2.5D HAU-Net with gated spatial attention for automatic hippocampus segmentation in MRI.

Reliable quantification of perivascular spaces (PVS) in the basal ganglia (BG) is of growing interest for understanding the glymphatic system but remains challenging in infants. To develop an automated deep learning method for BG and BG-PVS segmentation in infant brain MRI using an anatomy-informed pseudo-labeling approach. Retrospective, multi-cohort technical development, and validation study. Three cohorts: 150 neonates from the Developing Human Connectome Project (dHCP, 37-44 weeks of gestational age (GA); 76 males, 74 females), 133 infants from the Baby Connectome Project (BCP; ≤ 24 months; 70 males, 63 females) and 70 infants from an in-house dataset (30-41 weeks of GA; 36 males, 34 females). Manual ground-truth labels were generated by a trained researcher (dHCP, n = 150; BCP, n = 8; in-house, n = 10) and validated by a radiologist with 15 years of experience. Data included 3 T MRI with T1- and T2-weighted sequences: dHCP (inversion recovery turbo spin-echo [IR-TSE] and turbo spin-echo [TSE]), BCP (magnetization-prepared rapid gradient-echo [MPRAGE] and TSE), and in-house (MPRAGE and variable-flip-angle TSE). The proposed approach was compared with alternative automated approaches trained with different labeling strategies. Training/validation/test splits were 100/25/25 (dHCP), 100/25/8 (BCP), and 50/10/10 (in-house). Dice similarity coefficient (DSC), recall, positive predictive value, and Hausdorff distance were calculated for BG and BG-PVS quantification. Statistical significance was assessed using Wilcoxon signed-rank tests (p < 0.05), and quantification agreement was evaluated using Pearson's correlation, intraclass correlation coefficient (ICC), and mean absolute error (MAE). The proposed method improved accuracy (dHCP: BG DSC = 0.91 ± 0.03 and BG-PVS DSC = 0.78 ± 0.09; external datasets with fine-tuning: BG DSC = 0.86-0.89) and high agreement in PVS quantification with reference measurements (r = 0.90-0.99, ICC ≥ 0.96, MAE = 0.10). The proposed method seems to enable robust and annotation-efficient BG and BG-PVS segmentation in infants. 3. 1.

CGMC-Net: cross-layer guided and multi-branch coupling network for brain tumor segmentation of thick-slice MRI

To measure the morphology of the pelvic floor muscle and levator ani hiatus (LAH) using 3D MRI segmentation in females with (SUI) and to assess correlations with age, body mass index (BMI), and parity. This study includes 80 patients with SUI who underwent 3.0 T pelvic MRI. Patients were grouped based on age, BMI, and parity. Manual segmentation and 3D modeling of the pelvic floor muscle and LAH were performed to get volumetric and dimensional measurements. Statistical analysis included Kruskal-Wallis, Mann-Witney, and post hoc testing to evaluate group variations. The cohort included patients from 20 to 90 years old (mean 53.9 ± 11.7), with a mean BMI of 25.3 ± 3.4 and a mean parity of 1.85 ± 1.03. Significant variations in LAM surface (H(2) = 6.650, p = 0.036), flatness (H(2) = 6.471, p = 0.039), median intensity (H(2) = 8.218, p = 0.016), and curvature mean (H(2) = 7.382, p = 0.025) among age groups. The greatest LAM surface rankings were found in the 50-65 age range, whereas the highest LAM flatness ranks were found in those over 65 years old. BMI demonstrated a limited association, with overweight/obese participants exhibiting 11.3% larger LAM surface area compared to normal/underweight individuals (p = 0.044, r = 0.225). No significant associations were found between parity and any pelvic floor morphometric parameters (all p > 0.05). This study identifies age-dependent morphological changes in levator ani muscle architecture as the primary determinant of pelvic floor structural adaptation in women with stress urinary incontinence. These quantitative shape-based parameters offer novel diagnostic and therapeutic insights, advancing precision medicine approaches in urogynecological assessment and management.

