Biomedical Signal Processing and Control最新文献_第9页

Deep learning for diagnosing autism spectrum disorder using EEG with upright and inverted face processing tasks 深度学习在自闭症谱系障碍诊断中的应用

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-01 Epub Date: 2026-02-03 DOI: 10.1016/j.bspc.2026.109655

Fatemeh Alizadehziri, Ali Khadem

Background

Functional neuroimaging has been used to diagnose Autism Spectrum Disorder (ASD), but no study has combined EEG with face-processing tasks. Given the social-cognitive challenges in ASD, such as difficulty with eye contact, such tasks seem to be promising. We propose a novel diagnostic method using EEG data from face-perception tasks and deep learning.

Proposed method

We used raw EEG signals recorded from children (ASD and typically developing) during upright and inverted face-display tasks. The data was used without preprocessing as a time-by-channel matrix. We designed three lightweight 1D-CNN models to learn temporal patterns across channels and classify ASD.

Results

Using upright face stimuli, InceptionNet and ResNet achieved accuracies of 78.65% and 76.87%, respectively, with a final diagnosis rate of 91.6% for both. With inverted stimuli, accuracies were lower at 72.35% (InceptionNet) and 70.68% (ResNet), with final diagnosis rates of 83.3%. Performance was consistently better for upright faces.

Comparison with existing methods

To the best of our knowledge, this is the first EEG-based study for ASD classification using a face-perception task. Most existing methods rely on resting-state EEG, which does not probe specific social-cognitive deficits. Our task-driven approach provides a novel, more targeted framework for detecting ASD-related cognitive differences.

Conclusions

The method shows promising diagnostic results. The superior performance with upright faces aligns with the eye-avoidance and face-inversion effect hypotheses, highlighting the importance of facial orientation. This work establishes a new, insightful pipeline for ASD detection using task-based EEG and deep learning.

功能神经成像已被用于诊断自闭症谱系障碍（ASD），但尚未有研究将脑电图与面部处理任务结合起来。考虑到自闭症谱系障碍患者的社会认知挑战，比如眼神交流的困难，这些任务似乎很有希望。我们提出了一种利用面部感知任务和深度学习的脑电图数据进行诊断的新方法。我们使用来自儿童（ASD和典型发育）在直立和倒立面部展示任务中记录的原始脑电图信号。这些数据没有经过预处理，作为一个时间通道矩阵使用。我们设计了三个轻量级的1D-CNN模型来学习跨通道的时间模式并对ASD进行分类。结果使用直立面部刺激时，InceptionNet和ResNet的准确率分别为78.65%和76.87%，两者的最终诊出率均为91.6%。使用反向刺激时，准确率较低，分别为72.35% （InceptionNet）和70.68% (ResNet)，最终诊出率为83.3%。脸部直立的人的表现一直更好。与现有方法相比，据我们所知，这是第一个基于脑电图的研究，使用面部感知任务进行ASD分类。大多数现有的方法依赖于静息状态脑电图，它不能探测特定的社会认知缺陷。我们的任务驱动方法为检测自闭症相关的认知差异提供了一种新颖的、更有针对性的框架。结论该方法具有较好的诊断效果。直立面部的优越表现与避眼效应和脸反转效应假设一致，突出了面部方向的重要性。这项工作为基于任务的脑电图和深度学习的ASD检测建立了一个新的、有见地的管道。

