Computerized Medical Imaging and Graphics最新文献

{"title":"PCaSFUA-Net: Spatial frequency collaboration and uncertainty-aware fusion for multimodal prostate cancer segmentation","authors":"Chunyu Li,&nbsp;Mengxing Huang,&nbsp;Yuchun Li,&nbsp;Yue Li,&nbsp;Yu Zhang,&nbsp;Haixia Long,&nbsp;Tao Wang,&nbsp;Zhiming Bai","doi":"10.1016/j.compmedimag.2026.102806","DOIUrl":"10.1016/j.compmedimag.2026.102806","url":null,"abstract":"<div><div>Accurate prostate cancer segmentation is essential for disease progression assessment and prognostic evaluation. Multiparametric MRI (mpMRI) offers complementary structural and functional information, enabling more reliable lesion segmentation than single-modality MRI. However, existing methods still struggle to preserve high-frequency details and handle modality uncertainty, which limits multimodal fusion effectiveness and segmentation accuracy under ambiguous tumor boundaries, complex lesion morphology, and substantial inter-modality heterogeneity. To address these challenges, we propose PCaSFUA-Net, a multimodal prostate cancer segmentation network built upon a pretrained medical vision foundation model, enabling efficient task adaptation with minimal parameter updates. The proposed network integrates a unified multimodal enhancement and fusion (MEF) framework with a coarse-to-fine (C2F) strategy for spatial prompt refinement. Specifically, the MEF framework includes a spatial-frequency collaborative (SFC) module to enhance intra-modality boundary and texture representations and a modality uncertainty-aware fusion (MUAF) module to adaptively reweight multimodal features for robust cross-modal fusion. Extensive experiments on two public datasets and one private dataset show that PCaSFUA-Net consistently outperforms existing methods across multiple evaluation metrics.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"134 ","pages":"Article 102806"},"PeriodicalIF":5.5,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148738593","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Learning generalizable representations across Heterogeneous Acquisition Environments for Breast Ultrasound Diagnosis","authors":"Huanjun Wang,&nbsp;Shukang Zhang,&nbsp;Wentao Kong,&nbsp;Wei Shao,&nbsp;Daoqiang Zhang,&nbsp;Peng Wan","doi":"10.1016/j.compmedimag.2026.102799","DOIUrl":"10.1016/j.compmedimag.2026.102799","url":null,"abstract":"<div><div>Ultrasound image classification remains challenging in clinical practice, given substantial variations in image appearance across devices, acquisition protocols, and operator habits, often unrelated to underlying pathology and associated with inconsistent model performance and limited generalizability. To address this challenge, we propose <strong>HAE-BUS (Heterogeneous Acquisition Environments for Breast Ultrasound</strong>), a representation learning framework that aims to disentangle pathology-relevant features from acquisition-style bias and improve diagnostic consistency across heterogeneous acquisition conditions. Specifically, HAE-BUS uses Multimodal Large Language Models (MLLMs) as an offline semantic interface to characterize pathology-excluded acquisition cues, from which latent acquisition-style environments are inferred without relying on metadata. These inferred environments are then composed during training to suppress acquisition-style shortcuts while preserving pathology-relevant diagnostic features. Experiments conducted on two breast ultrasound benchmarks, including the public BUSBRA dataset and our private NDTH dataset, indicate that our approach obtains higher AUC and accuracy than the compared methods in the evaluated settings, along with more consistent performance across heterogeneous acquisition conditions, indicating improved robustness and suggesting potential for more reliable clinical translation.