This study investigated brain structural changes associated with NOTCH2NLC gene mutations in neuronal intranuclear inclusion disease (NIID) patients, focusing on the evolutionary implications of this human-specific gene in brain development. We analysed 41 NIID patients and 21 healthy controls using voxel-based morphometry and surface-based morphometry to assess differences in grey matter volume and cortical complexity. Spatial relationships between brain atrophy and white matter hyperintensity volume as well as cerebrospinal fluid fraction were examined. Additionally, we conducted exploratory Spearman correlation analyses to evaluate associations between regional grey matter volume and clinical variables, including GGC repeat length, disease duration, age at onset and cognitive scores. NIID patients exhibited extensive reductions in grey matter volume and cortical thinning in multiple brain regions, with pronounced effects in the prefrontal cortex and cerebellum. The parietal lobe, insula and posterior cingulate gyrus showed decreased gyrification index and fractal dimension, while certain regions of the temporal and frontal lobes showed increased gyrification index and fractal dimension. Furthermore, in the NIID group, white matter hyperintensity volume and cerebrospinal fluid fraction were negatively correlated with grey matter volume in the olfactory cortex, orbital gyrus, anterior cingulate gyrus, insula, amygdala and temporal pole. Exploratory analyses suggested that longer GGC repeats were associated with greater atrophy in the striatum, middle cingulate cortex, sensorimotor cortex and cerebellum; earlier age at onset with thalamic (mediodorsal/pulvinar), occipital and cerebellar atrophy; and poorer cognitive scores with atrophy in the anterior cingulate cortex, superior occipital gyrus and superior temporal pole. This study uncovers widespread and complex cerebral structural changes in NIID patients, predominantly affecting the prefrontal cortex, cerebellum, insula and limbic system structures. These findings provide new insights into the neuroanatomical basis of NIID and support the hypothesis that human-specific genetic innovations driving cortical expansion may concurrently confer selective vulnerability to neurodegeneration.

BACKGROUND:Recent NIH Data Management and Sharing (DMS) policy updates and NIH controlled-access data security requirements have increased attention to facial anonymization and controlled-access handling of shared head imaging data. This is particularly relevant for datasets submitted to or hosted by the Cancer Imaging Archive (TCIA), where NCI Cancer Imaging Program/TCIA implementation practices address imaging data containing potentially reconstructable facial anatomy. While intended to protect patient privacy and strengthen public trust, defacing can distort craniofacial geometry and alter image statistics, potentially compromising the fidelity and reproducibility of artificial intelligence (AI) models trained on such data. Existing studies primarily validate visual anonymization quality, but few have quantified its downstream impact on deep learning-based medical imaging tasks. Understanding this privacy-utility trade-off is crucial for responsible data sharing and compliant AI development. METHODS:We systematically evaluated three representative defacing algorithms, two invasive (QuickShear and Py-Deface) and one less destructive, facial replacement (mri_reface), across MRI and CT datasets from 600 subjects spanning three institutions. Model performance was assessed on three clinically relevant applications: (1) brain segmentation and Evans ratio biomarker quantification in normal pressure hydrocephalus (NPH) MRI using SLANT and FreeSurfer; (2) representative-slice selection and diagnostic reasoning for brain tumour MRI using vision-language models (VLMs); and (3) automated emergency head CT report generation using a fine-tuned Otter-based vision-language model. Each method's impact was quantified using Dice similarity, correlation metrics, reasoning accuracy, and natural-language generation scores (BLEU, METEOR, ROUGE, CIDEr). FINDINGS:Invasive algorithms caused significant degradation across all tasks. QuickShear reduced mean Dice scores by up to 9% and introduced 14-19% failure rates during quality control, while PyDeface induced smaller but measurable performance losses. mri_reface maintained 100% success without any failures and achieved segmentation, diagnostic, and report-generation accuracy within 3-5% of the original data. Evans ratio distributions remained statistically consistent between mri_reface and original images (p > 0.05), whereas invasive methods introduced broader variance. Across all VLM tasks, mri_reface preserved high correlation with radiologist-selected slices (r = 0.979) and stable report-generation quality (BLEU-4 = 0.11 ± 0.06 vs. 0.12 ± 0.07 for original). INTERPRETATION:Facial anonymization introduces a measurable privacy-utility trade-off that must be explicitly considered in the design of AI-ready medical imaging datasets. Invasive defacing compromises geometric and statistical integrity, reducing downstream model accuracy even outside facial regions. Facial replacement anonymization methods, such as mri_reface, effectively reconcile patient privacy with reproducibility, offering a practical path to NIH-compliant open data. Future regulatory and institutional policies should integrate quantitative privacy-utility assessment and mandate transparent reporting of anonymization pipelines to ensure that shared imaging data remain both ethically safe and scientifically valid under emerging digital health frameworks. FUNDING:This work was partially supported by the American Heart Association (Award No. 25IPA1454088), the National Institutes of Health (Award No. 1R03CA286693-01A1 and Award No. 1R01CA291826-01A1), the U.S. Department of Defense (Award No. HT94252510807), and the National Science Foundation (Award No. 2545071).

