Medical image analysis最新文献

{"title":"M³Surv: Fusing Multi-slide and Multi-omics for Memory-augmented robust Survival prediction.","authors":"Mingcheng Qu, Guang Yang, Donglin Di, Yue Gao, Yang Song, Lei Fan","doi":"10.1016/j.media.2025.103846","DOIUrl":"https://doi.org/10.1016/j.media.2025.103846","url":null,"abstract":"<p><p>Multimodal survival prediction is crucial for personalized oncology. However, existing methods typically integrate only Formalin-Fixed Paraffin-Embedded (FFPE) slides with a single omics type, such as genomics, overlooking Fresh Frozen (FF) slides that better preserve molecular information, as well as richer multi-omics data like proteomics and transcriptomics. More critically, the complete absence of certain modalities due to clinical constraints (e.g., time or cost) severely limits the applicability of conventional fusion models that rely on inter-modality correlations. To address these gaps, we propose M³Surv, a framework designed to integrate multi-pathology slides (both FF and FFPE) with multi-omics profiles. For multi-slide fusion, we design a divide-and-conquer hypergraph learning approach to capture both intra-slide higher-order cellular structures and inter-slide relationships, yielding a unified pathology representation. To enrich the biological context, we integrate multi-omics data and employ interactive cross-attention to fuse the pathological and omics modalities. To tackle the missing modality, we introduce a prototype-based memory bank. During training, this memory bank learns and stores representative pathology-omics feature prototypes. At inference, even if a modality is entirely missing, the model can query the bank with available features and robustly impute information from the most similar prototype. Extensive experiments on five TCGA cancer datasets and an in-house dataset demonstrate that M³Surv outperforms state-of-the-art methods, achieving an average 2.2% improvement in C-Index. The framework also shows strong stability across various missing modality scenarios, highlighting its clinical potential in real-world, data-incomplete scenarios.</p>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 Pt B","pages":"103846"},"PeriodicalIF":11.8,"publicationDate":"2025-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145345758","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An efficient end-to-end computational framework for the generation of ECG calibrated volumetric models of human atrial electrophysiology","authors":"Elena Zappon ,&nbsp;Luca Azzolin ,&nbsp;Matthias A.F. Gsell ,&nbsp;Franz Thaler ,&nbsp;Anton J. Prassl ,&nbsp;Robert Arnold ,&nbsp;Karli Gillette ,&nbsp;Mohammadreza Kariman ,&nbsp;Martin Manninger ,&nbsp;Daniel Scherr ,&nbsp;Aurel Neic ,&nbsp;Martin Urschler ,&nbsp;Christoph M. Augustin ,&nbsp;Edward J. Vigmond ,&nbsp;Gernot Plank","doi":"10.1016/j.media.2025.103822","DOIUrl":"10.1016/j.media.2025.103822","url":null,"abstract":"<div><div>Computational models of atrial electrophysiology (EP) are increasingly utilized for applications such as the development of advanced mapping systems, personalized clinical therapy planning, and the generation of virtual cohorts and digital twins. These models have the potential to establish robust causal links between simulated <em>in silico</em> behaviors and observed human atrial EP, enabling safer, cost-effective, and comprehensive exploration of atrial dynamics. However, current state-of-the-art approaches lack the fidelity and scalability required for regulatory-grade applications, particularly in creating high-quality virtual cohorts or patient-specific digital twins. Challenges include anatomically accurate model generation, calibration to sparse and uncertain clinical data, and computational efficiency within a streamlined workflow.</div><div>This study addresses these limitations by introducing novel methodologies integrated into an automated end-to-end workflow for generating high-fidelity digital twin snapshots and virtual cohorts of atrial EP. These innovations include: (i) automated multi-scale generation of volumetric biatrial models with detailed anatomical structures and fiber architecture; (ii) a robust method for defining space-varying atrial parameter fields; (iii) a parametric approach for modeling inter-atrial conduction pathways; and (iv) an efficient forward EP model for high-fidelity electro-cardiogram (ECG) computation.