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"134 ","pages":"Article 102799"},"PeriodicalIF":5.5,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148551190","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Integrating physics guidance into deep learning for cine cardiovascular segmentation","authors":"My Anh Tran-Nguyen,&nbsp;Quoc Khanh Luong,&nbsp;Minh Bao Kha,&nbsp;Tuan Linh Dang","doi":"10.1016/j.compmedimag.2026.102798","DOIUrl":"10.1016/j.compmedimag.2026.102798","url":null,"abstract":"<div><div>Cine cardiovascular magnetic resonance (CMR) provides dynamic sequences for quantifying clinically important indices such as ventricular volumes and ejection fraction. A fast and accurate deep segmentation model is therefore valuable for timely analysis, reducing manual workload and improving reproducibility in routine workflows. However, most existing segmentation models remain largely data-driven and lack explicit physical constraints, which can reduce prediction plausibility under realistic acquisition variability. In this work, we present PGE-UNet, an efficient segmentation framework for cine CMR that incorporates radiofrequency-related information to promote physically plausible outputs. Specifically, it employs the proposed Noise-Aware Encoder with self-supervised noise estimation and smoothing mechanism for robust feature encoding, and the Physics-Regularized Decoder that enforces Maxwell–Helmholtz consistency under a pseudo transmit-field prior. Since transmit-field measurements are typically unavailable in standard protocols, we further introduce a simulator-based procedure to synthesize the pseudo transmit-field prior, enabling physics-regularized training without additional mapping sequences. Experiments on three public cine CMR benchmarks demonstrated strong accuracy in both DICE and HD95, achieving foreground average DICE of 0.9152 on ACDC, 0.8605 on M&amp;Ms, and <span><math><mrow><mn>0.8548</mn><mo>±</mo><mn>0.0762</mn></mrow></math></span> on SCD, while remaining lightweight with only 1.60 million parameters in the default setting. Under CPU-only inference, the default configuration recorded an average latency of 72.38 ms per case, making it suitable for deployment in time-sensitive and resource-constrained environments. Our source code is publicly available at <span><span>https://github.com/QuocKhanhLuong/PGE-UNet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"134 ","pages":"Article 102798"},"PeriodicalIF":5.5,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148551279","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Cardiac motion tracking with bidirectional latent neural ODEs from cine cardiac MRI","authors":"Dongsheng Ruan,&nbsp;Ke Zhou,&nbsp;Chenyi Zhu,&nbsp;Chengjin Yu,&nbsp;Zhifan Gao,&nbsp;Mingfeng Jiang,&nbsp;Luyan Zheng,&nbsp;Huafeng Liu","doi":"10.1016/j.compmedimag.2026.102797","DOIUrl":"10.1016/j.compmedimag.2026.102797","url":null,"abstract":"<div><div>Accurate and temporally consistent cardiac motion tracking from cine MRI is essential for functional assessment and disease analysis. However, existing learning-based methods are typically formulated in discrete time and struggle to capture long-range temporal dependencies, often leading to accumulated errors and physically inconsistent motion. In this paper, we propose a unified unsupervised framework for cardiac motion tracking based on latent neural ordinary differential equations (ODEs). A frame-aware encoder extracts motion-sensitive features with temporal embeddings, which are evolved in a compact latent space via a neural ODE to model continuous-time deformation dynamics. To capture complex temporal variations in an efficient manner, we incorporate a conditional MLP-based dynamics module and further perform bidirectional forward–backward evolution within a unified framework. The evolved latent representations are decoded into inter-frame deformation fields, and a bidirectional Lagrangian regularization is introduced to enforce long-term temporal consistency and motion reversibility across the cardiac cycle. Extensive experiments on the ACDC and M&amp;Ms datasets demonstrate that the proposed method achieves state-of-the-art performance, producing temporally consistent and physiologically plausible motion fields with a lightweight and efficient architecture.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"134 ","pages":"Article 102797"},"PeriodicalIF":5.5,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148537490","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"WDBDM: Wavelet-based dual-branch diffusion model for low-dose CT and PET denoising","authors":"Qi Sun,&nbsp;Tongtong Li,&nbsp;Guowei Wang,&nbsp;Yanyan Huang,&nbsp;Shunjie Dong,&nbsp;Lequan Yu,&nbsp;Kuangyu Shi,&nbsp;Zhijun Yao,&nbsp;Yu Fu,&nbsp;Bin Hu","doi":"10.1016/j.compmedimag.2026.102785","DOIUrl":"10.1016/j.compmedimag.2026.102785","url":null,"abstract":"<div><div>X-ray computed tomography (CT) and positron emission tomography (PET) are imaging modalities for assessing various diseases. While normal-dose CT (NDCT) and normal-dose PET (NDPET) imaging