Recent advances in vision-language models and LLMs have introduced contextual anatomical reasoning into brain MRI segmentation. However, the field still suffers from a fundamental limitation: the absence of a unified anatomical definition of the structures being segmented. Existing datasets rely on labels produced by heterogeneous manual workflows, often lacking explicit anatomical criteria or consistent annotation standards. As a result, models learn and evaluate within isolated labeling systems, limiting cross-model comparison and valid anatomical measurements. To address these challenges, we introduce NeuroLangSeg, a language-guided framework that enforces a consistent anatomical protocol for subcortical segmentation. A key component of the framework is an anatomical-linguistic evaluator that acts as a training discriminator, encouraging the model to produce outputs by assessing shape characteristics, protocol-defined spatial relationships, and age- and sex-adjusted volumetric norms. Building upon this constraint, NeuroLangSeg integrates a pretrained image encoder with protocol-aligned anatomical prompts and a masked pseudo-labeling strategy, enabling data-efficient and interpretable learning under limited supervision. Together, these components yield anatomically consistent segmentations and support subject-level reporting grounded in a unified anatomical standard. Evaluation across diverse MRI datasets-including comparisons with state-of-the-art models-shows that NeuroLangSeg achieves +4.1 DSC / +8.0 NSD in in-site settings and +3.6 DSC / +14.5 NSD in cross-site generalization over the average baseline, enabled by its LLM-visual integration, while delivering anatomically verifiable predictions suitable for both research and clinical use. GitHub: https://github.com/jlliu2001/SAT_MPL.

HepaLite-FAU: A lightweight U-Net with SE-enhanced axial factorized blocks for cirrhotic liver MRI segmentation and severity assessment

A hybrid CNN–transformer network with cross-level multi-scale fusion and hierarchical attention for brain tumour MRI segmentation

Purpose To develop and systematically evaluate an iterative training approach, termed the expert-guided annotation loop, for efficient reference standard medical image segmentation, including assessment of two sample-selection strategies and real-world clinical implementation. Materials and Methods This retrospective study included ten datasets comprising 1948 CT or MRI examinations from autosomal dominant polycystic kidney disease, prostate cancer, uveal melanoma, thyroid eye disease, and non-small cell lung cancer patients. nnU-Net segmentation models were iteratively trained using an expert-guided annotation loop with random or active learning-based sample selection. In each iteration, additional samples were added to the training set, and model-generated presegmentations were corrected by expert radiologists to create reference standard annotations. Expert time required for manual segmentation versus presegmentations correction was measured. Model performance and efficiency were assessed using nonparametric tests, and cost savings were estimated for kidney and tumor segmentation using probabilistic sensitivity analysis. Feasibility of end-to-end no-code implementation was evaluated. Results Fifty-seven segmentation models were trained and evaluated. Final model Dice scores ranged from 0.67-0.97 for organ segmentation and from 0.64-0.69 for lung tumor segmentation across internal and external test sets. Maximum expert time savings were 90.3% for kidney and 48.2% for tumor segmentation (P < .001 and P = .003), corresponding to estimated per-examination cost savings of $14.30[95% CI: $5.94, $26.87] and $5.63[95% CI: $-7.26, $26.09], respectively. No-code execution of the expert-guided annotation loop was feasible. Conclusion The expert-guided annotation loop reduced expert annotation time and enabled estimated cost savings while producing high-quality reference standard CT and MRI segmentations. The no-code workflow was implemented in a clinical environment. © RSNA, 2026.

Accurate glioma MRI segmentation is critical for clinical decision-making, yet voxel-wise annotation of volumetric MR images is expensive and time-consuming. We aim to maximize segmentation quality while reducing annotation cost through selective labeling. We develop a tumor-focused active learning AL framework that mirrors real-world annotation workflows and combines three complementary techniques: (i) a patient-specific, tumor-focused 40-slice window to preserve context while reducing computational requirements; (ii) adiverseuncertainty committee spanning softmax entropy, margin, least confidence, variation ratio, and Bayesian AL by disagreement (BALD), where variation ratio and BALD are estimated via lightweight Monte Carlo dropout to capture epistemic uncertainty; and (iii)multi-criteria rank-level fusionvia Borda count to robustly aggregate uncertainty signals. In particular, we compute uncertainty both at theslicelevel and at theregion-of-interest (ROI)level by summarizing pixel-wise uncertainty within tumor ROIs, then fuse all techniques and granularities with Borda rank to form a single acquisition order. A segmentation network is trained on an initial small labeled subset and then iteratively selects additional slices for annotation. Evaluation on UCSF-PDGM and BraTS 2019 under realistic annotation budgets shows strong label efficiency and robustness: on UCSF-PDGM, near-full performance is achieved with only≈16%of the training data; on BraTS 2019, using≈32%of the data yields performance comparable to models trained on all data, even surpassing the full-data model. Overall, the proposed framework combines the tumor-focused window with MC-dropout-enhanced,ROI-awareuncertainty and Borda fusion to achieve accurate segmentation while aligning with real-world curation workflows.