OBJECTIVE:To determine whether retinal thinning in neuronal intranuclear inclusion disease (NIID) is associated with multilevel abnormalities across the visual system and with clinical severity. METHODS:Forty patients with NIID and 40 healthy controls underwent optical coherence tomography to measure peripapillary retinal nerve fiber layer (RNFL) and macular ganglion cell complex (GCC) thickness. Among patients with NIID, 37 underwent structural MRI for quantification of visual-region volumes and 30 underwent resting-state functional MRI for graph-theoretical assessment of visual-network topology. Cognitive function and activities of daily living were evaluated in the NIID cohort. Partial correlation and exploratory mediation analyses were used to examine associations among retinal, neuroimaging, and clinical measures. RESULTS:Patients with NIID showed diffuse thinning of the RNFL and GCC relative to controls, with mean GCC showing the best discrimination between groups. Thinner retinal measures were associated with poorer cognition, worse daily function, and lower mean cortical thickness. Structural MRI identified volume abnormalities in selected visual-system regions, particularly the lateral geniculate nucleus, early visual cortex, and dorsal/parietal regions, and retinal thickness correlated positively with the volumes of several visual regions. Poorer daily function was associated with a lower clustering coefficient of the visual network. Left V3d, the dorsal part of area V3 in the occipital visual cortex, partially mediated the association between retinal thinning and functional impairment. CONCLUSIONS:These findings support coordinated retina-brain involvement in NIID across retinal, structural, and network levels, and identify OCT-derived RNFL and GCC thickness as accessible, noninvasive candidate biomarkers of disease severity.

BACKGROUND/OBJECTIVES:Characterizing spinal cord multiple sclerosis (MS) lesions in MRI is critical for diagnosis, monitoring, and treatment evaluation. However, current automated approaches for lesion detection and segmentation are typically designed for specific MRI contrasts or acquisition sites, limiting their generalizability in real-world clinical settings where imaging protocols vary widely. This work proposes a robust multi-site, multi-contrast segmentation framework for spinal cord lesions. METHODS:The segmentation model was trained and evaluated on a large-scale dataset comprising 4428 annotated images from 1849 persons with MS across 23 imaging centers, encompassing six MRI contrasts (T1w, T2w, T2*w, PSIR, STIR, and UNIT1) acquired at 1.5 tesla (T), 3 T, and 7 T. RESULTS:Likert-type assessment performed by neuroradiologist ratings demonstrated superior generalization of the model compared to existing contrast-specific pipelines (p < 0.01). Additional experiments evaluated robustness across spinal levels, acquisition resolutions, binarization thresholds, and quantitative evaluation on external labeled datasets. CONCLUSIONS:The proposed model can achieve accurate and reliable spinal cord MS lesion segmentation across heterogeneous MRI data, addressing a key barrier to clinical translation. The model is available in the Spinal Cord Toolbox v7.2 and higher.Code repository: https://github.com/ivadomed/seg-sc-ms-lesion-multicontrast.

