Computerized Medical Imaging and Graphics最新文献_第6页

Deep spatiotemporal clutter filtering of transthoracic echocardiographic images: Leveraging contextual attention and residual learning 经胸超声心动图图像的深时空杂波滤波：利用上下文注意和残余学习。

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Computerized Medical Imaging and Graphics

Pub Date : 2025-11-28 DOI: 10.1016/j.compmedimag.2025.102665

Mahdi Tabassian , Somayeh Akbari , Sandro Queirós , Lamia Al Saikhan , Jan D’hooge

This study presents a deep autoencoder network for filtering reverberation clutter from transthoracic echocardiographic (TTE) images. Given the spatiotemporal nature of this type of clutter, the filtering network employs 3D convolutional layers to suppress it throughout the cardiac cycle. The design of the network incorporates two key features that contribute to the effectiveness of the filter: (1) an attention mechanism for focusing on cluttered regions and leveraging contextual information, and (2) residual learning for preserving fine image structures. A diverse set of artifact patterns was simulated and superimposed onto ultra-realistic synthetic TTE sequences from six ultrasound vendors, generating input for the filtering network. The corresponding artifact-free sequences served as ground-truth. The performance of the filtering network was evaluated using unseen synthetic and in vivo artifactual sequences. Results from the in vivo dataset confirmed the network’s strong generalization capabilities, despite being trained solely on synthetic data and simulated artifacts. The suitability of the filtered sequences for downstream processing was assessed by computing segmental strain curves. A significant reduction in the discrepancy between the strain profiles of the cluttered and clutter-free segments was observed after filtering. The trained network processes a TTE sequence in a fraction of a second, enabling real-time clutter filtering and potentially improving the precision of clinically relevant indices derived from TTE sequences. The source code of the proposed method and example video files of the filtering results are available at: https://github.com/MahdiTabassian/Deep-Clutter-Filtering/tree/main.

本研究提出了一种深度自编码器网络，用于过滤经胸超声心动图（TTE）图像中的混响杂波。考虑到这种杂波的时空性质，滤波网络采用3D卷积层在整个心脏周期内抑制它。该网络的设计包含两个关键特征，有助于过滤器的有效性：(1)关注混乱区域并利用上下文信息的注意机制，以及(2)残差学习以保留精细图像结构。不同的伪影模式被模拟并叠加到来自六个超声供应商的超逼真合成TTE序列上，为滤波网络生成输入。相应的无伪影序列作为基真值。使用未见的合成和活体人工序列来评估滤波网络的性能。活体数据集的结果证实了该网络强大的泛化能力，尽管仅在合成数据和模拟工件上进行训练。通过计算分段应变曲线来评估过滤序列对下游加工的适用性。过滤后，观察到杂乱和无杂乱段的应变曲线之间的差异显著减小。训练后的网络在不到一秒的时间内处理TTE序列，实现实时杂波过滤，并有可能提高从TTE序列中获得的临床相关指标的精度。所提出的方法的源代码和过滤结果的示例视频文件可在：https://github.com/MahdiTabassian/Deep-Clutter-Filtering/tree/main。

{"title":"Deep spatiotemporal clutter filtering of transthoracic echocardiographic images: Leveraging contextual attention and residual learning","authors":"Mahdi Tabassian , Somayeh Akbari , Sandro Queirós , Lamia Al Saikhan , Jan D’hooge","doi":"10.1016/j.compmedimag.2025.102665","DOIUrl":"10.1016/j.compmedimag.2025.102665","url":null,"abstract":"<div><div>This study presents a deep autoencoder network for filtering reverberation clutter from transthoracic echocardiographic (TTE) images. Given the spatiotemporal nature of this type of clutter, the filtering network employs 3D convolutional layers to suppress it throughout the cardiac cycle. The design of the network incorporates two key features that contribute to the effectiveness of the filter: (1) an <em>attention mechanism</em> for focusing on cluttered regions and leveraging contextual information, and (2) <em>residual learning</em> for preserving fine image structures. A diverse set of artifact patterns was simulated and superimposed onto ultra-realistic synthetic TTE sequences from six ultrasound vendors, generating input for the filtering network. The corresponding artifact-free sequences served as ground-truth. The performance of the filtering network was evaluated using unseen synthetic and <em>in vivo</em> artifactual sequences. Results from the <em>in vivo</em> dataset confirmed the network’s strong generalization capabilities, despite being trained solely on synthetic data and simulated artifacts. The suitability of the filtered sequences for downstream processing was assessed by computing segmental strain curves. A significant reduction in the discrepancy between the strain profiles of the cluttered and clutter-free segments was observed after filtering. The trained network processes a TTE sequence in a fraction of a second, enabling real-time clutter filtering and potentially improving the precision of clinically relevant indices derived from TTE sequences. The source code of the proposed method and example video files of the filtering results are available at: <span><span>https://github.com/MahdiTabassian/Deep-Clutter-Filtering/tree/main</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"127 ","pages":"Article 102665"},"PeriodicalIF":4.9,"publicationDate":"2025-11-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145662682","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}

引用次数: 0

BoneVisionNet: A deep learning approach for the classification of bone tumours from radiographs using a triple fusion attention network of transformer and CNNs with XAI visualizations BoneVisionNet：一种用于从x线照片中分类骨肿瘤的深度学习方法，使用变压器和cnn的三重融合关注网络与XAI可视化

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Computerized Medical Imaging and Graphics

Pub Date : 2025-11-19 DOI: 10.1016/j.compmedimag.2025.102668

Armaano Ajay , Karthik R , Akshaj Singh Bisht , Pranav Uppuluri

Diagnosis of bone tumours present numerous challenges due to the complexity of pathology and varying morphologies of bone tumours. Current methods rely on manual techniques that are time-consuming and prone to errors. Hence, there is a need for more accurate and automated methods to assist medical professionals. The proposed work aims to solve this challenge by developing a deep learning-based architecture for bone tumour classification using radiographs. The proposed BoneVisionNet is developed using a combination of three specialized DL networks. The proposed approach leverages Convolution-Enhanced Image Transformer for global feature extraction which is further refined using a Global Context Block (GCB). In parallel, the Attention Boosted Mid-Level Feature Extraction Network (ABMLFE-Net) targets mid-level features and DenseNet-169 focuses on local feature extraction. The feature maps from the ABMLFE-Net and DenseNet-169 are fused using element-wise multiplication and is followed by an Efficient Channel Attention (ECA) layer for feature refinement. The global features that are refined by GCB are concatenated with the enhanced feature maps from the ECA layer, resulting in an refined multi-scale feature map. The BoneVisionNet attained an accuracy of 84.35 % when tested on the BTXRD dataset, outperforming CNN and transformer-based networks for classifying bone tumours from radiographs. To the best of our knowledge, this study represents the first application of a triple-track architecture for the classification of bone tumours from X-ray images. XAI visualisations using Grad-CAM, LIME, and SHAP help to further validate the performance of the model by ensuring transparency in the decision-making process.