Uterine fibroids represent one of the most common gynecological diseases and accurate 3D reconstruction is a prerequisite for clinical diagnosis, surgical planning, and treatment evaluation. Routine clinical protocols typically involve a set of orthogonal sagittal, coronal, and transverse MRI scans to assess morphology. These images usually have anisotropic voxels with sparse 3D coverage and existing assessment schemes are often limited to different 2D views based on planar segmentation. Moreover, models trained on data from individual centers often fail to generalize to wider datasets. Accurate and consistent 3D reconstruction for quantitative uterine fibroid analysis, particularly with unsupervised domain adaptation (UDA), is an unmet clinical need. This paper proposes the Foundation Model-Guided Adaptive Segmentation (FGAS) framework for automatic annotation-free multi-planar uterine fibroid MRI segmentation. FGAS uses anatomical priors for pseudo-label optimization, and integrates multi-view consistency constraints and connected component control to reduce planar dependency and suppress false positives. Extensive experiments on clinical datasets demonstrate the superior performance of FGAS, improving the Dice similarity coefficient from 42.8% of baseline models to 70.6%, outperforming existing state-of-the-art UDA and multi-plane methods. These results indicate that FGAS can achieve robust, high-accuracy automated reconstruction for annotation-free multi-view and cross-domain image analysis.

Accurate MRI-based quantification of abdominal adipose tissue is critical for metabolic risk assessment but is limited by labor-intensive manual segmentation and the extensive labeled-data dependency of deep learning models. We introduce Dynamic Fuzzy-Gaussian Modeling (DynFGM), a fully automated, unsupervised framework for adipose tissue segmentation designed to operate without requiring training data, expert annotations, or anatomical priors. DynFGM was developed and validated on 776 abdominal MRI scans, using a benchmark cohort (n = 20) with expert ground truth segmentations and a large validation cohort (n = 756). The pipeline dynamically adapts its complexity for each MRI slice by using image intensity kurtosis to select the optimal number of tissue clusters. A fuzzy C-means (FCM) algorithm then initializes a Gaussian mixture model (GMM) for segmentation, providing a mathematically interpretable alternative to black-box neural networks. Finally, a radial distance transform with an adaptive cutoff differentiates subcutaneous (SAT) from visceral adipose tissue (VAT). Performance was evaluated against the ground truth using dice similarity coefficient (DSC) and intraclass correlation coefficient (ICC). DynFGM achieved strong spatial agreement with expert annotations (mean DSC: 0.94) and high volumetric reliability (ICC: 0.82-0.97), comparable to reported inter-expert variability. The framework reduced mean absolute volumetric error by 92.6% compared to standard FCM (482.2 cm3 vs. 6547.5 cm3). On the large validation cohort (n = 756), the method demonstrated operational stability, producing physiologically plausible adipose distributions with a low technical failure rate (3.0%). Furthermore,the computational throughput averaged 13.6s per participant on standard CPU (Intel® Core™ i9, 3.0GHz) hardware. DynFGM provides an interpretable and data-efficient approach for abdominal adipose tissue phenotyping, offering an alternative to supervised deep learning in settings where labeled data are limited or unavailable. By bridging the gap between manual segmentation and labeled-data-dependent AI, this unsupervised framework offers a scalable tool for population-level research and may serve as an automated labeling tool to facilitate future model development.

