma or ovarian cancer predicted the two-year onset of 162 treatment-associated complications, including therapy-related myelodysplastic syndromes, spanning eight clinical categories, achieving 1.5- to 6.1-fold enrichment above prevalence at the group level. It matched or outperformed non-sequential baselines across grouped endpoints (AUROC gains up to +0.11), demonstrating that longitudinal laboratory trajectories capture evolving complication-specific physiology inaccessible from isolated measurements. Predictions generalised across both cancers, divergence concentrating in disease-specific complications, and biomarker masking recovered signatures consistent with established pathophysiology. External validation on MIMIC-IV and MMRF CoMMpass confirmed transferability across independent healthcare systems (AUROC up to 0.85). Routine oncological laboratory data encode organ deterioration weeks to months before clinical onset, enabling complication-specific surveillance without additional testing infrastructure.

Abstract Background Choroid plexus (CP) volume is an emerging magnetic resonance imaging (MRI) biomarker in various disorders of the central nervous system (CNS). However, clinical translation is hindered by methodological heterogeneity and inconsistent anatomical coverage. Double inversion recovery (DIR) – a sequence providing dual-tissue suppression – is a promising candidate to improve CP segmentation. Methods The dataset included 93 scans across healthy subjects and individuals with multiple sclerosis (MS), divided into a training set ( n = 63), an internal test set ( n = 20), and an external test set ( n = 10). First, relative CP signal intensity and tissue contrast ratios on DIR were compared against fluid-attenuated inversion recovery (FLAIR) and T1-weighted (T1w) sequences (pre- and post-contrast). Reproducibility of manual CP segmentations was assessed via intraclass correlation coefficients (ICCs). Subsequently, we developed a 3D nnU-Net model for CP segmentation based on manually labeled DIR masks. Model performance was evaluated against manual segmentation using spatial overlap and volumetric error metrics. Finally, we compared our DIR-based model against three publicly available T1w- or FLAIR-based tools by assessing slice-wise volume distributions and voxel-wise density maps. Results DIR demonstrated the highest CP signal intensity and most consistent tissue contrast among evaluated MRI sequences ( p < 0.001). Intra- and inter-rater agreement for manual CP segmentations was robust (ICC = 0.92 and 0.83, respectively). The trained nnU-Net achieved high internal accuracy (Dice = 0.82) independent of scanner, diagnosis, or absolute CP volume, and generalized well to the external test set (Dice = 0.75). Compared to public T1w- and FLAIR-based models, DIR-based approaches (nnU-Net and manual) yielded significantly larger CP volumes ( p < 0.01). Axial volume distribution analysis attributed this difference to a distinct bimodal profile in DIR segmentations, more fully capturing the CP inside the temporal horn of the lateral ventricle ( p < 0.001 against T1w- and FLAIR-based models). Conclusions By leveraging the superior tissue contrast of DIR, our nnU-Net model achieves highly accurate CP segmentation that generalizes across scanners and captures the inferior extent of the C-shaped structure often missed by conventional models. This may improve standardization of CP volumetry and allow for more reliable studies in CNS disorders.