由于骨肿瘤的病理复杂性和不同的形态，骨肿瘤的诊断提出了许多挑战。目前的方法依赖于手工技术，既耗时又容易出错。因此，需要更准确和自动化的方法来协助医疗专业人员。提出的工作旨在通过开发基于深度学习的架构来解决这一挑战，该架构用于使用x射线片进行骨肿瘤分类。提出的BoneVisionNet是使用三个专门的深度学习网络的组合开发的。该方法利用卷积增强图像转换器进行全局特征提取，并使用全局上下文块（GCB）进一步细化。同时，注意力增强中级特征提取网络（ABMLFE-Net）针对中级特征，DenseNet-169侧重于局部特征提取。来自ABMLFE-Net和DenseNet-169的特征映射使用元素智能乘法进行融合，然后是高效通道注意（ECA）层进行特征细化。将经过GCB优化的全局特征与来自ECA层的增强特征映射相连接，得到精细化的多尺度特征映射。在BTXRD数据集上进行测试时，BoneVisionNet的准确率达到了84.35 %，优于CNN和基于变压器的网络，可以从x线照片中对骨肿瘤进行分类。据我们所知，这项研究代表了从x射线图像中对骨肿瘤进行分类的三轨道结构的首次应用。使用Grad-CAM、LIME和SHAP的XAI可视化有助于通过确保决策过程的透明度来进一步验证模型的性能。

{"title":"BoneVisionNet: A deep learning approach for the classification of bone tumours from radiographs using a triple fusion attention network of transformer and CNNs with XAI visualizations","authors":"Armaano Ajay , Karthik R , Akshaj Singh Bisht , Pranav Uppuluri","doi":"10.1016/j.compmedimag.2025.102668","DOIUrl":"10.1016/j.compmedimag.2025.102668","url":null,"abstract":"<div><div>Diagnosis of bone tumours present numerous challenges due to the complexity of pathology and varying morphologies of bone tumours. Current methods rely on manual techniques that are time-consuming and prone to errors. Hence, there is a need for more accurate and automated methods to assist medical professionals. The proposed work aims to solve this challenge by developing a deep learning-based architecture for bone tumour classification using radiographs. The proposed BoneVisionNet is developed using a combination of three specialized DL networks. The proposed approach leverages Convolution-Enhanced Image Transformer for global feature extraction which is further refined using a Global Context Block (GCB). In parallel, the Attention Boosted Mid-Level Feature Extraction Network (ABMLFE-Net) targets mid-level features and DenseNet-169 focuses on local feature extraction. The feature maps from the ABMLFE-Net and DenseNet-169 are fused using element-wise multiplication and is followed by an Efficient Channel Attention (ECA) layer for feature refinement. The global features that are refined by GCB are concatenated with the enhanced feature maps from the ECA layer, resulting in an refined multi-scale feature map. The BoneVisionNet attained an accuracy of 84.35 % when tested on the BTXRD dataset, outperforming CNN and transformer-based networks for classifying bone tumours from radiographs. To the best of our knowledge, this study represents the first application of a triple-track architecture for the classification of bone tumours from X-ray images. XAI visualisations using Grad-CAM, LIME, and SHAP help to further validate the performance of the model by ensuring transparency in the decision-making process.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"127 ","pages":"Article 102668"},"PeriodicalIF":4.9,"publicationDate":"2025-11-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145618356","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}

引用次数: 0

Hallucinated domain generalization network with domain-aware dynamic representation for medical image segmentation 基于域感知动态表示的幻觉域泛化网络在医学图像分割中的应用。

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Computerized Medical Imaging and Graphics

Pub Date : 2025-11-17 DOI: 10.1016/j.compmedimag.2025.102670

Minjun Wang, Houjin Chen, Yanfeng Li, Jia Sun, Luyifu Chen, Peng Liang

Due to variations in medical image acquisition protocols, segmentation models often exhibit degraded performance when applied to unseen domains. We argue that such degradation primarily stems from overfitting to source domains and insufficient dynamic adaptability to target domains. To address this issue, we propose a hallucinated domain generalization network with domain-aware dynamic representation for medical image segmentation, which introduces a novel ”hallucination during training, dynamic representation during testing” scheme to effectively improve generalization. Specifically, we design an uncertainty-aware dynamic hallucination module that achieves adaptive transformation through Bézier curves and estimates potential domain shift by introducing the uncertainty-aware offset variable driven by channel-wise variance, generating diverse synthetic images. This approach breaks the limitations of source domain distributions while preserving original anatomical structures, effectively alleviating the model’s overfitting to the specific styles of source domains. Furthermore, we develop a domain-aware dynamic representation module that treats source domain knowledge as a foundation for understanding unknown domains. Concretely, we obtain unbiased estimates of global style prototypes through domain-wise statistical aggregation and the momentum update strategy. Then, input features are mapped to the unified source domain space through global style prototypes and similarity weights, mitigating performance degradation caused by domain shift during the testing phase. Extensive experiments on four heterogeneously distributed fundus image datasets and six multi-center prostate MRI datasets demonstrate that our approach outperforms state-of-the-art methods.

由于医学图像采集协议的变化，分割模型在应用于未知领域时往往表现出性能下降。我们认为这种退化主要源于对源域的过度拟合和对目标域的动态适应性不足。针对这一问题，我们提出了一种具有领域感知动态表示的医学图像分割幻觉域泛化网络，该网络引入了一种新颖的“训练时产生幻觉，测试时动态表示”的方案，有效地提高了医学图像的泛化效果。具体来说，我们设计了一个不确定性感知的动态幻觉模块，通过bsamizier曲线实现自适应变换，并通过引入由信道方差驱动的不确定性感知偏移变量来估计潜在的域位移，生成不同的合成图像。该方法在保留原始解剖结构的同时，打破了源域分布的限制，有效缓解了模型对源域特定样式的过拟合。此外，我们开发了一个领域感知的动态表示模块，该模块将源领域知识作为理解未知领域的基础。具体而言，我们通过领域统计聚合和动量更新策略获得了全局样式原型的无偏估计。然后，通过全局样式原型和相似度权重将输入特征映射到统一的源域空间，减轻了在测试阶段因域漂移引起的性能下降。在四个非均匀分布的眼底图像数据集和六个多中心前列腺MRI数据集上进行的大量实验表明，我们的方法优于最先进的方法。