ensure high image quality, they often raise concerns regarding potential health risks from radiation exposure. This contradiction between reducing radiation dose and preserving diagnostic performance can be effectively addressed by reconstructing low-dose CT and low-dose PET images into high-quality images comparable to their normal-dose counterparts. In this study, we present WDBDM, a wavelet-based dual-branch diffusion framework for denoising low-dose data to generate normal-dose quality images. WDBDM consists of four main parts: Discrete Wavelet Transform (DWT), Low-Frequency Diffusion Branch (LFDB), High-Frequency Diffusion Branch (HFDB) and Fusion Module. Furthermore, we investigate the effectiveness of the Fusion Spatial–Frequency Convolution Module (FSFCM) in the diffusion branch, which can jointly extract spatial and frequency domain information, thereby significantly enhancing the feature representation capability of the model. Moreover, to prevent error propagation from imperfect recovery operators and to enable bidirectional guidance across high- and low-frequency components during sampling, we integrate HLF-MEMNet, a novel recovery network, into the WDBDM framework. This network leverages contextual information and high-low frequency mutual guidance constraints during the sampling process, preventing structural distortion and ensuring better alignment with the input at the next time step. Experiments on four public datasets and two imaging modalities demonstrate that WDBDM outperforms existing methods in denoising performance and generalization ability.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"133 ","pages":"Article 102785"},"PeriodicalIF":5.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148320393","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Segment Anything Model for medical image segmentation: A review","authors":"Hanguang Xiao,&nbsp;Shuai Liu,&nbsp;Xingyue Liu,&nbsp;Lingling Qian,&nbsp;Jiaxin Jiang","doi":"10.1016/j.compmedimag.2026.102789","DOIUrl":"10.1016/j.compmedimag.2026.102789","url":null,"abstract":"<div><div>The Segment Anything Model (SAM) is the first foundational model in general image segmentation, exhibiting outstanding generalization capability across a wide range of natural image tasks. However, medical images differ fundamentally from natural images in modality characteristics, semantic structure, and clinical objectives, resulting in limited systematic research on SAM adaptation to medical imaging. To bridge this gap and clarify the key challenges and developmental pathways for transferring general segmentation models to the medical domain, this review provides a comprehensive overview of recent advances and optimization strategies of SAM and its variants in medical image segmentation. Specifically, we first introduce the core principles of SAM and its research background in medical imaging, outlining the major technical challenges encountered in clinical contexts. Through an in-depth analysis and comparison of existing studies, we identify five key challenges, including the data gap, dimensionality barrier, precision bottleneck, semantic disconnect, and deployment hurdles. We then construct an integrated analytical framework to evaluate current adaptation mechanisms and optimization strategies. Furthermore, we discuss the potential and limitations of representative methods in clinical diagnosis, lesion detection, and surgical planning, summarizing the primary directions for improving model trustworthiness, interpretability, and privacy compliance. Finally, we highlight current research limitations and outline promising future directions, including the bidirectional integration of large language models, causal representation learning, and prompt optimization. We hope this work will serve as a valuable reference for future SAM-based medical image segmentation research and promote progress toward cross-modal integration, trustworthy artificial intelligence, and clinically deployable intelligent segmentation systems.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"133 ","pages":"Article 102789"},"PeriodicalIF":5.