cortical parcellation

Existing learning-based cortical surface parcellation methods lack in-depth analysis of performance improvement mechanisms—particularly their interplay with registration and atlas propagation. Method: We propose the first end-to-end joint cortical registration and parcellation framework, featuring deep coupling between the two tasks: a learnable atlas propagation module and a shallow fine-tuning subnetwork; a lightweight geometric-feature-driven architecture (using sulcal depth and curvature) that jointly optimizes diffeomorphic registration and label propagation. Results: On the Mindboggle dataset, our method achieves Dice scores exceeding 90%, significantly outperforming both conventional and state-of-the-art learning-based approaches. Ablation studies confirm registration quality as the key bottleneck limiting parcellation accuracy. Our framework enhances anatomical consistency and label fidelity of cortical parcellations while improving statistical power of brain atlases—thereby providing robust support for clinical applications such as neurosurgical planning.

This work addresses the challenges in neuroimaging research caused by data heterogeneity and incompatible tool interfaces, which often necessitate redundant implementations of routine operations. To overcome these limitations, the authors propose a unified, open-source Python framework that, for the first time, integrates volumetric, cortical surface, and streamline data within a single system. Through an object-oriented design, it provides consistent interfaces for loading, processing, and exporting data, while supporting BIDS compliance, FreeSurfer integration, diffusion MRI analysis, and GPU-accelerated visualization. The framework includes built-in color map and lookup table management, brain parcellation, multi-surface rendering, and tractography capabilities, enabling code-free workflow customization via JSON configuration. Developed for Python 3.9–3.12, it supports standard formats such as NIfTI, GIFTI, and TCK/TRK, substantially lowering technical barriers. The code is publicly available with comprehensive documentation and examples.

This study addresses key challenges in automated white matter tract segmentation—namely, high morphological similarity among streamlines, substantial inter-subject variability, and hemispheric symmetry-induced ambiguities. We propose the first end-to-end segmentation method based on a GPT architecture. Our approach introduces a novel tri-level representation: streamline-level (capturing fine-grained geometric structure), cluster-level (encoding tract-specific semantic context), and fusion-level (integrating multi-scale features). This hierarchical modeling preserves tract shape integrity while enhancing discriminative power. Evaluated on TractoInferno and 105HCP datasets, our method achieves state-of-the-art performance across Dice, Overlap, and Overreach metrics, with statistically significant improvements in cross-dataset generalization. The framework delivers a high-accuracy, robust, and broadly applicable tool for connectomic analysis, neurosurgical planning, and mechanistic studies of neurological disorders.

To address the efficiency–accuracy trade-off in tract segmentation from whole-brain diffusion MRI—particularly under large-scale data and clinical low-compute settings—this paper proposes the first lightweight, parallel streamline clustering framework leveraging global context modeling. Methodologically: (1) it introduces a flip-invariant streamline embedding with flip augmentation to explicitly encode streamline undirectedness; (2) it employs a Transformer-based architecture to enable global contextual interaction across streamline subsets, eliminating redundant local feature modeling; and (3) it incorporates a randomized sub-fiber graph partitioning scheme to enable highly efficient, GPU-free parallel inference on CPU. Experiments demonstrate that our method achieves over 100× speedup versus TractCloud, enabling real-time execution on standard clinical workstations without GPU acceleration, while matching or surpassing state-of-the-art accuracy on both healthy and pathological datasets.

This study addresses the limitations of existing brain segmentation and cortical parcellation tools when applied to ultra-high-field MRI (UHF-MRI) due to signal inhomogeneity and heterogeneity in contrast and resolution. To overcome these challenges, we propose a deep learning toolbox designed for multi-field-strength, multi-contrast, and multi-cohort scenarios, comprising two independently trained 3D U-Nets for 35-label whole-brain segmentation and 62-label Desikan–Killiany–Tourville (DKT) cortical parcellation, integrated with a volumetric measurement pipeline. Innovatively employing a domain randomization strategy trained on data from 238 subjects, our method achieves, for the first time, robust cortical parcellation on UHF-MRI. Experiments demonstrate superior performance over existing approaches across multiple heterogeneous datasets, with highly reliable parcellation and volumetric measurements that show strong agreement with standard protocols.

This study addresses the loss of patient-specific accuracy in template-space segmentation of small subcortical nuclei—such as the subthalamic nucleus (STN), red nucleus (RN), and substantia nigra (SN)—which is critical for Parkinson’s disease surgical planning. The authors propose a native-space segmentation pipeline based on UNet, leveraging multi-center 7T and 3T MRI data to systematically evaluate, for the first time, the advantages of native-space over template-based approaches. To mitigate domain shift between field strengths, they employ disentangled representation learning to generate synthetic 3T images. Performance is assessed using Dice coefficient and Hausdorff Distance at 95% (HD95). Results demonstrate that the native-space method significantly outperforms template registration on 7T data (STN Dice: 0.775 vs. 0.713) with superior boundary precision; however, direct transfer of the 7T model to 3T leads to performance degradation, and synthetic data yields only marginal improvement.

This study addresses the opacity of Alzheimer’s disease (AD) detection mechanisms caused by coupled brain segmentation and classification. We propose decoupling these stages and systematically benchmarking rapid deep learning approaches. Through a factorial design, we comprehensively evaluate various combinations of segmentation methods—including SynthSeg+, OpenMAP-T1, and zero/few-shot prompting with foundation models—alongside volumetry and classifiers to assess their impact on downstream tasks. Extensive validation on the OASIS-1 dataset reveals critical interaction effects among components, with all results quantified using BCa bootstrap 95% confidence intervals. This work establishes a rigorous methodological foundation and performance benchmark for AD detection, clarifying the specific contributions of individual pipeline stages to diagnostic accuracy.

研究通过比较多种损失函数方法，解决了纤维束分割在示踪组织学中的自动化问题，并提出新的空间诊断方法Excess32来评估分割质量。

This work addresses the challenge of simultaneously capturing high-resolution spatial details and efficiently modeling variable-length longitudinal brain scans for early Alzheimer’s disease (AD) prediction. The authors propose an anatomy-aware longitudinal Transformer framework that leverages an atlas-guided regional segmentation encoder to convert 3D MRI into anatomically semantic tokens, further integrating patient age information to enable linear-complexity modeling of arbitrary-length 4D structural MRI sequences. By circumventing the quadratic complexity bottleneck of conventional 4D Vision Transformers, the method achieves significant performance gains on ADNI, AIBL, and MIRIAD datasets, improving balanced accuracy by 7% for MCI-to-AD conversion prediction and by 5% for AD/CN classification. It also uncovers clinically consistent atrophy patterns and demonstrates generalizability to other neurodegenerative conditions such as frontotemporal dementia.

Manual annotation of fiber tracts in histological images of non-human primates is prohibitively costly, severely limiting the validation efficiency of diffusion MRI (dMRI) tractography. To address this challenge, this work proposes a novel approach that leverages ex vivo dMRI tractography results as a generative prior to synthesize 2D image patches with realistic foreground textures, which are then composited onto postmortem blockface photographs. Domain randomization is further incorporated to enhance data diversity. A 2D U-Net model trained on a combination of real and synthetic data achieves state-of-the-art segmentation performance using only one-third of the manually annotated data required by conventional methods. This hybrid training strategy significantly outperforms models trained solely on real data and demonstrates superior generalization across brain regions and varying fiber densities.