{"title":"Hallucinated domain generalization network with domain-aware dynamic representation for medical image segmentation","authors":"Minjun Wang, Houjin Chen, Yanfeng Li, Jia Sun, Luyifu Chen, Peng Liang","doi":"10.1016/j.compmedimag.2025.102670","DOIUrl":"10.1016/j.compmedimag.2025.102670","url":null,"abstract":"<div><div>Due to variations in medical image acquisition protocols, segmentation models often exhibit degraded performance when applied to unseen domains. We argue that such degradation primarily stems from overfitting to source domains and insufficient dynamic adaptability to target domains. To address this issue, we propose a hallucinated domain generalization network with domain-aware dynamic representation for medical image segmentation, which introduces a novel ”hallucination during training, dynamic representation during testing” scheme to effectively improve generalization. Specifically, we design an uncertainty-aware dynamic hallucination module that achieves adaptive transformation through Bézier curves and estimates potential domain shift by introducing the uncertainty-aware offset variable driven by channel-wise variance, generating diverse synthetic images. This approach breaks the limitations of source domain distributions while preserving original anatomical structures, effectively alleviating the model’s overfitting to the specific styles of source domains. Furthermore, we develop a domain-aware dynamic representation module that treats source domain knowledge as a foundation for understanding unknown domains. Concretely, we obtain unbiased estimates of global style prototypes through domain-wise statistical aggregation and the momentum update strategy. Then, input features are mapped to the unified source domain space through global style prototypes and similarity weights, mitigating performance degradation caused by domain shift during the testing phase. Extensive experiments on four heterogeneously distributed fundus image datasets and six multi-center prostate MRI datasets demonstrate that our approach outperforms state-of-the-art methods.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"126 ","pages":"Article 102670"},"PeriodicalIF":4.9,"publicationDate":"2025-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145566238","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}

引用次数: 0

PET/CT-based deep learning model predicts distant metastasis after SBRT for early-stage NSCLC: A multicenter study 基于PET/ ct的深度学习模型预测早期NSCLC SBRT后远处转移：一项多中心研究

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Computerized Medical Imaging and Graphics

Pub Date : 2025-11-15 DOI: 10.1016/j.compmedimag.2025.102663

Lu Yu , Shenlun Chen , Junyi Li , HeQing Yi , Jin Wang , Jianjiao Ni , Xiaoli Zheng , Hong Ge , Zhengfei Zhu , Ligang Xing , Petros Kalendralis , Leonard Wee , Andre Dekker , Zhen Zhang , Zhaoxiang Ye , Zhiyong Yuan

Distant metastasis (DM) is the most frequent recurrence mode following stereotactic body radiation therapy (SBRT) for early-stage non-small cell lung cancer (NSCLC). Assessing DM risk prior to treatment initiation is critical. This study aimed to develop and validate a deep learning fusion model, based on 18F-FDG PET/CT images, to predict DM risk. A total of 566 patients from 5 hospitals were allocated into a training set (n = 347), an internal test set (n = 139), and an external test set (n = 80). Deep learning features were extracted from CT, PET, and fusion images using a variational autoencoder. Metastasis-free survival prognostic models were developed via fully connected networks. The fusion model demonstrated superior predictive capability compared to the CT or PET models alone, achieving C-indices of 0.864 (training), 0.819 (internal test), and 0.782 (external test). The model successfully stratified patients into high- and low-risk groups with significantly differentiated MFS (e.g., training set: HR=8.425, p < 0.001; internal test set, HR=6.828, p < 0.001; external test set: HR=4.376, p = 0.011). It was identified as an independent prognostic factor for MFS (HR=14.387, p < 0.001). In conclusions, the 18F-FDG PET/CT deep learning-based fusion model provides a robust prediction of distant metastasis risk and MFS in early-stage NSCLC patients receiving SBRT. This tool may offer objective data to inform individualized treatment decisions.

远处转移（DM）是早期非小细胞肺癌（NSCLC）立体定向全身放射治疗（SBRT）后最常见的复发方式。在开始治疗前评估糖尿病风险至关重要。本研究旨在开发并验证基于18F-FDG PET/CT图像的深度学习融合模型，以预测糖尿病风险。来自5家医院的566名患者被分配到一个训练集（n = 347）、一个内部测试集（n = 139）和一个外部测试集（n = 80）。使用变分自编码器从CT、PET和融合图像中提取深度学习特征。无转移生存预后模型是通过完全连接的网络开发的。与单独的CT或PET模型相比，融合模型显示出更好的预测能力，其c指数分别为0.864（训练）、0.819（内部测试）和0.782（外部测试）。该模型成功地将患者分为MFS差异显著的高危组和低危组(例如，训练集：HR=8.425, p

{"title":"PET/CT-based deep learning model predicts distant metastasis after SBRT for early-stage NSCLC: A multicenter study","authors":"Lu Yu , Shenlun Chen , Junyi Li , HeQing Yi , Jin Wang , Jianjiao Ni , Xiaoli Zheng , Hong Ge , Zhengfei Zhu , Ligang Xing , Petros Kalendralis , Leonard Wee , Andre Dekker , Zhen Zhang , Zhaoxiang Ye , Zhiyong Yuan","doi":"10.1016/j.compmedimag.2025.102663","DOIUrl":"10.1016/j.compmedimag.2025.102663","url":null,"abstract":"<div><div>Distant metastasis (DM) is the most frequent recurrence mode following stereotactic body radiation therapy (SBRT) for early-stage non-small cell lung cancer (NSCLC). Assessing DM risk prior to treatment initiation is critical. This study aimed to develop and validate a deep learning fusion model, based on 18F-FDG PET/CT images, to predict DM risk. A total of 566 patients from 5 hospitals were allocated into a training set (n = 347), an internal test set (n = 139), and an external test set (n = 80). Deep learning features were extracted from CT, PET, and fusion images using a variational autoencoder. Metastasis-free survival prognostic models were developed via fully connected networks. The fusion model demonstrated superior predictive capability compared to the CT or PET models alone, achieving C-indices of 0.864 (training), 0.819 (internal test), and 0.782 (external test). The model successfully stratified patients into high- and low-risk groups with significantly differentiated MFS (e.g., training set: HR=8.425, p < 0.001; internal test set, HR=6.828, p < 0.001; external test set: HR=4.376, p = 0.011). It was identified as an independent prognostic factor for MFS (HR=14.387, p < 0.001). In conclusions, the 18F-FDG PET/CT deep learning-based fusion model provides a robust prediction of distant metastasis risk and MFS in early-stage NSCLC patients receiving SBRT. This tool may offer objective data to inform individualized treatment decisions.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"126 ","pages":"Article 102663"},"PeriodicalIF":4.9,"publicationDate":"2025-11-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145558111","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}

引用次数: 0

AtlasSeg: Atlas prior guided dual-U-Net for tissue segmentation in fetal brain MRI Atlas预先引导双u - net用于胎儿脑MRI组织分割

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Computerized Medical Imaging and Graphics

Pub Date : 2025-11-15 DOI: 10.1016/j.compmedimag.2025.102662

Haoan Xu , Tianshu Zheng , Xinyi Xu , Yao Shen , Jiwei Sun , Cong Sun , Guangbin Wang , Zhaopeng Cui , Dan Wu