The hypothalamus is a central regulator of neuroendocrine function and social behavior, yet its internal organization has remained difficult to examine in vivo in human neurodevelopmental research. Based on neuroendocrine models, we hypothesized that anterior hypothalamic subunits enriched in magnocellular neurosecretory nuclei-approximating the paraventricular (PVN) and supraoptic (SON) regions-form a structurally coherent axis linked to dimensional traits associated with autism spectrum disorder (ASD) and attention-deficit/hyperactivity disorder (ADHD). To test this hypothesis, we applied a convolutional neural network-based automated segmentation framework to high-resolution T1-weighted MRI data from a non-clinical adult sample (n = 23), enabling volumetric quantification of ten hypothalamic subunits. After adjustment for biological sex, age, and total intracranial volume, PVN- and SON-associated volumes showed a significant positive structural covariance (r = 0.51, p = 0.022), consistent with a coherent anterior magnocellular-rich network. Reduced SON-associated volume was significantly associated with higher autistic traits (Autism-Spectrum Quotient: r = - 0.45, p = 0.046) and greater ADHD symptom severity (CAARS: r = - 0.59, p = 0.006). Bootstrapped path analyses (5,000 iterations) further supported this network organization, revealing a robust association between PVN- and SON-associated volumes (β = 0.51, p = 0.013) and a significant indirect association between PVN-associated volume and ADHD symptoms via SON-associated volume (β = - 0.29, p = 0.006), with a marginal indirect association observed for autistic traits. Together, these preliminary findings provide in vivo evidence linking anterior hypothalamic structure to dimensional neurodevelopmental traits and highlight a potential neuroendocrine pathway relevant to psychiatric vulnerability.

Accurate segmentation of cardiac MRI is essential for assessment of cardiac function through biomarkers such as the left and right ventricular ejection fraction (LVEF, RVEF). Although AI methods have achieved high average segmentation accuracy, the precision of biomarkers for individual patients - quantified by estimation variance, remains critical for reliable diagnosis. Calibrated biomarkers, whose uncertainty accurately reflects the true variability, are highly desirable. However, existing evaluations predominantly focus on population-level segmentation accuracy, leaving biomarker-level uncertainty and calibration largely underexplored. Intrinsic anatomical ambiguity and annotation variability are major sources of biomarker variability and cannot be fully eliminated, even when training on a single annotation set. To address this, we propose a probabilistic segmentation framework that explicitly models aleatoric uncertainty with the goal of improving calibration in the biomarker space. The framework disentangles two key sources of uncertainty: (1) detection uncertainty, arising from ambiguous inclusion of basal or apical slices in 2D cardiac MRI, modeled via objectness probabilities; and (2) contour uncertainty, reflecting variability in ventricular boundary delineation, modeled through mean-variance regression of elliptic Fourier descriptors, a compact representation of closed contours. By propagating these uncertainties to derived biomarkers, the proposed method produces more informative and better-calibrated confidence estimates for ejection fraction. Compared to conventional pixel-wise approaches, our framework improves biomarker reliability, particularly in realistic settings dominated by annotation ambiguity and limited domain shift.

Accurate brain parcellation in diffusion MRI (dMRI) space is essential for advanced neuroimaging analyses. However, most existing approaches rely on anatomical MRI for segmentation and inter-modality registration, a process that can introduce errors and limit the versatility of the technique. In this study, we present a novel deep learning-based framework for direct parcellation based on the Desikan-Killiany (DK) atlas using only diffusion MRI-derived data. Our method utilizes a hierarchical, two-stage segmentation network: the first stage performs coarse parcellation into broad brain regions, and the second stage refines the segmentation to delineate more detailed subregions within each coarse category. We conduct an extensive ablation study to evaluate various diffusion-derived parameter maps, identifying a top-performing combination of fractional anisotropy, trace, sphericity, and maximum eigenvalue that enhances parcellation accuracy compared with previously used parameter choices. When evaluated on the Human Connectome Project, our approach achieves higher Dice Similarity Coefficients compared to existing state-of-the-art methods. On the Consortium for Neuropsychiatric Phenomics dataset, where reliable voxel-wise DK reference labels in diffusion space are not available, our method demonstrates label-free evidence of robustness across different image resolutions and acquisition protocols by producing more homogeneous parcellations as measured by the relative standard deviation within regions. This work represents a step toward more practical dMRI-based brain parcellation by avoiding the need for anatomical MRI and subject-specific anatomical-to-diffusion registration at inference time. The implementation of our method is publicly available on https://github.com/xmindflow/DKParcellationdMRI.