BACKGROUND:Long-term evidence for endovascular treatment in medium or distal vessel occlusion stroke is scarce. Three out of four randomised trials reported no benefit of endovascular treatment over best medical treatment at 90 days. We aimed to assess efficacy of endovascular treatment plus best medical treatment versus best medical treatment alone at 12 months in patients enrolled in the DISTAL trial, as well as overall survival. METHODS:DISTAL was an open-label, randomised trial with blinded endpoint assessment conducted at 55 hospitals in Europe and the Middle East. Adults (≥18 years) with acute ischaemic stroke due to medium or distal vessel occlusion (occlusion of co-dominant or non-dominant M2 or M3-M4 middle cerebral artery, A1-A3 anterior cerebral artery, or P1-P3 posterior cerebral artery) presenting from home within 6 h of last known well, or between 6 h and 24 h if neuroimaging demonstrated potentially salvageable tissue, were randomly assigned (1:1) through a centralised web-based system to endovascular treatment plus best medical treatment or best medical treatment alone. The prespecified primary outcome at 12 months was disability measured by use of the ordinal modified Rankin Scale (mRS; scores 5 and 6 combined) in the intention-to-treat population. The only safety outcome was overall survival. The trial is registered on ClinicalTrials.gov (NCT05029414) and is completed. FINDINGS:Between Dec 16, 2021, and July 10, 2024, we enrolled 553 patients. Ten patients declined post-hoc consent, leaving 543 participants in the analysis (239 [44%] females and 304 [56%] males; median age 77 years, IQR 68-84). 271 (50%) were assigned to endovascular treatment plus best medical treatment and 272 (50%) to best medical treatment alone. The median NIHSS score at admission was 6 (IQR 5-9); 355 (65%) participants received intravenous thrombolysis. Predominant occlusion locations were the M2 (239 [44%]), M3 (146 [27%]), P2 (73 [13%]), and P1 (30 [6%]) segments. 12-month data were available for 524 (97%) participants. The median mRS score was 2 (IQR 1-4) in the endovascular treatment plus best medical treatment group and 2 (1-4) in the best medical treatment alone group. There was no difference in 12-month mRS distribution between endovascular treatment plus best medical treatment and best medical treatment alone (adjusted common odds ratio for better functional outcome 0·81, 95% CI 0·59-1·12; p=0·20). Overall survival was similar between the two groups (hazard ratio 1·46, 95% CI 0·93-2·30; p=0·10). INTERPRETATION:In patients with a medium or distal vessel occlusion stroke, endovascular treatment plus best medical treatment was not associated with a reduction of disability or death at 12 months compared with best medical treatment. These results are consistent with the 90-day results. Routine endovascular treatment is therefore not supported for patients with mild-to-moderate medium or distal vessel occlusion stroke. FUNDING:Swiss National Science Foundation, Gottfried und Julia Bangerter-Rhyner-Foundation, Medtronic, Stryker Neurovascular, Phenox, Rapid Medical, and Penumbra.

To develop and evaluate an automated workflow for the setup of single-vertebra finite element (FE) simulations from clinical CT data. Specifically, we quantified how automated endplate identification, vertebra-specific coordinate system definition, and load-application-point assignment influence the simulated fracture-load estimates. We analyzed 113 vertebrae from 70 patients that had previously undergone manual FE setup. The automated pipeline identified vertebral endplates, assigned a vertebra-specific coordinate system, and assigned the load application point for axial compression simulations. Automated setups were visually graded as good, acceptable, or bad using predefined criteria. Agreement with manual reference models was evaluated, and the effects of each setup component on fracture-load estimates were quantified separately. Of 113 vertebrae, 53 (47

Vertebral fractures are severe complications of osteoporosis but are frequently missed on computed tomography (CT). Differentiating true fractures from non-osteoporotic vertebral height loss remains challenging; deep learning (DL) models may improve detection and grading. In this retrospective study, we evaluated eight human raters with different expertise (three students, three residents, and two attendings), four DL models, and one DL-based commercial software using the public Vertebral Segmentation (VerSe) 19 20 datasets. Vertebral fractures were graded using the semiquantitative Genant scale (0–3). Diagnostic performance was evaluated using interrater agreement and classification metrics, with significance tested via generalized linear mixed models across patient and vertebral levels for the thoracolumbar spine and its regional subsets. Consensus readings by a senior neuroradiologist and an experienced resident, informed by clinical data, served as the reference standard. 3548 thoracic and lumbar vertebrae from 331 patients were analyzed. 190 (5.4

PURPOSE:Opportunistic CT-based screening for osteoporosis has been shown to be cost-effective and may help to reduce the existing treatment gap in patients. Reproducibility has not yet been investigated at thoracic levels, which is becoming increasingly relevant with upcoming lung cancer screening programs. METHODS:Scans from 178 patients with two consecutive CT examinations within one month were retrospectively included. In all available reconstructions, automated extraction of trabecular volumetric bone mineral density (vBMD) was performed, including HU-to-BMD conversion and contrast phase correction. Mean trabecular vBMD was calculated for two thoracic regions (T5-T7 and T8-T10). Reproducibility was calculated as the root mean square coefficient of variation (RMSCV) under six reproducibility scenarios, ranging from different reconstructions of the same scan to measurements across different scanners and contrast phases. RESULTS:In total, 914 vBMD observations were analyzed. RMSCV values ranged from 0.83% to 4.90% for T5-T7 and from 1.4% to 6.8% for T8-T10. One-way ANOVA showed a significant overall difference between reproducibility scenarios in both vertebral regions (all p < 0.001), with post hoc analyses demonstrating a stepwise increase in RMSCV with technical complexity. CONCLUSION:Our results support the clinical applicability of opportunistic osteoporosis screening at the thoracic spine across diverse CT acquisition protocols. With the upcoming national lung cancer screening programs utilizing low-dose CT (LDCT), an extraordinary opportunity emerges to address the substantial osteoporosis treatment gap through opportunistic bone health assessment. ADVANCES IN KNOWLEDGE:This work is novel in extending reproducibility validation for opportunistic osteoporosis screening from the lumbar spine to thoracic vertebrae.

