Medical image analysis最新文献

{"title":"QCResUNet: Joint subject-level and voxel-level segmentation quality prediction","authors":"Peijie Qiu ,&nbsp;Satrajit Chakrabarty ,&nbsp;Phuc Nguyen ,&nbsp;Soumyendu Sekhar Ghosh ,&nbsp;Aristeidis Sotiras","doi":"10.1016/j.media.2025.103718","DOIUrl":"10.1016/j.media.2025.103718","url":null,"abstract":"<div><div>Deep learning has made significant strides in automated brain tumor segmentation from magnetic resonance imaging (MRI) scans in recent years. However, the reliability of these tools is hampered by the presence of poor-quality segmentation outliers, particularly in out-of-distribution samples, making their implementation in clinical practice difficult. Therefore, there is a need for quality control (QC) to screen the quality of the segmentation results. Although numerous automatic QC methods have been developed for segmentation quality screening, most were designed for cardiac MRI segmentation, which involves a single modality and a single tissue type. Furthermore, most prior works only provided subject-level predictions of segmentation quality and did not identify erroneous parts segmentation that may require refinement. To address these limitations, we proposed a novel multi-task deep learning architecture, termed QCResUNet, which produces subject-level segmentation-quality measures as well as voxel-level segmentation error maps for each available tissue class. To validate the effectiveness of the proposed method, we conducted experiments on assessing its performance on evaluating the quality of two distinct segmentation tasks. First, we aimed to assess the quality of brain tumor segmentation results. For this task, we performed experiments on one internal (Brain Tumor Segmentation (BraTS) Challenge 2021, <span><math><mrow><mi>n</mi><mo>=</mo><mn>1</mn><mo>,</mo><mn>251</mn></mrow></math></span>) and two external datasets (BraTS Challenge 2023 in Sub-Saharan Africa Patient Population (BraTS-SSA), <span><math><mrow><mi>n</mi><mo>=</mo><mn>40</mn></mrow></math></span>; Washington University School of Medicine (WUSM), <span><math><mrow><mi>n</mi><mo>=</mo><mn>175</mn></mrow></math></span>). Specifically, we first performed a three-fold cross-validation on the internal dataset using segmentations generated by different methods at various quality levels, followed by an evaluation on the external datasets. Second, we aimed to evaluate the segmentation quality of cardiac Magnetic Resonance Imaging (MRI) data from the Automated Cardiac Diagnosis Challenge (ACDC, <span><math><mrow><mi>n</mi><mo>=</mo><mn>100</mn></mrow></math></span>). The proposed method achieved high performance in predicting subject-level segmentation-quality metrics and accurately identifying segmentation errors on a voxel basis. This has the potential to be used to guide human-in-the-loop feedback to improve segmentations in clinical settings.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103718"},"PeriodicalIF":11.8,"publicationDate":"2025-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145047579","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":"HarmonicEchoNet: Leveraging harmonic convolutions for automated standard plane detection in fetal heart ultrasound videos","authors":"Md Mostafa Kamal Sarker ,&nbsp;Divyanshu Mishra ,&nbsp;Mohammad Alsharid ,&nbsp;Netzahualcoyotl Hernandez-Cruz ,&nbsp;Rahul Ahuja ,&nbsp;Olga Patey ,&nbsp;Aris T. Papageorghiou ,&nbsp;J. Alison Noble","doi":"10.1016/j.media.2025.103758","DOIUrl":"10.1016/j.media.2025.103758","url":null,"abstract":"<div><div>Fetal echocardiography offers non-invasive and real-time imaging acquisition of fetal heart images to identify congenital heart conditions. Manual acquisition of standard heart views is time-consuming, whereas automated detection remains challenging due to high spatial similarity across anatomical views with subtle local image appearance variations. To address these challenges, we introduce a very lightweight frequency-guided deep learning-based model named HarmonicEchoNet that can automatically detect heart standard views in a transverse sweep or freehand ultrasound scan of the fetal heart.</div><div>HarmonicEchoNet uses harmonic convolution blocks (HCBs) and a harmonic spatial and channel squeeze-and-excitation (hscSE) module. The HCBs apply a Discrete Cosine Transform (DCT)-based harmonic decomposition to input features, which are then combined using learned weights. The hscSE module identifies significant regions in the spatial domain to improve feature extraction of the fetal heart anatomical structures, capturing both spatial and channel-wise dependencies in an ultrasound image. The combination of these modules improves model performance relative to recent CNN-based, transformer-based, and CNN+transformer-based image classification models.