Pseudo kinetics-driven federated diffusion hemodynamic framework for breast tumor segmentation in pre-contrast MRI

A selective memory network for medical image segmentation in 3D MRI

This study conducts quantitative and qualitative analyses to investigate the relationship between human-annotated and AI-derived uncertainty in cardiac MRI segmentation, aiming to enhance the reliability of AI-based cardiac MRI segmentation models and foster better human-AI collaboration. The CMRI dataset used in the experiments consists of 483 scans, each with two types of labels: an annotated segmentation mask and an uncertainty score per CMRI slice, both provided by clinicians. First, the AI-derived uncertainty estimated by the fuzzy-based algorithm is utilized to indicate the quality of segmentation. Multiple levels of uncertainties are derived from the method, including class-wise, slice-wise, subject-wise, etc. Subsequently, they are compared to the human-annotated uncertainty scores. Finally, qualitative analyses are conducted with clinicians to investigate all uncertainty measures potentially impacting real clinical applications. Experimental results show a strong inverse correlation between AI-derived uncertainty and Dice score, a standard metric for segmentation quality, indicating that lower uncertainty predicts higher segmentation quality. Additionally, it is found that human-annotated uncertainty coincides with AI-derived uncertainty for some anatomical structures (e.g., papillary muscle). However, high human-annotated uncertainty does not necessarily correlate with low AI segmentation quality, and there is no obvious association between human-annotated uncertainty and the size of the segmented structure. Concluded from qualitative analysis with clinicians, humans are better at utilizing prior knowledge (e.g. cardiac structural and contextual information) for uncertainty scoring, while the current AI method lacks this capability and is mainly data-driven for decision-making and uncertainty estimation. AI-derived uncertainty could be utilized as quality control for CMRI segmentation. Humans utilize structural and contextual information to formulate uncertainty, while AI models currently lack this capability.

Cardiac MRI (CMRI) is crucial for assessing cardiovascular structure and function, yet manual analysis is time-consuming and subjective. This paper introduces CMRI Insight, an open-source, Python-based GUI tool designed for automated cardiac segmentation and motion tracking like CardioTrackNet in CMRI. Leveraging deep learning models like CardSegNet and MECardNet, it enables accurate delineation of the left ventricle, right ventricle, and myocardium with mean Dice scores of 95.8% for the left ventricle, 94.2% for the right ventricle, and 95.0% for the myocardium. The tool offers 3D mesh reconstruction, strain analysis, bull’s-eye visualization, and support for custom models. CMRI Insight enhances efficiency and reproducibility in cardiac imaging research and clinical evaluation, providing a user-friendly platform for comprehensive CMRI analysis.

Deep learning-driven MRI segmentation of choroid plexus volume: a novel biomarker for cognitive impairment in type 2 diabetes mellitus.

The advent of deep learning has significantly advanced the state of the art in cardiac magnetic resonance (CMR) image segmentation. However, most models remain task-specific, which hinders the development and validation of approaches generalizable across different clinical centers, imaging protocols, or scanner vendors. Vision foundation models (VFMs), pre-trained on large-scale natural image datasets, offer powerful and transferable representations under the "pre-training and fine-tuning" paradigm. Nevertheless, adapting them to CMR segmentation faces two major challenges: (1) the substantial domain gap between natural and medical images limits transferability; and (2) extracting domain-agnostic features from diverse domain styles represents a key bottleneck for domain generalization with VFMs. To address these issues, we propose a scenario-activated fine-tuning initialization strategy that adapts VFMs to MRI characteristics and employs singular value decomposition to extract principal components for parameter-efficient tuning. This approach enables robust domain-generalized CMR segmentation. Additionally, we apply Haar wavelet transforms to disentangle style information from domain-invariant content. The former helps stabilize scene content, while the latter captures scene style and mitigates its impact on domain-generalized semantic segmentation. Experiments under various CMR domain generalization segmentation settings demonstrate the state-of-the-art performance of our FA-SedLoRA framework and its versatility across different VFMs.