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148266871","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"RGCNN-nnUNet: Recurrent group equivariant nnU-Net for robust brain tissue segmentation on stroke NCCT","authors":"Xiang Li,&nbsp;Fengyuan Wang,&nbsp;Saurabh Bagchi,&nbsp;John Julius Volpi,&nbsp;Timea Hodics,&nbsp;Stephen T.C. Wong,&nbsp;Kelvin Wong","doi":"10.1016/j.compmedimag.2026.102792","DOIUrl":"10.1016/j.compmedimag.2026.102792","url":null,"abstract":"<div><div>Computed tomography (CT), particularly non-contrast CT (NCCT), is crucial in emergency settings for rapid stroke diagnosis due to its affordability and accessibility, yet subtle density variations in early ischemia are often obscured by high noise and low soft-tissue contrast in NCCT. To address this, we present RGCNN-nnUNet, a novel architecture designed for high-fidelity brain tissue segmentation. We introduce a Recurrent Group Convolutional Cell that leverages group equivariance to extract consistent anatomical features and stabilize iterative refinement. Evaluations on multi-center datasets demonstrate that RGCNN-nnUNet achieves state-of-the-art performance, reaching Dice coefficients of 0.885 for white matter and 0.86 for deep gray matter, while delivering a 14.3% and 10.6% reduction in Hausdorff Distance (HD95). To validate clinical impact, we conducted a blinded, independent reader study with a neuroimager. Results show that integrating RGCNN-nnUNet’s segmented tissue-specific Z-score maps with AI-denoised NCCT significantly improves diagnostic sensitivity compared to standard-of-care radiology reports, with an increased sensitivity by 29%. By providing precise anatomical boundaries and stable tissue-weighting, our framework enables the detection of subtle infarct signals.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"133 ","pages":"Article 102792"},"PeriodicalIF":5.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148383118","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"ScribSAM: A robust scribble-supervised framework for spatiotemporal segmentation of breast lesions in ultrasound videos","authors":"Long Chen,&nbsp;Qingqing Zheng,&nbsp;Yulong Guo,&nbsp;Qiong Wang","doi":"10.1016/j.compmedimag.2026.102790","DOIUrl":"10.1016/j.compmedimag.2026.102790","url":null,"abstract":"<div><div>Automated segmentation of breast lesions in ultrasound videos is critical for clinical applications but remains hindered by the reliance on expensive pixel-wise annotations. While scribble supervision offers a user-friendly alternative, its potential for ultrasound video segmentation remains underexplored. In this regard, we propose ScribSAM, a novel scribble-supervised framework built on the Medical Segment Anything Model (MedSAM) for robust ultrasound video segmentation. ScribSAM integrates two key innovations: a <strong>flow-guided scribble propagation module</strong> that leverages optical flow to efficiently propagate sparse scribble annotations across frames while preserving temporal consistency, and a <strong>bidirectional cross-attention module</strong> that fuses MedSAM’s global ViT embeddings with 3D CNN local-temporal embeddings for enhanced spatiotemporal feature learning. Extensive experiments on the scribble-annotated variants of two ultrasound video datasets, BUV2022 and US-VOS, demonstrate ScribSAM’s superiority. It surpasses state-of-the-art scribble-supervised methods by 5.53% (BUV2022) and 8.31% (US-VOS) in Dice score, outperforming some fully supervised methods and substantially narrowing the gap with the best fully supervised methods, while using only 4% of the annotated pixels. Code and dataset will be released at <span><span>https://github.com/003-GH/ScribSAM</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"133 ","pages":"Article 102790"},"PeriodicalIF":5.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148303830","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multimodal clinical data integration for prognosis of pulmonary embolism: A comparative study","authors":"Domenico Paolo,&nbsp;Paolo Soda,&nbsp;Matteo Tortora,&nbsp;Alessandro Bria,&nbsp;Rosa Sicilia","doi":"10.1016/j.compmedimag.2026.102794","DOIUrl":"10.1016/j.compmedimag.2026.102794","url":null,"abstract":"<div><div>Pulmonary embolism (PE) is a life-threatening condition for which accurate risk stratification is essential for clinical decision-making. Most existing prognostic models rely on a single data modality, typically structured electronic health records (EHR), limiting their ability to capture the heterogeneous nature of the disease. Multimodal approaches combining EHR, computed tomography (CT) imaging, and free-text clinical reports provide complementary information, but the optimal fusion strategy for prognosis prediction remains unclear. In