</div><div>We use four datasets from two private studies, PULSE (Perception Ultrasound by Learning Sonographic Experience) and CAIFE (Clinical Artificial Intelligence in Fetal Echocardiography), to develop and evaluate HarmonicEchoNet models. Experimental results show that HarmonicEchoNet is 10–15 times faster than ConvNeXt, DeiT, and VOLO, with an inference time of just 3.9 ms. It also achieves 2%–7% accuracy improvement in classifying fetal heart standard planes compared to these baselines. Furthermore, with just 19.9 million parameters compared to ConvNeXt’s 196.24 million, HarmonicEchoNet is nearly ten times more parameter-efficient.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103758"},"PeriodicalIF":11.8,"publicationDate":"2025-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144908924","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":"Category-specific unlabeled data risk minimization for ultrasound semi-supervised segmentation","authors":"Lu Xu, Mingyuan Liu, Boxuan Wei, Yihua He, Zhifan Gao, Hongbin Han, Jicong Zhang","doi":"10.1016/j.media.2025.103773","DOIUrl":"https://doi.org/10.1016/j.media.2025.103773","url":null,"abstract":"Achieving accurate computer-aided analysis of ultrasound images is challenging, since not only its image artifacts but also the difficulties in collecting large-scale pixel-wise annotations from experts for training. Semi-supervised segmentation is a solution for learning from labeled and unlabeled data, which mainly focuses on generating pseudo annotations for unlabeled data or learning consistent features in enhanced views of images to enhance model generalization. However, anatomically, diverse learning difficulties across tissues are overlooked, and, technically, the estimation and minimization of empirical risk for unlabeled training data are largely ignored. Motivated by them, this work proposes a semi-supervised segmentation model, named CSUDRM, with two modules. The former is called category-specific distribution alignment (CSDA), which learns more consistent feature representations of the same class across labeled and unlabeled data. Moreover, it enhances feature space intra-class compactness and inter-class discrepancy and provides category-specific penalties for more robust learning. The latter one is Unlabeled Data Risk Minimization (UDRM). It minimizes the risk on the entire training data, which distinguishes it from most existing works that merely optimize labels. The risk of unlabeled data is estimated by a novel learnable class prior estimator, with the help of distributional hints from CSDA. This design could reinforce the robustness of the model and achieve stable segmentation. CSUDRM achieves state-of-the-art performances on four ultrasound datasets. Extensive ablation studies, including quantitative comparisons, feature space visualization, and robustness analysis, demonstrate the superiority of our designs.","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"13 1","pages":""},"PeriodicalIF":10.9,"publicationDate":"2025-08-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144924344","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":"Supervised white matter bundle segmentation in glioma patients with transfer learning","authors":"Chiara Riccardi ,&nbsp;Ludovico Coletta ,&nbsp;Sofia Ghezzi ,&nbsp;Gabriele Amorosino ,&nbsp;Luca Zigiotto ,&nbsp;Jorge Jovicich ,&nbsp;Silvio Sarubbo ,&nbsp;Paolo Avesani","doi":"10.1016/j.media.2025.103766","DOIUrl":"10.1016/j.media.2025.103766","url":null,"abstract":"<div><div>In clinical settings, the virtual dissection of white matter tracts represents an informative source of information for monitoring neurological conditions or to support the planning of a treatment. The implementation of this task through data-driven methodologies and, in particular, deep learning models demonstrates promising evidence of achieving high accuracy when applied to healthy individuals. However, the lack of large clinical datasets and the profound differences between healthy and clinical populations hinder the translation of these results to patients. Here, we investigated for the first time the effectiveness of transfer learning in adapting a deep learning architecture trained on a healthy population to glioma patients. Importantly, we provided the first thorough characterization of domain shift and its complexity, distinguishing systematic (i.e. measurement and pre-processing related) from tumor-specific components. Our results suggest that (i) models trained on a large normative healthy population have a significant performance drop when the inference is carried out on patients; (ii) transfer learning can be an effective strategy to overcome the shortage of clinical data and to manage the systematic shift; (iii) fine-tuning of the learning model cannot accommodate large white matter deformations induced by the tumor. The results were coherent across the five white matter bundles and the three input modalities tested, highlighting their robustness and generalizability. Our work provides valuable insights for advancing automated white matter segmentation in clinical populations and enhancing clinical transfer learning applications.