The precise and comprehensive diagnosis of complex brain disorders relies on non-invasive computed tomography (CT) and magnetic resonance imaging (MRI) in conjunction with multi-modal clinical information. Here, we present Brainfound, a multi-modal foundation model for brain medical imaging that integrates image-text contrastive learning with a diffusion-based generative framework. The model was pre-trained on more than 3 million brain CT slices and 7 million brain MRI slices paired with clinical reports. In multi-center evaluations, Brainfound demonstrates state-of-the-art performance across seven tasks, including brain disease diagnosis, lesion segmentation, MRI enhancement, cross-modality translation, automatic report generation, zero-shot disease classification, and human-AI dialogue. It substantially outperforms leading models in automated report generation and clinical question answering for brain imaging, and its performance approaches that of expert physicians. These findings highlight the potential of Brainfound for accelerating diagnosis, support treatment decisions, and advance human-in-the-loop brain health care.

Cerebral vasculature plays a critical role in brain function. Accurate characterization of its normative organization and distribution is essential for identifying vascular abnormalities. However, existing cerebrovascular templates do not adequately account for age-related anatomical variability and typically rely on structural-image-driven registration without explicitly leveraging vascular information for alignment. To address these limitations, we present a set of open, age-specific cerebrovascular templates constructed from time-of-flight magnetic resonance angiography (TOF-MRA) data of 1,288 healthy adults spanning the adult lifespan (19-92 years). Fine-scale vascular structures were automatically segmented using CereVessPro and registered through a novel two-stage framework that integrates age-specific structural alignment with vessel-guided refinement. The resulting resources comprise a series of openly accessible maps, including age-specific MRA templates, vessel probability maps, vessel density maps, and vessel radius maps from young to older adulthood. This framework is designed to improve vascular correspondence across subjects by explicitly addressing age-dependent anatomical differences and incorporating vascular features during alignment. Validation analyses revealed biologically plausible age-related changes in vascular morphology and vessel calibre, particularly in major cerebral arteries. Together, these data provide an age-resolved reference for cerebrovascular mapping, improving vascular alignment and supporting more reliable group-level analyses. This resource may facilitate more sensitive detection of vascular abnormalities in both research and clinical contexts.

The RANO criteria remain the cornerstone for evaluating adult gliomas; however, they often fail in spinal cord gliomas due to anatomical constraints, molecular heterogeneity, and distinct biological behavior. Firstly, central nervous system (CNS) dissemination is a hallmark feature of spinal cord gliomas and a critical driver of mortality, distinguishing them from their brain counterparts where such spread is rare. ‌Moreover, T1-weighted contrast-enhanced MRI typically reveals non-enhancing primary lesions‌, a characteristic feature of ‌H3 K27M-mutant diffuse midline gliomas (DMG)‌ that constitute over 40% of spinal cord gliomas, while ‌non-contrast T2-weighted imaging demonstrates sensitivity and reproducibility. Given the narrow and elongated anatomy of the spinal cord and the unique surgical strategy required for spinal cord gliomas, we advocate for volumetric assessment as the primary evaluation method, utilizing millimeters (mm) rather than centimeters (cm) as the measurement unit, and consider all visually identifiable lesions as measurable. Furthermore, the spinal cord exhibits super-functional integration across motor, sensory, and reflex pathways, thereby accentuating the importance of clinical manifestations and neurological functional assessment in accurately and promptly tracking disease progression. Our objective is to develop the specialized response assessment criteria for spinal cord gliomas to serve clinical trials.

Graphical AbstractFor image description, please refer to the figure legend and surrounding text.