Accurate automatic tissue segmentation in fetal brain MRI is a crucial step in clinical diagnosis but remains challenging, particularly due to the dynamically changing anatomy and tissue contrast during fetal development. Existing segmentation networks can only implicitly learn age-related features, leading to a decline in accuracy at extreme gestational ages (GAs). To improve segmentation performance throughout gestation, we introduce AtlasSeg, a dual-U-shape convolution network that explicitly integrates GA-specific information as guidance. By providing a publicly available fetal brain atlas with segmentation labels corresponding to relevant GAs, AtlasSeg effectively extracts age-specific patterns in the atlas branch and generates precise tissue segmentation in the segmentation branch. Multi-scale spatial attention feature fusions are constructed during both encoding and decoding stages to enhance feature flow and facilitate better information interactions between two branches. We compared AtlasSeg with ten well-established networks with a seven-tissue segmentation task in our in-house and two public datasets, achieving the highest average Dice similarity coefficient. The improvement was particularly evident in extreme early or late GA cases, where training data was scare. Furthermore, AtlasSeg exhibited minimal performance degradation on low-quality images with contrast changes and noise, attributed to its anatomical shape priors. Overall, AtlasSeg demonstrated enhanced segmentation accuracy, better consistency across fetal ages, and robustness to perturbations, making it a powerful tool for reliable fetal brain MRI tissue segmentation, particularly suited for diagnostic assessments during early gestation.

胎儿脑MRI准确的自动组织分割是临床诊断的关键一步，但仍然具有挑战性，特别是由于胎儿发育过程中解剖结构和组织对比度的动态变化。现有的分割网络只能隐式学习与年龄相关的特征，导致极端胎龄（GAs）的准确性下降。为了提高整个妊娠期的分割性能，我们引入了AtlasSeg，这是一个双u形卷积网络，它明确地集成了ga特定的信息作为指导。AtlasSeg通过提供公开可用的胎儿脑图谱，并提供相应GAs的分割标签，有效地提取图谱分支中的年龄特异性模式，并在分割分支中生成精确的组织分割。在编码和解码阶段构建多尺度空间注意特征融合，增强特征流，促进两个分支之间更好的信息交互。我们将atlseg与10个完善的网络进行了比较，其中包括我们内部的7个组织分割任务和两个公共数据集，获得了最高的平均Dice相似系数。在训练数据匮乏的极端早期或晚期GA案例中，这种改善尤为明显。此外，由于其解剖形状先验，atlseg在具有对比度变化和噪声的低质量图像上表现出最小的性能下降。总体而言，AtlasSeg显示出更高的分割准确性，更好的跨胎龄一致性和对扰动的稳健性，使其成为可靠的胎儿脑MRI组织分割的强大工具，特别适用于妊娠早期的诊断评估。

{"title":"AtlasSeg: Atlas prior guided dual-U-Net for tissue segmentation in fetal brain MRI","authors":"Haoan Xu , Tianshu Zheng , Xinyi Xu , Yao Shen , Jiwei Sun , Cong Sun , Guangbin Wang , Zhaopeng Cui , Dan Wu","doi":"10.1016/j.compmedimag.2025.102662","DOIUrl":"10.1016/j.compmedimag.2025.102662","url":null,"abstract":"<div><div>Accurate automatic tissue segmentation in fetal brain MRI is a crucial step in clinical diagnosis but remains challenging, particularly due to the dynamically changing anatomy and tissue contrast during fetal development. Existing segmentation networks can only implicitly learn age-related features, leading to a decline in accuracy at extreme gestational ages (GAs). To improve segmentation performance throughout gestation, we introduce AtlasSeg, a dual-U-shape convolution network that explicitly integrates GA-specific information as guidance. By providing a publicly available fetal brain atlas with segmentation labels corresponding to relevant GAs, AtlasSeg effectively extracts age-specific patterns in the atlas branch and generates precise tissue segmentation in the segmentation branch. Multi-scale spatial attention feature fusions are constructed during both encoding and decoding stages to enhance feature flow and facilitate better information interactions between two branches. We compared <strong>AtlasSeg</strong> with ten well-established networks with a seven-tissue segmentation task in our in-house and two public datasets, achieving the highest average Dice similarity coefficient. The improvement was particularly evident in extreme early or late GA cases, where training data was scare. Furthermore, AtlasSeg exhibited minimal performance degradation on low-quality images with contrast changes and noise, attributed to its anatomical shape priors. Overall, AtlasSeg demonstrated enhanced segmentation accuracy, better consistency across fetal ages, and robustness to perturbations, making it a powerful tool for <strong>reliable fetal brain MRI tissue segmentation,</strong> particularly suited for <strong>diagnostic assessments</strong> during <strong>early gestation.</strong></div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"126 ","pages":"Article 102662"},"PeriodicalIF":4.9,"publicationDate":"2025-11-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145571229","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}

引用次数: 0

Colorectal disease diagnosis with deep triple-stream fusion and attention refinement 结直肠疾病的深三流融合与注意精细化诊断

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Computerized Medical Imaging and Graphics

Pub Date : 2025-11-15 DOI: 10.1016/j.compmedimag.2025.102669

Abdulfattah Ba Alawi , Abdullah Ammar Karcioglu , Ferhat Bozkurt

Colorectal cancer constitutes a significant proportion of global cancer-related mortality, underscoring the imperative for robust and early-stage diagnostic methodologies. In this study, we propose a novel end-to-end deep learning framework that integrates multiple advanced mechanisms to enhance the classification of colorectal disease from histopathologic and endoscopic images. Our model, named TripleFusionNet, leverages a unique triple-stream architecture by combining the strengths of EfficientNetB3, ResNet50, and DenseNet121, enabling the extraction of rich, multi-level feature representations from input images. To augment discriminative feature modeling, a Multi-Scale Attention Module is integrated, which concurrently performs spatial and channel-wise recalibration, thereby enabling the network to emphasize diagnostically salient regions. Additionally, we incorporate a Squeeze-Excite Refinement Block (SERB) to selectively enhance informative channel activations while attenuating noise and redundant signals. Feature representations from the individual backbones are adaptively fused through a Progressive Gated Fusion mechanism that dynamically learns context-aware weighting for optimal feature integration and redundancy mitigation. We validate our approach on two colorectal benchmarks: CRCCD_V1 (14 classes) and LC25000 (binary). On CRCCD_V1, the best performance is obtained by a conventional classifier trained on our 256-D TripleFusionNet embeddings—SVM (RBF) reaches 96.63% test accuracy with macro F1 96.62%, with the Stacking Ensemble close behind. With five-fold cross-validation, it yields comparable out-of-fold means (0.964 with small standard deviations), confirming stability across partitions. End-to-end image-based baselines, including TripleFusionNet, are competitive but are slightly surpassed by embedding-based classifiers, highlighting the utility of the learned representation. On LC25000, our method attains 100% accuracy. Beyond accuracy, the approach maintains strong precision, recall, F1, and ROC–AUC, and the fused embeddings transfer effectively to multiple conventional learners (e.g., Random Forest, XGBoost). These results confirm the potential of the model for real-world deployment in computer-aided diagnosis workflows, particularly within resource-constrained clinical settings.