</div><div>We evaluated this workflow on a cohort of 50 atrial fibrillation (AF) patients, producing high-quality meshes suitable for reaction-eikonal and reaction–diffusion models, demonstrating the ability to efficiently simulate atrial ECGs under parametrically controlled conditions, and, as a proof-of-concept, the feasibility of calibrating models to clinical P-wave in four patients. These advancements represent a critical step towards scalable, precise, and clinically applicable digital twin models and virtual cohorts, enabling enhanced patient-specific predictions and therapeutic planning.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103822"},"PeriodicalIF":11.8,"publicationDate":"2025-10-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145308404","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Rapid whole brain motion-robust mesoscale in-vivo MR imaging using multi-scale implicit neural representation","authors":"Jun Lyu ,&nbsp;Lipeng Ning ,&nbsp;William Consagra ,&nbsp;Qiang Liu ,&nbsp;Richard J. Rushmore ,&nbsp;Berkin Bilgic ,&nbsp;Yogesh Rathi","doi":"10.1016/j.media.2025.103830","DOIUrl":"10.1016/j.media.2025.103830","url":null,"abstract":"<div><div>High-resolution whole-brain in vivo MR imaging at mesoscale resolutions remains challenging due to long scan durations, motion artifacts, and limited signal-to-noise ratio (SNR). While acquiring multiple anisotropic scans from rotated slice orientations offers a practical compromise, reconstructing accurate isotropic volumes from such inputs remains non-trivial due to the lack of high-resolution ground truth and the presence of inter-scan motion. To address these challenges, we proposes <u>Ro</u>tating-<u>v</u>iew sup<u>er</u>-resolution (ROVER)-MRI, an unsupervised framework based on multi-scale implicit neural representations (INR), enabling accurate recovery of fine anatomical details from multi-view thick-slice acquisitions. ROVER-MRI employs coordinate-based neural networks to implicitly and continuously encode image structures at multiple spatial scales, simultaneously modeling anatomical continuity and correcting inter-view motion through an integrated registration mechanism. Validation on ex-vivo monkey brain data and multiple in-vivo human datasets demonstrates substantially improved reconstruction performance compared to bi-cubic interpolation and state-of-the-art regularized least-squares super-resolution reconstruction (LS-SRR) with 2-fold reduction in scan time. Notably, ROVER-MRI enables whole-brain in-vivo T2-weighted imaging at <span><math><mrow><mn>180</mn><mspace></mspace><mi>μ</mi><mi>m</mi></mrow></math></span> isotropic resolution in just 17 min on a 7T scanner, achieving a 22.4% reduction in relative error compared to LS-SRR. We also demonstrate improved SNR using ROVER-MRI compared to a time-matched 3D GRE acquisition. Quantitative results on several datasets demonstrate better sharpness of the reconstructed images with ROVER-MRI for different super-resolution factors (5 to 11). These findings highlight ROVER-MRI’s potential as a rapid, accurate, and motion-resilient mesoscale imaging solution, promising substantial advantages for neuroimaging studies.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103830"},"PeriodicalIF":11.8,"publicationDate":"2025-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145268284","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-contrast low-field MRI acceleration with k-space progressive learning and image-space hybrid attention fusion","authors":"Xiaohan Xing ,&nbsp;Qi Chen ,&nbsp;Lequan Yu ,&nbsp;Liang Qiu ,&nbsp;Lingting Zhu ,&nbsp;Lei Xing ,&nbsp;Lianli Liu","doi":"10.1016/j.media.2025.103833","DOIUrl":"10.1016/j.media.2025.103833","url":null,"abstract":"<div><div>Multi-contrast MRI provides complementary tissue information for diagnosis and treatment planning but is limited by the long acquisition time and system noise, which deteriorates at low field strength. To jointly accelerate and denoise multi-contrast MRI acquired at low field strength, we present a novel dual-domain framework designed to reconstruct high-quality multi-contrast MR images from k-space data corrupted by under-sampling and system noise. Our dual-domain framework first enhances k-space data quality through a k-space <em>Low-to-High Frequency Progressive</em> (LHFP) learning network, and then further refines the k-space outputs with an image-space <em>Hybrid Attention Fusion Network</em> (HAFNet). In k-space learning, the magnitude imbalance between the low- and high-frequency components may cause the network to be dominated by low-frequency components, leading to sub-optimal recovery of high-frequency components. To tackle this challenge, the two-stage LHFP learning network