{"title":"Deep learning for diagnosing autism spectrum disorder using EEG with upright and inverted face processing tasks","authors":"Fatemeh Alizadehziri, Ali Khadem","doi":"10.1016/j.bspc.2026.109655","DOIUrl":"10.1016/j.bspc.2026.109655","url":null,"abstract":"<div><h3>Background</h3><div>Functional neuroimaging has been used to diagnose Autism Spectrum Disorder (ASD), but no study has combined EEG with face-processing tasks. Given the social-cognitive challenges in ASD, such as difficulty with eye contact, such tasks seem to be promising. We propose a novel diagnostic method using EEG data from face-perception tasks and deep learning.</div></div><div><h3>Proposed method</h3><div>We used raw EEG signals recorded from children (ASD and typically developing) during upright and inverted face-display tasks. The data was used without preprocessing as a time-by-channel matrix. We designed three lightweight 1D-CNN models to learn temporal patterns across channels and classify ASD.</div></div><div><h3>Results</h3><div>Using upright face stimuli, InceptionNet and ResNet achieved accuracies of 78.65% and 76.87%, respectively, with a final diagnosis rate of 91.6% for both. With inverted stimuli, accuracies were lower at 72.35% (InceptionNet) and 70.68% (ResNet), with final diagnosis rates of 83.3%. Performance was consistently better for upright faces.</div></div><div><h3>Comparison with existing methods</h3><div>To the best of our knowledge, this is the first EEG-based study for ASD classification using a face-perception task. Most existing methods rely on resting-state EEG, which does not probe specific social-cognitive deficits. Our task-driven approach provides a novel, more targeted framework for detecting ASD-related cognitive differences.</div></div><div><h3>Conclusions</h3><div>The method shows promising diagnostic results. The superior performance with upright faces aligns with the eye-avoidance and face-inversion effect hypotheses, highlighting the importance of facial orientation. This work establishes a new, insightful pipeline for ASD detection using task-based EEG and deep learning.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109655"},"PeriodicalIF":4.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174596","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

LGPS: A lightweight GAN-based approach for polyp segmentation in colonoscopy images LGPS：一种轻量级的基于gan的结肠镜图像息肉分割方法

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-01 Epub Date: 2026-02-03 DOI: 10.1016/j.bspc.2026.109777

Fiseha Berhanu Tesema , Alejandro Guerra-Manzanares , Tianxiang Cui , Qian Zhang , Moses M. Solomon , Xiangjian He

Accurate and real-time polyp segmentation is essential for early colorectal cancer detection, yet remains challenging due to small, low-contrast lesions and the computational demands of many deep learning models. We propose LGPS, a Lightweight GAN-based Polyp Segmentation framework that achieves high segmentation accuracy with minimal computational overhead. LGPS integrates a MobileNetV2-based generator enhanced with Residual Squeeze-and-Excitation (ReSE) blocks and a discriminator equipped with Convolutional Conditional Random Fields (ConvCRF) to refine spatial coherence and boundary precision. A hybrid loss function — combining Binary Cross-Entropy, Weighted IoU, and Dice Loss — improves class imbalance handling and sensitivity to small or blurry polyps. LGPS is an extremely compact model, requiring only 1.07 million parameters — over 17

\times

smaller than many recent transformer- and CNN-based SOTA architectures — while preserving high segmentation accuracy. In quantitative evaluation, LGPS achieves a Dice score of 0.7299 and an IoU of 0.7867 on the multi-center PolypGen dataset, the most challenging benchmark in polyp segmentation. On the widely used CVC-ClinicDB dataset, LGPS attains the highest IoU (0.9238) among lightweight and transformer-based approaches. The model also runs at 100.08 FPS on 256 × 256 inputs, demonstrating true real-time capability. Stratified evaluation and qualitative results further confirm its robustness on small and low-contrast lesions. These results highlight LGPS as a computationally efficient yet high-performing framework suitable for real-time clinical deployment. The code is available at https://github.com/Falmi/LGPS/.

准确和实时的息肉分割对于早期结直肠癌检测至关重要，但由于病灶小、对比度低，以及许多深度学习模型的计算需求，仍然具有挑战性。我们提出LGPS，一种基于gan的轻量级多边形分割框架，以最小的计算开销实现高分割精度。LGPS集成了一个基于mobilenetv2的生成器，增强了残余挤压和激励（ReSE）块，以及一个配备卷积条件随机场（ConvCRF）的鉴别器，以改善空间相干性和边界精度。混合损失函数-结合二元交叉熵，加权IoU和骰子损失-改善类不平衡处理和灵敏度小或模糊的息肉。LGPS是一个非常紧凑的模型，只需要107万个参数-比许多最近的变压器和基于cnn的SOTA架构小17倍以上-同时保持高分割精度。在定量评价中，LGPS在多中心polygen数据集上的Dice得分为0.7299，IoU为0.7867，这是息肉分割中最具挑战性的基准。在广泛使用的CVC-ClinicDB数据集上，LGPS在轻量级和基于变压器的方法中获得了最高的IoU（0.9238）。该模型还在256 × 256输入下以100.08 FPS运行，展示了真正的实时能力。分层评价和定性结果进一步证实了其对小病变和低对比病变的稳健性。这些结果突出了LGPS作为一种计算效率高且高性能的框架，适合于实时临床部署。代码可在https://github.com/Falmi/LGPS/上获得。