this study, we present a comprehensive comparative study of multimodal fusion strategies for PE prognosis using the INSPECT dataset, which includes structured EHR, CT images, and clinical text with mortality labels. We develop unimodal baselines and multiple bimodal and trimodal models using early, intermediate, late, and cross-fusion strategies within a unified framework. Models are evaluated for 1-, 6-, and 12-month mortality prediction. Multimodal models consistently outperform unimodal baselines across all prediction horizons, with MCC improvements up to +0.13 for 1-month mortality (0.399 vs 0.269) and +0.11 for 6-month mortality (0.479 vs 0.367). The best overall performance is achieved by late fusion models, reaching an MCC of 0.497 for 12-month prediction, compared to 0.454 for the best unimodal model. Interestingly, bimodal combinations often outperform trimodal fusion in short-term prediction, suggesting that improper integration of multiple modalities may introduce noise and reduce effectiveness. These results highlight the importance of fusion design in multimodal clinical AI and provide practical guidance for developing robust prognosis models for pulmonary embolism. Code is publicly available at <span><span>https://github.com/arco-group/INSPECT-CS</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"133 ","pages":"Article 102794"},"PeriodicalIF":5.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148406699","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Self-supervised isotropic reconstruction for abnormality detection in anisotropic MRI","authors":"Jui-Yo Hsu,&nbsp;Pin-Hsun Lian,&nbsp;Tzu-Yi Chuang,&nbsp;Yi-Hwa Chen,&nbsp;Gary Han Chang","doi":"10.1016/j.compmedimag.2026.102786","DOIUrl":"10.1016/j.compmedimag.2026.102786","url":null,"abstract":"<div><div>Accelerating musculoskeletal magnetic resonance imaging (MRI) while preserving diagnostic detail remains challenging because acquiring fully‑isotropic ground‑truth volumes is clinically costly. In routine practice, anisotropic scans with reduced through-plane resolution degrade multiplanar visualization and slice-by-slice review in reformatted planes, obscure subtle abnormalities spanning only a few slices, and limit automated three-dimensional (3D) analyses that assume comparable spatial resolution across axes. We present a two‑stage, fully self‑supervised pipeline that learns directly from anisotropic scans—obviating any paired high‑resolution data—and converts highly anisotropic (8:1) turbo‑spin‑echo volumes into isotropic images and 3D abnormality maps. Unlike prior self-supervised super-resolution methods, Stage 1 uses a single forward multi-view generative adversarial network (GAN) with patch-based contrastive and adversarial objectives rather than a backward/cycle-consistency approach. Stage 2 leverages an anatomy-conditioned denoising-diffusion model for healthy counterfactual generation, yielding voxel-wise lesion maps without external annotations.</div><div>On 2225 Osteoarthritis Initiative knee scans from five different imaging centres, the framework reduced Fréchet inception distance from 407.4 → 254.4 (coronal) and 429.9 → 266.9 (axial), achieved the best Kernel Inception Distance (KID) / Learned Perceptual Image Patch Similarity (LPIPS) scores among competing unsupervised methods, and was preferred in 65–67% of blinded orthopedist comparisons. Crucially, isotropic enhancement propagated to downstream tasks: femur–tibia segmentation F1 scores increased and previously confluent bone‑marrow lesions were separated into discrete entities, enabling precise volumetric quantification. Robustness experiments demonstrated consistent gains across five imaging centers, synthetic noise/contrast perturbations, and transfer of the resolution-enhancement module to two additional MRI protocols, supporting robustness across sites and acquisition protocols.</div><div>By eliminating the need for ground‑truth isotropic images while surpassing state‑of‑the‑art unsupervised super‑resolution in both perceptual quality and clinical utility, our method may facilitate retrospective cohort studies and prospective scan-time reduction in heterogeneous knee MRI settings, with preliminary transferability to additional protocols.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"133 ","pages":"Article 102786"},"PeriodicalIF":5.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148363894","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}