结直肠癌在全球癌症相关死亡率中占很大比例，因此需要强有力的早期诊断方法。在这项研究中，我们提出了一个新的端到端深度学习框架，该框架集成了多种先进的机制，以增强从组织病理学和内窥镜图像中对结直肠疾病的分类。我们的模型名为TripleFusionNet，通过结合EfficientNetB3、ResNet50和DenseNet121的优势，利用独特的三流架构，能够从输入图像中提取丰富的、多层次的特征表示。为了增强判别特征建模，集成了一个多尺度注意力模块，该模块同时执行空间和通道重新校准，从而使网络能够强调诊断上的显著区域。此外，我们还结合了一个压缩激励细化块（塞族），以选择性地增强信息通道激活，同时衰减噪声和冗余信号。来自各个主干网的特征表示通过渐进门控融合机制自适应融合，该机制动态学习上下文感知权重，以实现最佳特征集成和冗余缓解。我们在两个结肠直肠基准上验证了我们的方法：CRCCD_V1（14类）和LC25000（二进制）。在CRCCD_V1上，在我们的256-D TripleFusionNet嵌入上训练的传统分类器-支持向量机（RBF）的测试准确率达到96.63%，宏F1为96.62%，Stacking Ensemble紧随其后。通过五倍交叉验证，它产生了可比较的叠外均值（0.964，标准差较小），确认了跨分区的稳定性。端到端基于图像的基线，包括TripleFusionNet，是有竞争力的，但被基于嵌入的分类器稍微超越，突出了学习表征的实用性。在LC25000上，我们的方法达到了100%的准确率。除了准确性之外，该方法还保持了很强的精度、召回率、F1和ROC-AUC，并且融合的嵌入有效地转移到多个传统的学习器（例如，Random Forest, XGBoost）。这些结果证实了该模型在计算机辅助诊断工作流程中的实际应用潜力，特别是在资源有限的临床环境中。

{"title":"Colorectal disease diagnosis with deep triple-stream fusion and attention refinement","authors":"Abdulfattah Ba Alawi , Abdullah Ammar Karcioglu , Ferhat Bozkurt","doi":"10.1016/j.compmedimag.2025.102669","DOIUrl":"10.1016/j.compmedimag.2025.102669","url":null,"abstract":"<div><div>Colorectal cancer constitutes a significant proportion of global cancer-related mortality, underscoring the imperative for robust and early-stage diagnostic methodologies. In this study, we propose a novel end-to-end deep learning framework that integrates multiple advanced mechanisms to enhance the classification of colorectal disease from histopathologic and endoscopic images. Our model, named <strong>TripleFusionNet</strong>, leverages a unique triple-stream architecture by combining the strengths of EfficientNetB3, ResNet50, and DenseNet121, enabling the extraction of rich, multi-level feature representations from input images. To augment discriminative feature modeling, a <em>Multi-Scale Attention Module</em> is integrated, which concurrently performs spatial and channel-wise recalibration, thereby enabling the network to emphasize diagnostically salient regions. Additionally, we incorporate a <em>Squeeze-Excite Refinement Block (SERB)</em> to selectively enhance informative channel activations while attenuating noise and redundant signals. Feature representations from the individual backbones are adaptively fused through a <em>Progressive Gated Fusion mechanism</em> that dynamically learns context-aware weighting for optimal feature integration and redundancy mitigation. We validate our approach on two colorectal benchmarks: CRCCD_V1 (14 classes) and LC25000 (binary). On CRCCD_V1, the best performance is obtained by a conventional classifier trained on our 256-D <em>TripleFusionNet</em> embeddings—SVM (RBF) reaches <strong>96.63%</strong> test accuracy with macro F1 <strong>96.62%</strong>, with the Stacking Ensemble close behind. With five-fold cross-validation, it yields comparable out-of-fold means (<strong>0.964</strong> with small standard deviations), confirming stability across partitions. End-to-end image-based baselines, including <em>TripleFusionNet</em>, are competitive but are slightly surpassed by embedding-based classifiers, highlighting the utility of the learned representation. On LC25000, our method attains <strong>100%</strong> accuracy. Beyond accuracy, the approach maintains strong precision, recall, F1, and ROC–AUC, and the fused embeddings transfer effectively to multiple conventional learners (e.g., Random Forest, XGBoost). These results confirm the potential of the model for real-world deployment in computer-aided diagnosis workflows, particularly within resource-constrained clinical settings.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"126 ","pages":"Article 102669"},"PeriodicalIF":4.9,"publicationDate":"2025-11-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145520750","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}

引用次数: 0

Anatomy-informed deep learning and radiomics for neurofibroma segmentation in whole-body MRI 全身MRI中神经纤维瘤分割的解剖学信息深度学习和放射组学

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Computerized Medical Imaging and Graphics

Pub Date : 2025-11-14 DOI: 10.1016/j.compmedimag.2025.102667

Georgii Kolokolnikov , Marie-Lena Schmalhofer , Lennart Well , Said Farschtschi , Victor-Felix Mautner , Inka Ristow , René Werner

Background and Objectives:

Neurofibromatosis type 1 (NF1) is a genetic disorder characterized by the development of multiple neurofibromas (NFs) throughout the body. Accurate segmentation of these tumors in whole-body magnetic resonance imaging (WB-MRI) is critical for quantifying tumor burden and clinical decision-making. This study aims to develop a pipeline for NF segmentation in fat-suppressed T2-weighted WB-MRI that incorporates anatomical context and radiomics to improve accuracy and specificity.

Methods:

The proposed pipeline consists of three stages: (1) anatomy segmentation using MRSegmentator and refinement with a high-risk NF zone; (2) NF segmentation using an ensemble of 3D anisotropic anatomy-informed U-Nets; and (3) tumor candidate classification using radiomic features to filter false positives. The study used 109 WB-MRI scans from 74 NF1 patients, divided into training and three test sets representing in-domain (3T), domain-shifted (1.5T), and low tumor burden scenarios. Evaluation metrics included per-scan and per-tumor Dice Similarity Coefficient (DSC), Volume Overlap Error (VOE), Absolute Relative Volume Difference (ARVD), and per-scan F1 score. Statistical significance was assessed using Wilcoxon signed-rank tests with Bonferroni correction.

Results:

On the in-domain test set, the proposed ensemble of 3D anisotropic anatomy-informed U-Nets with tumor candidate classification achieved a per-scan DSC of 0.64, outperforming 2D nnU-Net (DSC: 0.52) and 3D full-resolution nnU-Net (DSC: 0.54). Performance was maintained on the domain-shift test set (DSC: 0.51) but declined on low tumor burden cases (DSC: 0.23). Preliminary inter-reader variability analysis showed model-to-expert agreement (DSC: 0.67–0.69) comparable to inter-expert agreement (DSC: 0.69).