Abstract Cerebral oxygen extraction fraction (OEF) reflects the balance between cerebral oxygen delivery and metabolic demand, but its normative evolution across the human lifespan remains unknown. Here we used rapid, non-contrast TRUST MRI to establish a multisite normative model of global cerebral OEF in 2,025 healthy individuals aged 0-93 years from 17 imaging sites. OEF increased from the neonatal period to middle adulthood, followed by a slower rise and plateau in later life, with the fastest change occurring during early development and no significant sex differences. Individual OEF deviation scores were associated with vascular risk burden in healthy adults. Applying the model to 885 patients revealed disease-related OEF alterations, including positive deviations in pediatric obstructive sleep apnea, autoimmune disorders, brain tumors, mild cognitive impairment and dementia. OEF deviation further tracked tumor grade and Ki-67 proliferation. These findings establish lifespan OEF charting as a scalable framework for individualized physiological neuroimaging.

BACKGROUND:Pleomorphic xanthoastrocytoma (PXA) carries highly variable recurrence risk, yet reliable preoperative stratification tools are lacking. We developed a recurrence prediction model using Visually Accessible Rembrandt Images (VASARI) MRI features in the largest standardized PXA imaging cohort reported to date. MATERIALS AND METHODS:This retrospective dual-center study included an imaging cohort of 155 patients with histopathologically confirmed PXA. Of these, 116 patients with available follow-up and complete data for the variables required for survival modeling comprised the survival cohort. Twenty-seven VASARI features were independently assessed by two radiologists. Univariate and multivariable Cox proportional hazards regression identified independent recurrence predictors. Model discrimination was quantified by Harrell's concordance index (C-index) with bootstrap internal validation (1000 iterations). Calibration was explored using tertile-based calibration plots at 24 and 36 months and the integrated Brier score. RESULTS:Tumor recurrence occurred in 27 of 116 patients (23.3%) in the survival cohort. Three imaging features were independently associated with recurrence: diffusion restriction (HR=5.10; 95% CI, 2.143-12.148; P < .001), midline crossing (HR=3.82; 95% CI, 1.662-8.769; P = .002), and mixed cystic-solid non-classic morphology (HR=3.42; 95% CI, 1.353-8.625; P = .009). The model achieved good discrimination (apparent C-index=0.834; optimism-corrected C-index=0.79 [95% CI, 0.70-0.86]). Exploratory calibration showed approximate agreement between predicted and observed outcomes, although estimates were imprecise because of the limited number of events. The integrated Brier score was 0.126 compared with 0.179 for the null model. CONCLUSIONS:A three-feature VASARI-based model showed potential for preoperative recurrence risk stratification in PXA without specialized software or postprocessing. These findings are hypothesis-generating and require independent external validation before clinical application.

Purpose To develop and validate a deep neural network that simultaneously segments brain tumors and anatomic structures, regardless of the contrast and resolution of the input scans, and can effortlessly adapt to unseen modalities. Materials and Methods The authors included various MRI scans from patients with and without brain tumors from four different datasets. Patient data were divided into a training set and a test set. The authors' method, TumorSynth, combines a Bayesian generative model and a deep learning segmentation model. The generative model creates paired synthetic labels and images with simulated tumors and brain tissues, providing a rich dataset for training the segmentation model. The authors quantitatively compared its performance with that of other widely used methods by calculating Dice similarity coefficients (DSCs). Results A total of 1971 patients with and without tumors were included in the study (training set, n = 351 patients; test set, n = 1620 patients). The median DSCs for segmentation (authors' method vs reference standard) were 0.89 (IQR, 0.83-0.95; P < .001) for the unaffected brain volume and 0.89 (IQR, 0.84-0.94; P < .001) for the tumor region. There were no differences in parcellation performance when an MRI sequence was missing (P = .07). In cross-modality validation, the authors' method achieved DSC values of 0.88 for apparent diffusion coefficient, 0.85 for diffusion-weighted imaging, 0.80 for susceptibility-weighted imaging, and 0.79 for fractional anisotropy images. The authors observed a 4% false-positive rate when processing tumor-free MR images. Conclusion The authors developed a deep neural network for brain tumor and tissue segmentation, validated its performance across standard structural MRI sequences, and determined its generalizability to unseen data. Keywords: Segmentation, Neuro-Oncology, CNS, Deep Learning, Neurosurgery Supplemental material is available online for this article. © RSNA, 2026.