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103766"},"PeriodicalIF":11.8,"publicationDate":"2025-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144899164","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 4D-flow MRI reconstruction based on partially-independent generative modeling and complex-difference sparsity constraint","authors":"Zhongsen Li ,&nbsp;Aiqi Sun ,&nbsp;Haining Wei ,&nbsp;Wenxuan Chen ,&nbsp;Chuyu Liu ,&nbsp;Haozhong Sun ,&nbsp;Chenlin Du ,&nbsp;Rui Li","doi":"10.1016/j.media.2025.103769","DOIUrl":"10.1016/j.media.2025.103769","url":null,"abstract":"<div><div>4D-flow MRI can provide spatiotemporal quantification of in-vivo blood flow velocity, which holds significant diagnostic value for various vascular diseases. Due to the large data size, 4D-flow MRI typically requires undersampling to shorten the scan time and employs reconstruction algorithms to recover images. Recently, deep learning methods have emerged for 4D-flow MRI reconstruction, but most of them are supervised algorithms, which have two major problems. First, supervised methods require high-quality fully sampled data for network training, which is usually very limited for 4D-flow MRI. Second, concerns are raised about the algorithm’s generalization ability since the morphology and velocity distribution vary in different vascular beds. In this work, we propose an unsupervised method for 4D-flow MRI reconstruction based on the deep image prior framework, which exploits the structural prior of convolutional neural networks for generative image recovery. Our method has three central components. First, we design a partially-independent network to improve the parameter efficiency and reduce the model size for 4D-flow MRI generation. Second, we incorporate the complex difference sparsity constraint to improve the accuracy of image phase recovery. Third, we introduce a joint generative and sparse optimization goal, and propose a “pretraining + ADMM finetuning” optimization algorithm for solution. Comprehensive experiments were conducted on two in-house acquired 4D-flow MRI datasets: an aorta dataset and a brain vessel dataset, compared with compressed-sensing algorithms and supervised deep-learning methods. The results demonstrate the superior reconstruction performance and generalization capability of the proposed method.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103769"},"PeriodicalIF":11.8,"publicationDate":"2025-08-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144895765","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 doppler-exclusive computational diagnostic framework to enhance conventional 2-D clinical ultrasound with 3-D mitral valve dynamics and cardiac hemodynamics","authors":"Nikrouz Bahadormanesh ,&nbsp;Mohamed Abdelkhalek ,&nbsp;Zahra Keshavarz-Motamed","doi":"10.1016/j.media.2025.103772","DOIUrl":"10.1016/j.media.2025.103772","url":null,"abstract":"<div><div>Mitral valve diseases are prevalent cardiac conditions especially by aging. With their high prevalence, the accessibility, accuracy, and reliability of the diagnostic methods are crucial. Mitral valve dynamics assessment could offer crucial insights into the progression of cardiac deterioration and recovery, significantly influencing patient care, intervention planning, and critical clinical decision-making in scenarios with potentially life-threatening risks. In this study, we developed a Doppler-exclusive computational diagnostic framework to assess mitral valve motion and dynamics as well as cardiac hemodynamics in patients non-invasively and at no risk to the patients. The framework was developed based on transthoracic echocardiogram (TTE) data (N=20), validated against transesophageal echocardiography (TEE) data (N=12) as well as CT data (N=4). In addition, we demonstrated the framework’s diagnostic abilities by providing novel and clinically-relevant analyses and interpretations of clinical data. Based on our findings, patient-specific left ventricular pressure was a strong predictor of stress levels in our cohort of 20 patients, despite being neglected by previous studies. There was a very strong negative correlation between the 3-D finite element-based coaptation area and vena Contracta width (R = -0.8; <em>p</em> &lt; 0.001). Furthermore, the LV conicity index, as the geometrical parameter showing left ventricle dilatation, had a strong positive correlation with end diastolic von Mises stress, used for quantification of leaflet tethering (R = 0.78; <em>p</em> &lt; 0.001). Finally, the patient-specific left ventricular pressure, and the rest