Conclusions:

The proposed pipeline achieves the highest performance among established methods for automated NF segmentation in WB-MRI and approaches expert-level consistency. The integration of anatomical context and radiomics enhances robustness. Nonetheless, segmentation performance decreases in low tumor burden scenarios, indicating a key area for future methodological improvements. Additionally, the limited inter-reader agreement observed among experts underscores the inherent complexity and ambiguity of the NF segmentation task.

背景和目的：1型神经纤维瘤病（NF1）是一种以全身多发性神经纤维瘤（NFs）发展为特征的遗传性疾病。在全身磁共振成像（WB-MRI）中准确分割这些肿瘤对于量化肿瘤负担和临床决策至关重要。本研究旨在开发一种在脂肪抑制的t2加权WB-MRI中进行NF分割的管道，该管道结合解剖学背景和放射组学来提高准确性和特异性。方法：该流程包括三个阶段：(1)使用MRSegmentator进行解剖分割，并使用高危NF区进行细化；(2)利用三维各向异性的U-Nets集合进行NF分割；(3)利用放射学特征对候选肿瘤进行分类，过滤假阳性。该研究使用74例NF1患者的109张WB-MRI扫描，分为训练集和三个测试集，分别代表域内（3T）、域移位（1.5T）和低肿瘤负荷情景。评估指标包括每次扫描和每个肿瘤骰子相似系数（DSC）、体积重叠误差（VOE）、绝对相对体积差（ARVD）和每次扫描F1评分。采用Wilcoxon符号秩检验和Bonferroni校正评估统计学显著性。结果：在域内测试集上，所提出的具有候选肿瘤分类的三维各向异性解剖学信息的U-Nets集合的每次扫描DSC为0.64，优于2D nnU-Net （DSC: 0.52）和3D全分辨率nnU-Net （DSC: 0.54）。在域移测试集上表现良好（DSC: 0.51），但在低肿瘤负荷情况下表现下降（DSC: 0.23）。初步的读者间变异性分析显示，模型与专家的一致性（DSC: 0.67-0.69）与专家间的一致性（DSC: 0.69）相当。结论：所提出的管道在现有的WB-MRI自动NF分割方法中达到了最高的性能，并达到了专家级的一致性。解剖学背景和放射组学的整合增强了鲁棒性。尽管如此，在低肿瘤负荷情况下，分割性能下降，这表明了未来方法改进的关键领域。此外，在专家之间观察到的有限的读者间协议强调了NF分词任务固有的复杂性和模糊性。

{"title":"Anatomy-informed deep learning and radiomics for neurofibroma segmentation in whole-body MRI","authors":"Georgii Kolokolnikov , Marie-Lena Schmalhofer , Lennart Well , Said Farschtschi , Victor-Felix Mautner , Inka Ristow , René Werner","doi":"10.1016/j.compmedimag.2025.102667","DOIUrl":"10.1016/j.compmedimag.2025.102667","url":null,"abstract":"<div><h3>Background and Objectives:</h3><div>Neurofibromatosis type 1 (NF1) is a genetic disorder characterized by the development of multiple neurofibromas (NFs) throughout the body. Accurate segmentation of these tumors in whole-body magnetic resonance imaging (WB-MRI) is critical for quantifying tumor burden and clinical decision-making. This study aims to develop a pipeline for NF segmentation in fat-suppressed T2-weighted WB-MRI that incorporates anatomical context and radiomics to improve accuracy and specificity.</div></div><div><h3>Methods:</h3><div>The proposed pipeline consists of three stages: (1) anatomy segmentation using MRSegmentator and refinement with a high-risk NF zone; (2) NF segmentation using an ensemble of 3D anisotropic anatomy-informed U-Nets; and (3) tumor candidate classification using radiomic features to filter false positives. The study used 109 WB-MRI scans from 74 NF1 patients, divided into training and three test sets representing in-domain (3T), domain-shifted (1.5T), and low tumor burden scenarios. Evaluation metrics included per-scan and per-tumor Dice Similarity Coefficient (DSC), Volume Overlap Error (VOE), Absolute Relative Volume Difference (ARVD), and per-scan F1 score. Statistical significance was assessed using Wilcoxon signed-rank tests with Bonferroni correction.</div></div><div><h3>Results:</h3><div>On the in-domain test set, the proposed ensemble of 3D anisotropic anatomy-informed U-Nets with tumor candidate classification achieved a per-scan DSC of 0.64, outperforming 2D nnU-Net (DSC: 0.52) and 3D full-resolution nnU-Net (DSC: 0.54). Performance was maintained on the domain-shift test set (DSC: 0.51) but declined on low tumor burden cases (DSC: 0.23). Preliminary inter-reader variability analysis showed model-to-expert agreement (DSC: 0.67–0.69) comparable to inter-expert agreement (DSC: 0.69).</div></div><div><h3>Conclusions:</h3><div>The proposed pipeline achieves the highest performance among established methods for automated NF segmentation in WB-MRI and approaches expert-level consistency. The integration of anatomical context and radiomics enhances robustness. Nonetheless, segmentation performance decreases in low tumor burden scenarios, indicating a key area for future methodological improvements. Additionally, the limited inter-reader agreement observed among experts underscores the inherent complexity and ambiguity of the NF segmentation task.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"126 ","pages":"Article 102667"},"PeriodicalIF":4.9,"publicationDate":"2025-11-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145520826","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}

引用次数: 0

DuetMatch: Harmonizing semi-supervised brain MRI segmentation via decoupled branch optimization DuetMatch：通过解耦分支优化协调半监督脑MRI分割。

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Computerized Medical Imaging and Graphics

Pub Date : 2025-11-11 DOI: 10.1016/j.compmedimag.2025.102666

Thanh-Huy Nguyen , Hoang-Thien Nguyen , Vi Vu , Ba-Thinh Lam , Phat Huynh , Tianyang Wang , Xingjian Li , Ulas Bagci , Min Xu

The limited availability of annotated data in medical imaging makes semi-supervised learning increasingly appealing for its ability to learn from imperfect supervision. Recently, teacher-student frameworks have gained popularity for their training benefits and robust performance. However, jointly optimizing the entire network can hinder convergence and stability, especially in challenging scenarios. To address this for medical image segmentation, we propose DuetMatch, a novel dual-branch semi-supervised framework with asynchronous optimization, where each branch optimizes either the encoder or decoder while keeping the other frozen. To improve consistency under noisy conditions, we introduce Decoupled Dropout Perturbation, enforcing regularization across branches. We also design Pairwise CutMix Cross-Guidance to enhance model diversity by exchanging pseudo-labels through augmented input pairs. To mitigate confirmation bias from noisy pseudo-labels, we propose Consistency Matching, refining labels using stable predictions from frozen teacher models. Extensive experiments on benchmark brain MRI segmentation datasets, including ISLES2022 and BraTS, show that DuetMatch consistently outperforms state-of-the-art methods, demonstrating its effectiveness and robustness across diverse semi-supervised segmentation scenarios.