BACKGROUND:The T2-FLAIR mismatch (T2FM) sign is a highly specific imaging biomarker for isocitrate dehydrogenase (IDH)-mutant astrocytomas; however, its strict definition limits clinical applicability. PURPOSE:To determine whether an expanded T2FM (eT2FM) phenotype could improve identification of IDH-mutant astrocytomas and capture prognostic and biological information. STUDY TYPE:Retrospective. POPULATION:349 patients (50.14% male; mean age, 41.08 ± 10.40 years) with IDH-mutant astrocytomas. FIELD STRENGTH/SEQUENCE:3 T, T1-weighted gradient-echo- or spin-echo-based images, and T2-weighted, T2 fluid-attenuated inversion recovery (FLAIR), and contrast-enhanced T1-weighted images mainly acquired using spin-echo-based techniques. ASSESSMENT:The eT2FM phenotype was defined by incorporating spatially heterogeneous T2 FLAIR signals beyond the classic T2FM sign. Survival differences were compared between eT2FM and non-eT2FM tumors. Survival associations were evaluated. Metabolomic profiling was explored. STATISTICAL TESTS:Kaplan-Meier analysis, weighted log-rank test, Cox regression, and metabolomic analyses. Significance was set at p < 0.05. RESULTS:The eT2FM phenotype was identified in 116 of 349 patients (33.24%; mean age, 39.08 ± 9.51 years), exceeding classic T2FM (50/349, 14.33%; mean age, 36.40 ± 9.10 years). Kaplan-Meier analysis showed more favorable overall survival for eT2FM than non-eT2FM tumors, with restricted mean survival time (RMST) up to 36 months of 35.18 and 32.08 months, respectively (absolute difference, 3.10 months). Similar findings were observed in the external validation cohort, in which eT2FM phenotype was identified in 48 of 146 patients (32.88%; mean age, 37.77 ± 11.05 years). Kaplan-Meier analysis also showed more favorable overall survival for eT2FM than non-eT2FM tumors, with RMSTs of 36.00 and 33.81 months, respectively (absolute difference, 2.19 months). The eT2FM phenotype was associated with better prognosis in univariable Cox analysis (HR, 0.265; 95% CI, 0.120-0.585), but not independently in multivariable analysis (HR, 2.014; 95% CI, 0.380-10.673; p = 0.41). Exploratory metabolomics identified 1078 differentially abundant features between groups. DATA CONCLUSION:The eT2FM phenotype extends the clinical utility of classic T2FM sign and delineates a subgroup of IDH-mutant astrocytomas with favorable clinicopathologic features. LEVEL OF EVIDENCE: 4: TECHNICAL EFFICACY:Stage 4.