first recovers low-frequency components and then emphasizes high-frequency learning through patient-specific adaptive prediction of the low-high frequency boundary. In image domain learning, the challenge of efficiently capturing long-range dependencies across the multi-contrast images is resolved through <em>Hybrid Window-based Attention Fusion</em> (HWAF) modules, which integrate features by alternately computing self-attention within dense and dilated windows. Extensive experiments on the BraTs MRI and M4Raw low-field MRI datasets demonstrate the superiority of our method over state-of-the-art MRI reconstruction methods. Our source code will be made publicly available upon acceptance.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103833"},"PeriodicalIF":11.8,"publicationDate":"2025-10-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145268285","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Semi-supervised Fetal Brain Parcellation via Hierarchical Learning Framework","authors":"Shijie Huang ,&nbsp;Kai Zhang ,&nbsp;Fangmei Zhu ,&nbsp;Zhongxiang Ding ,&nbsp;Geng Chen ,&nbsp;Dinggang Shen","doi":"10.1016/j.media.2025.103835","DOIUrl":"10.1016/j.media.2025.103835","url":null,"abstract":"<div><div>Automatic parcellation of fetal brain regions using magnetic resonance (MR) images has become a valuable tool for studying prenatal brain growth and development. However, manual segmentation on large-scale fetal brain images is challenging, leading to limited annotated data availability. Although previous works have made progress, they are limited by not considering hierarchical nature and complementary information between brain regions. To overcome this limitation, we introduce a novel method to hierarchically segment the fetal brain into 87 distinct regions. The method employs a three-level coarse-to-fine network with the coarse level providing prior information to aid the fine level for fine segmentation. The first level predicts 8 brain regions, the second level refines the first-level 8 brain regions into 36 regions, and the final level refines further into 87 regions. This design hierarchically decomposes the fine difficult-to-achieve segmentation task into the coarse relatively-easy-to-achieve tasks by using guiding information from coarse level. Additionally, we introduce a data augmentation module to simulate variations in imaging conditions. To ensure robust segmentation performance under diverse imaging conditions, the network is trained in a semi-supervised manner using simulated data combined with a small set of labeled real data. In this way, we address the issue of limited high-quality labeled data, and enhance the model’s robustness to MR scanner variability. Extensive experiments on 558 neonatal subjects from the dHCP dataset and 176 fetal brain MR images demonstrate excellent segmentation performance of our method in terms of Dice score (91.42%), outperforming the second best nnUNet (88.77%).</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103835"},"PeriodicalIF":11.8,"publicationDate":"2025-10-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145280438","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A saliency detection-inspired method for optic disc and cup segmentation","authors":"Fan Guo ,&nbsp;Wentao Liu ,&nbsp;Jin Tang","doi":"10.1016/j.media.2025.103836","DOIUrl":"10.1016/j.media.2025.103836","url":null,"abstract":"<div><div>Glaucoma, as one of the leading causes of blindness worldwide, requires early diagnosis for effective patient treatment. Accurate segmentation of the optic cup and optic disc, along with the calculation of the cup-to-disc ratio (CDR), is central to glaucoma screening. However, traditional semantic segmentation methods face significant challenges in handling complex fundus images due to interference from background structures such as blood vessels. To address this, this paper proposes a saliency detection-inspired method for optic cup and disc segmentation, extending saliency detection to a three-class task (optic cup, optic disc, and background). The approach incorporates an Edge-guided Multi-scale Feature Extraction Module (EMFEM), a Global Context Information Enhancement Module (GCIEM), and a Self-Interaction Module (SIM) to integrate multi-level features and improve segmentation performance. Additionally, a ConvNeXtV2-based feature extraction network and improved loss functions—including Cross-Entropy Loss, Consistency-Enhanced Loss (CEL), and Edge-Gradient-Aware Tversky Loss (EAL)—are employed to optimize saliency focus and boundary refinement. Experimental results demonstrate that the proposed method outperforms mainstream segmentation algorithms on six public datasets. It achieves the highest Dice coefficients of 0.9073 for optic cup segmentation on the Drishti-GS dataset, and 0.9734 and 0.8965 for optic cup and disc segmentation on the Rim-One dataset, respectively. The method exhibits strong robustness and generalizability, offering a promising direction for glaucoma-assisted diagnosis and medical image segmentation.