医学影像中标注数据的有限可用性使得半监督学习越来越有吸引力，因为它能够从不完善的监督中学习。最近，师生框架因其培训优势和强大的性能而受到欢迎。然而，联合优化整个网络可能会影响收敛性和稳定性，特别是在具有挑战性的场景下。为了解决医学图像分割中的这个问题，我们提出了DuetMatch，这是一种具有异步优化的新型双分支半监督框架，其中每个分支优化编码器或解码器，同时保持另一个分支冻结。为了提高噪声条件下的一致性，我们引入了解耦的Dropout摄动，强制跨分支的正则化。我们还设计了Pairwise CutMix Cross-Guidance，通过增强输入对交换伪标签来增强模型的多样性。为了减轻来自噪声伪标签的确认偏差，我们提出一致性匹配，使用来自冻结教师模型的稳定预测来改进标签。在包括ISLES2022和BraTS在内的基准脑MRI分割数据集上进行的大量实验表明，DuetMatch始终优于最先进的方法，证明了其在各种半监督分割场景中的有效性和鲁棒性。

{"title":"DuetMatch: Harmonizing semi-supervised brain MRI segmentation via decoupled branch optimization","authors":"Thanh-Huy Nguyen , Hoang-Thien Nguyen , Vi Vu , Ba-Thinh Lam , Phat Huynh , Tianyang Wang , Xingjian Li , Ulas Bagci , Min Xu","doi":"10.1016/j.compmedimag.2025.102666","DOIUrl":"10.1016/j.compmedimag.2025.102666","url":null,"abstract":"<div><div>The limited availability of annotated data in medical imaging makes semi-supervised learning increasingly appealing for its ability to learn from imperfect supervision. Recently, teacher-student frameworks have gained popularity for their training benefits and robust performance. However, jointly optimizing the entire network can hinder convergence and stability, especially in challenging scenarios. To address this for medical image segmentation, we propose <em>DuetMatch</em>, a novel dual-branch semi-supervised framework with asynchronous optimization, where each branch optimizes either the encoder or decoder while keeping the other frozen. To improve consistency under noisy conditions, we introduce <strong>Decoupled Dropout Perturbation</strong>, enforcing regularization across branches. We also design <strong>Pairwise CutMix Cross-Guidance</strong> to enhance model diversity by exchanging pseudo-labels through augmented input pairs. To mitigate confirmation bias from noisy pseudo-labels, we propose <strong>Consistency Matching</strong>, refining labels using stable predictions from frozen teacher models. Extensive experiments on benchmark brain MRI segmentation datasets, including ISLES2022 and BraTS, show that DuetMatch consistently outperforms state-of-the-art methods, demonstrating its effectiveness and robustness across diverse semi-supervised segmentation scenarios.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"126 ","pages":"Article 102666"},"PeriodicalIF":4.9,"publicationDate":"2025-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145558034","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}

引用次数: 0

CRAD: Cognitive Aware Feature Refinement with Missing Modalities for Early Alzheimer’s Progression Prediction 认知意识特征细化与缺失模式在早期阿尔茨海默病进展预测中的应用。

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Computerized Medical Imaging and Graphics

Pub Date : 2025-11-10 DOI: 10.1016/j.compmedimag.2025.102664

Fei Liu , Shiuan-Ni Liang , Mohamed Hisham Jaward , Huey Fang Ong , Huabin Wang , Alzheimer’s Disease Neuroimaging Initiative , Australian Imaging Biomarkers and Lifestyle flagship study of ageing

Accurate diagnosis and early prediction of Alzheimer’s disease (AD) often require multiple neuroimageing modalities, but in many cases, only one or two modalities are available. This missing modality hinders the accuracy of diagnosis and is a critical challenge in clinical practice. Multimodal knowledge distillation (KD) offers a promising solution by aligning complete knowledge from multimodal data with that of partial modalities. However, current methods focus on aligning high-level features, which limit their effectiveness due to insufficient transfer of reliable knowledge. In this work, we propose a novel Consistency Refinement-driven Multi-level Self-Attention Distillation framework (CRAD) for Early Alzheimer’s Progression Prediction, which enables the cross-modal transfer of more robust shallow knowledge with self-attention to refine features. We develop a multi-level distillation module to progressively distill cross-modal discriminating knowledge, enabling lightweight yet reliable knowledge transfer. Moreover, we design a novel self-attention distillation module (PF-CMAD) to transfer disease-relevant intermediate knowledge, which leverages feature self-similarity to capture cross-modal correlations without introducing trainable parameters, enabling interpretable and efficient distillation. We incorporate a consistency-evaluation-driven confidence regularization strategy within the distillation process. This strategy dynamically refines knowledge using adaptive distillation controllers that assess teacher confidence. Comprehensive experiments demonstrate that our method achieves superior accuracy and robust cross-dataset generalization performance using only MRI for AD diagnosis and early progression prediction. The code is available at https://github.com/LiuFei-AHU/CRAD.

阿尔茨海默病（AD）的准确诊断和早期预测通常需要多种神经成像方式，但在许多情况下，只有一种或两种方式可用。这种缺失的模式阻碍了诊断的准确性，是临床实践中的一个关键挑战。多模态知识蒸馏（KD）通过将来自多模态数据的完整知识与部分模态的知识进行比对，提供了一种很有前途的解决方案。然而，目前的方法主要集中在高级特征的对齐上，由于可靠知识的转移不足，限制了它们的有效性。在这项工作中，我们提出了一种新的一致性改进驱动的多层次自关注蒸馏框架（CRAD）用于早期阿尔茨海默氏症的进展预测，该框架能够跨模态转移更强大的具有自关注的浅层知识来改进特征。我们开发了一个多级蒸馏模块，逐步提取跨模态的鉴别知识，实现轻量级但可靠的知识转移。此外，我们设计了一种新的自关注蒸馏模块（PF-CMAD）来转移疾病相关的中间知识，该模块利用特征自相似性来捕获跨模态相关性，而不引入可训练的参数，从而实现可解释和高效的蒸馏。我们在蒸馏过程中加入了一致性评估驱动的置信度正则化策略。该策略使用评估教师信心的自适应蒸馏控制器动态地提炼知识。综合实验表明，该方法仅使用MRI进行AD诊断和早期进展预测，具有优越的准确性和鲁棒的跨数据集泛化性能。代码可在https://github.com/LiuFei-AHU/CRAD上获得。