length of the chords played a primary role in the biomechanical behavior of the mitral leaflets. The developed framework, while aligned with the current clinical metrics, could provide a strong add-on to the established clinical practice for the diagnosis of mitral valve diseases. Notably, this framework is novel in that it relies solely on standard Doppler ultrasound inputs, requiring no additional imaging or invasive measurements to achieve 3-D assessment. Clinically, the DE-MV-Dyn can be seamlessly applied in routine echocardiography exams to provide clinicians with new patient-specific metrics (e.g., leaflet stress, strain, and dynamic coaptation measures) for improved diagnosis and personalized mitral valve therapy planning.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103772"},"PeriodicalIF":11.8,"publicationDate":"2025-08-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144899167","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":"Paired phase and magnitude reconstruction neural network for multi-shot diffusion magnetic resonance imaging","authors":"Qiaoling Lin ,&nbsp;Xuanchu Chen ,&nbsp;Boxuan Shi ,&nbsp;Chen Qian ,&nbsp;Mingyang Han ,&nbsp;Liuhong Zhu ,&nbsp;Dafa Shi ,&nbsp;Xiaoyong Shen ,&nbsp;Wanjun Hu ,&nbsp;Dan Ruan ,&nbsp;Yi Guo ,&nbsp;Jianjun Zhou ,&nbsp;Xiaobo Qu","doi":"10.1016/j.media.2025.103771","DOIUrl":"10.1016/j.media.2025.103771","url":null,"abstract":"<div><div>Diffusion weighted imaging (DWI) is an important magnetic resonance imaging modality that reflects the diffusion of water molecules and has been widely used in tumor diagnosis. Higher image resolution is possible through multi-shot sampling but raises the challenge of suppressing image artifacts and noise when combining multi-shot data. Conventional methods introduce the magnitude and/or phase priors and regularize the reconstructed image in an iterative computing process, which suffers from slow computational speed. Deep learning offers a valuable solution to this challenge. In this work, traditional methods are adopted to generate the training labels offline. Then, a neural network is designed for paired phase and magnitude reconstruction. Last, the network is further improved by incorporating a high signal-to-noise ratio (SNR) b0 image with small geometric distortions. Compared with the state-of-the-art deep learning approach, results on simulated and in vivo data demonstrate that the proposed method enables sub-second fast reconstruction and achieves better objective evaluation criteria. Besides, a study by six radiologists on image quality confirms that the proposed method is within the excellent range and provides higher scores of image artifact suppression and more stable overall quality as well as SNR. This work provides a solution for fast and promising image reconstruction for multi-shot DWI.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103771"},"PeriodicalIF":11.8,"publicationDate":"2025-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144893533","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":"Leveraging sparse annotations for leukemia diagnosis on the large leukemia dataset","authors":"Abdul Rehman ,&nbsp;Talha Meraj ,&nbsp;Aiman Mahmood Minhas ,&nbsp;Ayisha Imran ,&nbsp;Mohsen Ali ,&nbsp;Waqas Sultani ,&nbsp;Mubarak Shah","doi":"10.1016/j.media.2025.103760","DOIUrl":"10.1016/j.media.2025.103760","url":null,"abstract":"<div><div>Leukemia is the 10th most frequently diagnosed cancer and one of the leading causes of cancer-related deaths worldwide. Realistic analysis of leukemia requires white blood cell (WBC) localization, classification, and morphological assessment. Despite deep learning advances in medical imaging, leukemia analysis lacks a large, diverse multi-task dataset, while existing small datasets lack domain diversity, limiting real-world applicability. To overcome dataset challenges, we present a large-scale WBC dataset named ‘Large Leukemia Dataset’ (LLD) and novel methods for detecting WBC with their attributes. Our contribution here is threefold. First, we present a large-scale Leukemia dataset collected through Peripheral Blood Films (PBF) from 48 patients, through multiple microscopes, multi-cameras, and multi-magnification. To enhance diagnosis explainability and medical expert acceptance, each leukemia cell is annotated at 100x with 7 morphological attributes, ranging from Cell Size to Nuclear Shape. Secondly, we propose a multi-task model that not only detects WBCs but also predicts their attributes, providing an interpretable and clinically meaningful solution. Third, we propose a method for WBC detection with attribute analysis using sparse annotations. This approach reduces the annotation burden on hematologists, requiring them to mark only a small area within the field of view. Our method enables the model to leverage the entire field of view rather than