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103836"},"PeriodicalIF":11.8,"publicationDate":"2025-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145268288","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Fusion and pure feature extraction framework for intraoperative hyperspectral of thyroid lesion","authors":"Shuang Xue ,&nbsp;Zhoufeng Zhang ,&nbsp;Siyuan Li ,&nbsp;Jian Du ,&nbsp;Hang Zhao ,&nbsp;Meijie Qi ,&nbsp;Chenglong Tao","doi":"10.1016/j.media.2025.103832","DOIUrl":"10.1016/j.media.2025.103832","url":null,"abstract":"<div><div>Thyroid cancer has remained one of the most prevalent endocrine malignancies. In routine surgery, thyroid cancer analysis involves two time-consuming steps: intraoperative frozen section preparation and manual microscopic examination. Recently, info-rich hyperspectral intelligence analysis has been studied, reducing subjective bias but only optimizing the intraoperative second step and the model complexity, ignoring the independent features that possess substance fingerprints. To bridge the gaps, we developed a hyperspectral recognition algorithm called PS4EM-SN for intraoperatively ex-vivo macro thyroid lesion, which comprised a pure spectral with pure spatial(SPS) learning framework and a spatial–spectral fusion embed mechanism(SSEM) coupled with cascade attention. The cascade attention mechanism, integrating Squeeze-and-Excitation (SE) and Non-Local (NOL) blocks, enhanced robustness to the outliers of SSEM and improved generalization. The experimental results were satisfactory in differentiating non-malignant and malignant regions with 93.91% average accuracy. Given its hyperspectral multifaceted performance, our method promises a digital solution for intraoperative thyroid diagnosis.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103832"},"PeriodicalIF":11.8,"publicationDate":"2025-10-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145268287","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Anatomically and metabolically informed diffusion for unified denoising and segmentation in low-count PET imaging","authors":"Menghua Xia ,&nbsp;Kuan-Yin Ko ,&nbsp;Der-Shiun Wang ,&nbsp;Ming-Kai Chen ,&nbsp;Qiong Liu ,&nbsp;Huidong Xie ,&nbsp;Liang Guo ,&nbsp;Wei Ji ,&nbsp;Jinsong Ouyang ,&nbsp;Reimund Bayerlein ,&nbsp;Benjamin A. Spencer ,&nbsp;Quanzheng Li ,&nbsp;Ramsey D. Badawi ,&nbsp;Georges El Fakhri ,&nbsp;Chi Liu","doi":"10.1016/j.media.2025.103831","DOIUrl":"10.1016/j.media.2025.103831","url":null,"abstract":"<div><div>Positron emission tomography (PET) image denoising, along with lesion and organ segmentation, are critical steps in PET-aided diagnosis. However, existing methods typically treat these tasks independently, overlooking inherent synergies between them as correlated steps in the analysis pipeline. In this work, we present the anatomically and metabolically informed diffusion (AMDiff) model, a unified framework for denoising and lesion/organ segmentation in low-count PET imaging. By integrating multi-task functionality and exploiting the mutual benefits of these tasks, AMDiff enables direct quantification of clinical metrics, such as total lesion glycolysis (TLG), from low-count inputs. The AMDiff model incorporates a semantic-informed denoiser based on diffusion strategy and a denoising-informed segmenter utilizing nnMamba architecture. The segmenter constrains denoised outputs via a lesion-organ-specific regularizer, while the denoiser enhances the segmenter by providing enriched image information through a denoising revision module. These components are connected via a warming-up mechanism to optimize multi-task interactions. Experiments on multi-vendor, multi-center, and multi-noise-level datasets demonstrate the superior performance of AMDiff. For test cases below 20% of the clinical count levels from participating sites, AMDiff achieves TLG quantification biases of −21.60±47.26%, outperforming its ablated versions which yield biases of −30.83±59.11% (without the lesion-organ-specific regularizer) and −35.63±54.08% (without the denoising revision module). By leveraging its internal multi-task synergies, AMDiff surpasses standalone PET denoising and segmentation methods. Compared to the benchmark denoising diffusion model, AMDiff reduces the normalized root-mean-square error for lesion/liver by 22.92/17.27% on average. Compared to the benchmark nnMamba segmentation model, AMDiff improves lesion/liver Dice coefficients by 10.17/2.02% on average.