{"title":"CRAD: Cognitive Aware Feature Refinement with Missing Modalities for Early Alzheimer’s Progression Prediction","authors":"Fei Liu , Shiuan-Ni Liang , Mohamed Hisham Jaward , Huey Fang Ong , Huabin Wang , Alzheimer’s Disease Neuroimaging Initiative , Australian Imaging Biomarkers and Lifestyle flagship study of ageing","doi":"10.1016/j.compmedimag.2025.102664","DOIUrl":"10.1016/j.compmedimag.2025.102664","url":null,"abstract":"<div><div>Accurate diagnosis and early prediction of Alzheimer’s disease (AD) often require multiple neuroimageing modalities, but in many cases, only one or two modalities are available. This missing modality hinders the accuracy of diagnosis and is a critical challenge in clinical practice. Multimodal knowledge distillation (KD) offers a promising solution by aligning complete knowledge from multimodal data with that of partial modalities. However, current methods focus on aligning high-level features, which limit their effectiveness due to insufficient transfer of reliable knowledge. In this work, we propose a novel Consistency Refinement-driven Multi-level Self-Attention Distillation framework (CRAD) for Early Alzheimer’s Progression Prediction, which enables the cross-modal transfer of more robust shallow knowledge with self-attention to refine features. We develop a multi-level distillation module to progressively distill cross-modal discriminating knowledge, enabling lightweight yet reliable knowledge transfer. Moreover, we design a novel self-attention distillation module (PF-CMAD) to transfer disease-relevant intermediate knowledge, which leverages feature self-similarity to capture cross-modal correlations without introducing trainable parameters, enabling interpretable and efficient distillation. We incorporate a consistency-evaluation-driven confidence regularization strategy within the distillation process. This strategy dynamically refines knowledge using adaptive distillation controllers that assess teacher confidence. Comprehensive experiments demonstrate that our method achieves superior accuracy and robust cross-dataset generalization performance using only MRI for AD diagnosis and early progression prediction. The code is available at <span><span>https://github.com/LiuFei-AHU/CRAD</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"126 ","pages":"Article 102664"},"PeriodicalIF":4.9,"publicationDate":"2025-11-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145507841","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}

引用次数: 0

Multi-representational deep transfer learning for classifying hemorrhagic metastases and non-neoplastic intracranial hematomas in multi-modal brain MRI scans 多表征深度迁移学习在多模态脑MRI扫描中分类出血性转移和非肿瘤性颅内血肿。

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Computerized Medical Imaging and Graphics

Pub Date : 2025-11-03 DOI: 10.1016/j.compmedimag.2025.102661

Luyue Yu , Linyang Cui , Jiachen Cui , Aixi Qu , Dexin Yu , Qiang Wu , Ju Liu

With an increasing incidence of malignant tumors, occurrence of brain metastases (BMs) has increased. BM represents the most common adult malignant brain tumors. BM is associated with hemorrhages, cystic necrosis, and calcification, which leads to significant diagnostic challenges when differentiating between hemorrhagic brain metastasis (HBM) and non-neoplastic intracranial hematomas (nn-ICH). This study addressed the limitations of small sample sizes, limited imaging features, and underutilized machine learning techniques reported in previous radiomic studies and introduced a novel multi-representation deep transfer learning (MRDTL) framework. Compared to existing radiomics feature analysis methods, MRDTL utilizes multi-modal MRI scans with two substantial merits: (1) A multi-representation fusion (MRF) module which extracted typical feature combinations by explicitly learning the complementarities between multi-modal sequences and multiple representations; (2) a neighborhood embedding (NE) module that measured metrics and clustering on cross-centric data to enhance transferable representations and improve model generalization. On the self-constructed HBMRI dataset, MRDTL outperformed five other baseline methods in AUC, F1-score, and accuracy. It improved accuracy to 94.5% and 93.5% in Co-site and Separate site testing, respectively, and overall provided more reliable diagnostic insights.

随着恶性肿瘤发病率的增加，脑转移瘤（brain metastasis, BMs）的发生率也在增加。脑脊髓瘤是最常见的成人恶性脑肿瘤。脑转移与出血、囊性坏死和钙化有关，这在区分出血性脑转移（HBM）和非肿瘤性颅内血肿（nn-ICH）时带来了重大的诊断挑战。本研究解决了以往放射学研究中样本量小、成像特征有限以及机器学习技术未充分利用的局限性，并引入了一种新的多表示深度迁移学习（MRDTL）框架。与现有的放射组学特征分析方法相比，MRDTL利用了多模态MRI扫描，具有两个显著优点：(1)多表征融合（MRF）模块，通过明确学习多模态序列和多个表征之间的互补性来提取典型特征组合；(2)邻域嵌入（NE）模块，该模块测量跨中心数据的度量和聚类，以增强可转移表示和提高模型泛化。在自行构建的HBMRI数据集上，MRDTL在AUC、f1评分和准确性方面优于其他5种基线方法。在联合位点和单独位点测试中，准确率分别提高到94.5%和93.5%，总体上提供了更可靠的诊断见解。

{"title":"Multi-representational deep transfer learning for classifying hemorrhagic metastases and non-neoplastic intracranial hematomas in multi-modal brain MRI scans","authors":"Luyue Yu , Linyang Cui , Jiachen Cui , Aixi Qu , Dexin Yu , Qiang Wu , Ju Liu","doi":"10.1016/j.compmedimag.2025.102661","DOIUrl":"10.1016/j.compmedimag.2025.102661","url":null,"abstract":"<div><div>With an increasing incidence of malignant tumors, occurrence of brain metastases (BMs) has increased. BM represents the most common adult malignant brain tumors. BM is associated with hemorrhages, cystic necrosis, and calcification, which leads to significant diagnostic challenges when differentiating between hemorrhagic brain metastasis (HBM) and non-neoplastic intracranial hematomas (nn-ICH). This study addressed the limitations of small sample sizes, limited imaging features, and underutilized machine learning techniques reported in previous radiomic studies and introduced a novel multi-representation deep transfer learning (MRDTL) framework. Compared to existing radiomics feature analysis methods, MRDTL utilizes multi-modal MRI scans with two substantial merits: (1) A multi-representation fusion (MRF) module which extracted typical feature combinations by explicitly learning the complementarities between multi-modal sequences and multiple representations; (2) a neighborhood embedding (NE) module that measured metrics and clustering on cross-centric data to enhance transferable representations and improve model generalization. On the self-constructed HBMRI dataset, MRDTL outperformed five other baseline methods in AUC, F1-score, and accuracy. It improved accuracy to 94.5% and 93.5% in Co-site and Separate site testing, respectively, and overall provided more reliable diagnostic insights.</div></div>","PeriodicalId":50631,"journal":{"name":"Computerized Medical Imaging and Graphics","volume":"126 ","pages":"Article 102661"},"PeriodicalIF":4.9,"publicationDate":"2025-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145460601","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}

引用次数: 0