BACKGROUND AND OBJECTIVES:Longitudinally extensive transverse myelitis (LETM) is a core feature of neuromyelitis optica spectrum disorder (NMOSD), which leads to spinal cord atrophy. We aim to determine annual spinal cord atrophy rates and influencing factors in NMOSD. METHODS:Spinal cord structural metrics, including cross-sectional area (CSA), anteroposterior diameter (AP), and right-left diameter (RL), were calculated using the Spinal Cord Toolbox from brain 3D sagittal T1-weighted images covering the C1-C3 levels. Normative references were constructed based on healthy participants and used to derive the deviation scores (Z-scores). Annualized atrophy rates were estimated using a linear mixed-effects model. Associations with clinical measures, rituximab treatment, and neurofilament light chain (NfL) and glial fibrillary acidic protein (GFAP) were assessed using partial Pearson correlation, multivariate linear regression, and linear mixed-effects models. RESULTS:In this longitudinal study, 72 patients with NMOSD (233 MRI scans; mean age 40.0 years; 84.7% female) from the CLUE cohort and 2,162 healthy participants were included. In patients with NMOSD, annual atrophy rates were -0.27/y for CSA (95% CI -0.35 to -0.20, p < 0.001), -0.23/y for AP (95% CI -0.33 to -0.13, p < 0.001), and -0.14/y for RL (95% CI -0.20 to -0.09, p < 0.001). A faster AP atrophy rate was associated with longer T2 lesion length and a higher number of relapses (r = -0.35/-0.25, p < 0.05). Prolonged interval from initial LETM to baseline MRI was independently associated with a slower CSA atrophy rate (β = 4.62 × 10-5; 95% CI 0.25 to 9.00 × 10-5, p = 0.043), while older age (β = -25.17 × 10-5; 95% CI -39.12 to -11.22 × 10-5, p < 0.001) and lesion length (β = -55.66 × 10-5; 95% CI -104.50 to -6.83 × 10-5, p = 0.029) were related to faster RL atrophy. Rituximab was associated with slower RL atrophy compared with other immunosuppressive therapies (β = 0.02; 95% CI 0.01-0.05; p = 0.044). Elevated serum GFAP levels were correlated with accelerated CSA (r = -0.42/-0.45), AP (r = -0.49/-0.45), and RL (r = -0.37/-0.44) atrophy at baseline/follow-up (all p < 0.01). Increased NfL levels were correlated with faster CSA and RL atrophy at follow-up (p = 0.009, p = 0.031). DISCUSSION:Spinal cord atrophy in NMOSD progresses annually and is influenced by disease activity and duration. Rituximab was associated with attenuated progression of spinal cord atrophy. Moreover, serum NfL and GFAP may serve as potential biomarkers.

To develop and validate a deep learning system (DLS) model predicting hematoma expansion (HE) based on non-contrast (NC) CT and a score combining with clinical variables. The multicenter retrospective dataset (R), the multicenter prospective dataset (P1), and the single-center prospective dataset (P2) enrolled 2350, 460, and 96 intracerebral hemorrhage (ICH) patients for analysis, respectively. The DLS model was developed, validated, and tested in R-development (R-dev), R-validation (R-val), and P1, respectively. After exploring clinical predictors of HE using multivariable logistic regression on P1-development (P1-dev), a five-point score “ARCHES” (Ai-Reinforced intraCerebral Hemorrhage hematoma Expansion Score) combining clinical predictors with the DLS model was created. We compared the discrimination of the ARCHES, DLS, with other models using the receiver operating characteristic (ROC) and DeLong test. The areas under the curve (AUC) of the DLS model were 0.781 (95

Magnetic resonance imaging (MRI) radiomics has shown promise in glioma grading and isocitrate dehydrogenase (IDH) mutation prediction, but traditional whole-tumor approaches overlook intratumoral heterogeneity, limiting diagnostic accuracy and interpretability. This study aims to explore cellularity habitat-based MRI radiomics for precise grading and IDH mutation status prediction in adult-type diffuse glioma (ADG). A total of 625 ADG patients were retrospectively collected. Whole-tumor volumes of interest (VOIs) were delineated on four conventional MRI sequences (T1WI, T2WI, T2-FLAIR, and CE-T1WI) and segmented into three cellularity habitats using apparent diffusion coefficient (ADC)-based K-means clustering: H1 (low ADC), H2 (medium ADC), and H3 (high ADC). Radiomic features were extracted from individual and combined habitats, and predictive models were developed using a disentangled-learning-based multi-sequence fusion network (DMSFN). Performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy (ACC), sensitivity (SEN), and specificity (SPE). The optimal habitats for ADG grading (Grade 2 vs. Grade 3 + 4, Grade 2 + 3 vs. Grade 4) and IDH prediction were H1 + 2, H1 + 2, and H2 + 3, respectively. Combining T1WI, CE-T1WI, and T2-FLAIR sequences yielded the highest AUCs of 0.9360, 0.9605, and 0.8721 in the training set, and 0.8070, 0.8236, and 0.8180 in the independent test set. Shapley Additive exPlanation (SHAP) analysis identified key radiomic features contributing to model predictions, with CE-T1WI features consistently demonstrating high discriminative power. Integrating ADC-derived cellularity habitats with MRI radiomics significantly improves the accuracy and biological interpretability of ADG grading and IDH mutation status prediction, offering a robust, non-invasive approach for glioma characterization. Retrospectively registered.