just the annotated regions, enhancing learning efficiency and diagnostic accuracy. From diagnosis explainability to overcoming domain-shift challenges, the presented datasets can be used for many challenging aspects of microscopic image analysis. The datasets, code, and demo are available at: <span><span>https://im.itu.edu.pk/sparse-leukemiaattri/</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103760"},"PeriodicalIF":11.8,"publicationDate":"2025-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144899195","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":"Completing spatial transcriptomics data for gene expression prediction benchmarking","authors":"Daniela Ruiz,&nbsp;Paula Cárdenas,&nbsp;Leonardo Manrique,&nbsp;Daniela Vega,&nbsp;Gabriel M. Mejia,&nbsp;Pablo Arbeláez","doi":"10.1016/j.media.2025.103754","DOIUrl":"10.1016/j.media.2025.103754","url":null,"abstract":"<div><div>Spatial Transcriptomics is a groundbreaking technology that integrates histology images with spatially resolved gene expression profiles. Among the various Spatial Transcriptomics techniques available, Visium has emerged as the most widely adopted. However, its accessibility is limited by high costs, the need for specialized expertise, and slow clinical integration. Additionally, gene capture inefficiencies lead to significant dropout, corrupting acquired data. To address these challenges, the deep learning community has explored the gene expression prediction task directly from histology images. Yet, inconsistencies in datasets, preprocessing, and training protocols hinder fair comparisons between models. To bridge this gap, we introduce SpaRED, a systematically curated database comprising 26 public datasets, providing a standardized resource for model evaluation. We further propose SpaCKLE, a state-of-the-art transformer-based gene expression completion model that reduces mean squared error by over 82.5% compared to existing approaches. Finally, we establish the SpaRED benchmark, evaluating eight state-of-the-art prediction models on both raw and SpaCKLE-completed data, demonstrating SpaCKLE substantially improves the results across all the gene expression prediction models. Altogether, our contributions constitute the most comprehensive benchmark of gene expression prediction from histology images to date and a stepping stone for future research on Spatial Transcriptomics.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103754"},"PeriodicalIF":11.8,"publicationDate":"2025-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144899197","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":"MSARAE: Multiscale adversarial regularized autoencoders for cortical network classification","authors":"Yihui Zhu ,&nbsp;Yue Zhou ,&nbsp;Xiaotong Zhang ,&nbsp;Yueying Li ,&nbsp;Yonggui Yuan ,&nbsp;Youyong Kong","doi":"10.1016/j.media.2025.103775","DOIUrl":"10.1016/j.media.2025.103775","url":null,"abstract":"<div><div>Due to privacy regulations and technical limitations, current research on the cerebral cortex frequently faces challenges, including limited data availability. The number of samples significantly influences the performance and generalization ability of deep learning models. In general, these models require sufficient training data to effectively learn underlying distributions and features, enabling strong performance on unseen samples. A limited sample size can lead to overfitting, thereby weakening the model’s generalizability. To address these challenges from a data augmentation perspective, we propose a Multi-Scale Adversarial Regularized Autoencoder (MSARAE) for augmenting and classifying cortical structural connectivity. The approach begins with data preprocessing and the construction of cortical structural connectivity networks. To better capture cortical features, the model leverages Laplacian eigenvectors to enhance topological information. Structural connectivity is then generated using variational autoencoders, with multi-scale graph convolutional layers serving as encoders to capture graph representations at different scales. An adversarial regularization mechanism is introduced to minimize the distribution discrepancy in the latent space. By training a discriminator, the model encourages the encoder to produce latent representations that closely match the distribution of real data, thereby improving its representational capacity. Finally, extensive experiments on the major depression disorder (MDD) dataset, the Human Connectome Project (HCP) dataset, and the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset demonstrated the superiority of the model.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"107 ","pages":"Article 103775"},"PeriodicalIF":11.8,"publicationDate":"2025-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144895890","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}