The finite element (FE) method is a cornerstone of patient-specific biomechanical analysis, yet most workflows assign isotropic linear elastic behaviour, and neglect bone's intrinsic anisotropic and non-linear response to load. We present PBMGA (Python-based Bone material grouping and anisotropy), a novel open-source tool that automates the calculation and element-specific assignment of non-linear and transversely isotropic (and, in principle, more general anisotropic) bone material parameters using user-defined equations. PBMGA integrates three customisable material grouping strategies: Percentual Thresholding, Adaptive Clustering, and Equidistant Grouping, to compress the number of unique material sets, significantly reducing computational complexity in downstream FE simulations without compromising accuracy. Its modular architecture supports seamless integration with existing preprocessing workflows and scalable analysis of large clinical datasets. By combining accurate material modelling with high-throughput capability, PBMGA enhances biomechanical prediction and paves the way for more efficient clinical diagnostics and treatment planning.

ObjectiveTo assess the discriminatory ability of vertebra-specific volumetric bone mineral density (vBMD), finite element analysis-derived fracture load (FEA-derived FL), and texture analysis (TA) features for incidental vertebral fractures, and to compare performance between thoracic and lumbar levels.Materials and methodsWe retrospectively reviewed baseline and follow-up thoracolumbar CT scans from 420 patients and identified 11 patients with incidental vertebral fractures contributing to 20 fractured vertebrae (7 females; mean age 65.5years). For each fractured vertebra, three level-matched control vertebrae from patients without fractures were selected, yielding 58 controls across 29 control patients (total 78 vertebrae). Parameters evaluated include vBMD, FEA-derived FL, and TA features (24 total). Discriminatory ability was assessed using area under the curve (AUC) values.ResultsvBMD, FEA-derived FL, and 4 of 24 TA features showed group-wise differences between fractured and control vertebrae groups. AUCs were 0.76 [95% CI 0.55-0.90] (vBMD) and 0.73 [95% CI 0.52-0.90] (FEA-derived FL); selected texture features ranged 0.70-0.72. Region-stratified AUC point estimates were higher in the lumbar than in the thoracic vertebrae, but the 95% CIs were wide/overlapping; comparisons are descriptive.ConclusionvBMD had the numerically largest AUC point estimate for discriminating fractured from control vertebrae; FEA-derived FL was similar, and selected texture features showed modest discrimination with comparable point estimates across lumbar and thoracic levels, generating the hypothesis of less region dependence. Regional comparisons are descriptive. Findings are exploratory and intended to prioritize candidate measures for validation and future multivariable modeling before any clinical application.

The human spine commonly consists of seven cervical, twelve thoracic, and five lumbar vertebrae. However, enumeration anomalies may result in individuals having eleven or thirteen thoracic vertebrae and four or six lumbar vertebrae. Although the identification of enumer- ation anomalies has potential clinical implications for chronic back pain and operation planning, the thoracolumbar junction is often poorly as- sessed and rarely described in clinical reports. Additionally, even though multiple deep-learning-based vertebra labeling algorithms exist, there is a lack of methods to automatically label enumeration anomalies. Our work closes that gap by introducing "Vertebra Identification with Anomaly Handling" (VERIDAH), a novel vertebra labeling algorithm based on multiple classification heads combined with a weighted vertebra sequence prediction algorithm. We show that our approach surpasses existing mod- els on T2w TSE sagittal (98.30