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103831"},"PeriodicalIF":11.8,"publicationDate":"2025-10-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145268289","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Nested resolution mesh-graph CNN for automated extraction of liver surface anatomical landmarks","authors":"Xukun Zhang ,&nbsp;Jinghui Feng ,&nbsp;Peng Liu ,&nbsp;Minghao Han ,&nbsp;Yanlan Kang ,&nbsp;Jingyi Zhu ,&nbsp;Le Wang ,&nbsp;Xiaoying Wang ,&nbsp;Sharib Ali ,&nbsp;Lihua Zhang","doi":"10.1016/j.media.2025.103825","DOIUrl":"10.1016/j.media.2025.103825","url":null,"abstract":"<div><div>The anatomical landmarks on the liver (mesh) surface, including the falciform ligament and liver ridge, are composed of triangular meshes of varying shapes, sizes, and positions, making them highly complex. Extracting and segmenting these landmarks is critical for augmented reality-based intraoperative navigation and monitoring. The key to this task lies in comprehensively understanding the overall geometric shape and local topological information of the liver mesh. However, due to the liver’s variations in shape and appearance, coupled with limited data, deep learning methods often struggle with automatic liver landmark segmentation. To address this, we propose a two-stage automatic framework combining mesh-CNN and graph-CNN. In the first stage, dynamic graph convolution (DGCNN) is employed on low-resolution meshes to achieve rapid global understanding, generating initial landmark proposals at two levels, “dilation” and “erosion”, and mapping them onto the original high-resolution surface. Subsequently, a refinement network based on mesh convolution fuses these landmark proposals from edge features along the local topology of the high-resolution mesh surface, producing refined segmentation results. Additionally, we incorporate an anatomy-aware Dice loss to address resolution imbalance and better handle sparse anatomical regions. Extensive experiments on two liver datasets, both in-distribution and out-of-distribution, demonstrate that our method accurately processes liver meshes of different resolutions, outperforming state-of-the-art methods. The reconstructed liver mesh dataset and the source code are available at <span><span>https://github.com/xukun-zhang/MeshGraphCNN</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103825"},"PeriodicalIF":11.8,"publicationDate":"2025-10-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145268983","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Unsupervised multimodal surface registration with geometric deep learning","authors":"Mohamed A. Suliman ,&nbsp;Logan Z.J. Williams ,&nbsp;Abdulah Fawaz ,&nbsp;Emma C. Robinson","doi":"10.1016/j.media.2025.103821","DOIUrl":"10.1016/j.media.2025.103821","url":null,"abstract":"<div><div>This paper introduces GeoMorph, a novel geometric deep-learning framework designed for image registration of cortical surfaces. The registration process consists of two main steps. First, independent feature extraction is performed on each input surface using graph convolutions, generating low-dimensional feature representations that capture important cortical surface characteristics. Subsequently, features are registered in a deep-discrete manner to optimize the overlap of common structures across surfaces by learning displacements of a set of control points. To ensure smooth and biologically plausible deformations, we implement regularization through a deep conditional random field implemented with a recurrent neural network. Experimental results demonstrate that GeoMorph surpasses existing deep-learning methods by achieving improved alignment with smoother deformations. Furthermore, GeoMorph exhibits competitive performance compared to classical frameworks. Such versatility and robustness suggest strong potential for various neuroscience applications. Code is made available at <span><span>https://github.com/mohamedasuliman/GeoMorph</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103821"},"PeriodicalIF":11.8,"publicationDate":"2025-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145308399","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}