Synthesizing missing modalities in multi-modal magnetic resonance imaging (MRI) is vital for ensuring diagnostic completeness, particularly when full acquisitions are infeasible due to time constraints, motion artifacts, and patient tolerance. Recent unified synthesis models have enabled flexible synthesis tasks by accommodating various input-output configurations. However, their training and evaluation are typically restricted to a single dataset, limiting their generalizability across diverse clinical datasets and impeding practical deployment. To address this limitation, we propose PMM-Synth, a personalized MRI synthesis framework that not only supports various synthesis tasks but also generalizes effectively across heterogeneous datasets. PMM-Synth is jointly trained on multiple multi-modal MRI datasets that differ in modality coverage, disease types, and intensity distributions. It achieves cross-dataset generalization through three core innovations: a Personalized Feature Modulation module that dynamically adapts feature representations based on dataset identifier to mitigate the impact of distributional shifts; a Modality-Consistent Batch Scheduler that facilitates stable and efficient batch training under inconsistent modality conditions; and a selective supervision loss to ensure effective learning when ground truth modalities are partially missing. Evaluated on four clinical multi-modal MRI datasets, PMM-Synth consistently outperforms state-of-the-art methods in both one-to-one and many-to-one synthesis tasks, achieving superior PSNR and SSIM scores. Qualitative results further demonstrate improved preservation of anatomical structures and pathological details. Additionally, downstream tumor segmentation and radiological reporting studies suggest that PMM-Synth holds potential for supporting reliable diagnosis under real-world modality-missing scenarios.

To develop and validate a deep learning (DL) model based on conventional MRI for preoperative differentiation of brain metastasis (BM), glioblastoma (GBM), and primary central nervous system lymphoma (PCNSL), and to evaluate its diagnostic utility as a decision-support tool for radiologists. This multicenter retrospective study included 1,298 patients with histopathologically confirmed BM (n = 426), GBM (n = 456), or PCNSL (n = 416). A 2.5D ResNet50-based DL model was developed using axial T2-weighted and contrast-enhanced T1-weighted MRI. A clinical model (CM) incorporating demographic and semantic imaging features was constructed using multivariable logistic regression. Model performance was evaluated in independent internal and external test sets using the area under the receiver operating characteristic curve (AUC) and accuracy with 95

Background:Radiogenomics allows identification of radiological biomarkers for genomic phenotypes. In glioblastoma, these biomarkers could potentially complement patient stratification strategies. We aim to develop and analytically validate radiological biomarkers that capture immune cell signatures within IDH-wildtype glioblastoma microenvironment using radiogenomic analysis. Methods:This was a retrospective multicenter study using curated open-access anonymized imaging and genomic data from TCGA-GBM, CPTAC, IvyGAP, REMBRANDT, and CGGA datasets. Imaging data consisted of MRI-based radiomic features extracted from necrotic core, enhancing and edema regions of deep learning-based autosegmented tumors. Radiomic feature selections were performed using nested cross-validated LASSO. Support vector machine and ensemble models were trained using seventeen immune and cell-specific score labels extracted from deconvoluted transcriptomic data using pan-cancer and glioblastoma immune signature matrices as reference standards. Seventeen classifier models trained in 3 cross-cohort strategies were validated on 3 held-out datasets assessing stability and generalizability. Results:One-hundred-and-seventy-six patients were included in the study. The immune-related radiomic signatures obtained after feature selection were shape, first order and higher order radiomic features. Models predicting macrophage subtype immune signature showed stable mean performance on balanced accuracy (0.67) and precision (0.89) metrics for 3 independent holdout datasets with ensemble model outperforming support vector machine model. Conclusion:Radiogenomic models noninvasively predicted the macrophage subtype M0 immune signature in IDH-wildtype glioblastoma. These biomarkers have the potential to stratify patients for immunotherapy within prospective glioblastoma clinical trials.