PURPOSE:Normative modeling allows to assess individual MRI data against a normative reference cohort. Currently, however, mainly one-dimensional features such as cortical thickness (CTh) are considered, which neglect the complex network architecture of the brain. Here, we assess whether informing normative modeling with a complex feature reflecting complex brain network architecture can enhance personalized medicine for disorders affecting complex brain networks such as multiple sclerosis (MS). METHOD:We calculated normative trajectories of small-worldedness (SWI) and CTh across the lifespan from T1w data (456 RRMS, 84 HC). We investigated differences in RRMS and HC trajectories by evaluating the best-fit of different polynomial functions. Additionally, we calculated trajectories for clinical (EDSS scores) and neuropsychological (cognition/fatigue) variables in RRMS and studied a potential predictive value of network changes using Granger Causality Testing. RESULTS:Cognition (via MuSIC composite) exhibited an inverse U-shaped quadratic trajectory, peaking mid-life before declining, akin to SWI. We found that in RRMS, small-worldedness follows a quadratic trend across the lifespan, while CTh exhibits a linear negative trend. In HCs, CTh followed a linear negative trend (β = -.012 mm/year, p < 0.001), mirroring patients. EDSS, similar to CTh, exhibited a linear accumulation throughout age. We found a Granger Causal relationship between small-worldedness and fatigue (observed F-value = 6.704), which lagged behind network changes. CONCLUSION:We conclude that informing normative modeling with a complex feature reflecting complex brain network architecture can enhance personalized medicine for disorders affecting complex brain network structures such as MS. Additionally, we provide initial evidence that MS-fatigue might partly result from changes in network structure, which needs to be carefully evaluated in future studies.

Abstract Purpose High-resolution T2-weighted imaging is essential for preoperative assessment before cochlear implantation. Compressed sensing (CS) with AI-based reconstruction (CSAI) reduces acquisition times whilst preserving image quality. Although CSAI has been established in various clinical applications, its performance in inner ear imaging remains unclear. This study assesses CSAI-optimized T2-DRIVE sequences at different resolutions and acquisition times for visualizing inner ear structures. Methods In 30 healthy participants, CS T2-DRIVE was acquired at isotropic resolutions of 0.65, 0.5, and 0.4 mm. 0.5 and 0.4 mm datasets were also reconstructed using a commercially available AI-based reconstruction algorithm. Three raters independently assessed the imaging quality of anatomical landmarks (cochlea, semicircular canals, vestibulocochlear nerve), artifacts, and signal-to-noise ratio (SNR) using a 5-point Likert scale. Each rater re-rated a subset of images after ≥ 4 weeks. Inter- and intra-rater reliability were calculated using quadratically weighted Cohen’s kappa, and differences between sequences were analyzed using cumulative link mixed models (CLMM). Results 0.4 mm isotropic imaging exhibited lower SNR compared to CSAI 0.5 mm, regardless of reconstruction algorithm ( p < 0.001). Across all raters, CSAI T2 at 0.5 mm resolution significantly improved delineation of the cochlea and vestibulocochlear nerve compared to 0.65 mm imaging ( p < 0.001), while assessability of semicircular canals was reduced ( p = 0.082). Acquisition times increased with higher resolutions (0.65/0.5/0.4 mm: 4:02/4:25/4:34 min). Conclusion AI-driven reconstruction algorithms enable statistically significant improvements in imaging of key inner-ear structures with minimal increases in scan time at 0.5 mm resolution.

BACKGROUND:Accurate device sizing is crucial for successful flow diverter (FD) therapy in intracranial aneurysms. This study assesses the accuracy and clinical utility of the Ankyras virtual simulation software (Mentice AB, Gothenburg, Sweden) in predicting FD length across multiple device types and manufacturers. METHODS:We retrospectively analyzed 193 FDs (7 device types) deployed in 180 patients with 230 intracranial aneurysms. Simulation-based prediction of effective lengths (simulated length (SL)) and nominal manufacturer specifications (labeled length) were compared with measured in vivo lengths (measured length (ML)). Performance was evaluated using simulation accuracy (SA), absolute error (AE), relative error (RE), length ratio, and correlation analysis. RESULTS:Virtual simulation demonstrated good clinical usability, achieving a mean absolute deviation of only -1.38 mm at the proximal landing zone compared with ML. Simulation-based predictions showed significantly superior accuracy compared with nominal manufacturer specifications (SA 89.6±11.3% vs 81.8±13.7%, P<0.001), with mean AE reduced by 58% (-1.38 mm vs -3.34 mm) and mean RE by 42% (10.5% vs 18.2%). Strong correlations between SL and ML (r=0.900) validated predictive reliability across all tested device types. Centerline correction technology further enhanced parameters such as SA to 94.8±9.5%. CONCLUSION:Virtual simulation-based prediction of effective FD length using the Ankyras software showed reliable and clinically meaningful results, enabling accurate estimations of the proximal landing zone and overall length. Ankyras, along with comparable simulation platforms, may provide considerable potential to facilitate FD selection and implementation, especially in anatomically complex cases.