{"title":"LGPS: A lightweight GAN-based approach for polyp segmentation in colonoscopy images","authors":"Fiseha Berhanu Tesema , Alejandro Guerra-Manzanares , Tianxiang Cui , Qian Zhang , Moses M. Solomon , Xiangjian He","doi":"10.1016/j.bspc.2026.109777","DOIUrl":"10.1016/j.bspc.2026.109777","url":null,"abstract":"<div><div>Accurate and real-time polyp segmentation is essential for early colorectal cancer detection, yet remains challenging due to small, low-contrast lesions and the computational demands of many deep learning models. We propose LGPS, a Lightweight GAN-based Polyp Segmentation framework that achieves high segmentation accuracy with minimal computational overhead. LGPS integrates a MobileNetV2-based generator enhanced with Residual Squeeze-and-Excitation (ReSE) blocks and a discriminator equipped with Convolutional Conditional Random Fields (ConvCRF) to refine spatial coherence and boundary precision. A hybrid loss function — combining Binary Cross-Entropy, Weighted IoU, and Dice Loss — improves class imbalance handling and sensitivity to small or blurry polyps. LGPS is an extremely compact model, requiring only 1.07 million parameters — over 17<span><math><mo>×</mo></math></span> smaller than many recent transformer- and CNN-based SOTA architectures — while preserving high segmentation accuracy. In quantitative evaluation, LGPS achieves a Dice score of 0.7299 and an IoU of 0.7867 on the multi-center PolypGen dataset, the most challenging benchmark in polyp segmentation. On the widely used CVC-ClinicDB dataset, LGPS attains the highest IoU (0.9238) among lightweight and transformer-based approaches. The model also runs at 100.08 FPS on 256 × 256 inputs, demonstrating true real-time capability. Stratified evaluation and qualitative results further confirm its robustness on small and low-contrast lesions. These results highlight LGPS as a computationally efficient yet high-performing framework suitable for real-time clinical deployment. The code is available at <span><span>https://github.com/Falmi/LGPS/</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109777"},"PeriodicalIF":4.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174589","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

Robust source-free few-shot brain tumor segmentation via style perturbation and heatmap-guided consistency 基于风格扰动和热图引导一致性的鲁棒无源少射脑肿瘤分割

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-01 Epub Date: 2026-02-05 DOI: 10.1016/j.bspc.2026.109790

Li Liu , Khairunnisa Hasikin , Kaijian Xia , Khin Wee Lai

Cross-domain few-shot segmentation (CD-FSS) has shown great potential for alleviating annotation scarcity and mitigating domain discrepancies in medical image analysis. However, existing CD-FSS approaches typically require access to source-domain data, which conflicts with clinical privacy regulations and data-isolation constraints. To address this issue, we propose SHCNet, a source-free few-shot brain tumor segmentation framework for MRI. SHCNet consists of three key components: (i) a feature-level style perturbation module (EnhanceStyle) that improves robustness to domain-specific appearance variations; (ii) a heatmap-guided structure alignment mechanism (HGSA) for enforcing spatial saliency consistency between support and query features; and (iii) a semantic consistency alignment (SCA) module that enhances intra-class compactness and inter-class separability via foreground–background distance constraints and triplet loss. Extensive experiments show that SHCNet significantly outperforms the state-of-the-art source-free method ABCDFSS across diverse datasets. On BraTS 2020, SHCNet (ResNet-50) achieves mean DSCs of 76.22% and 80.36% under 1-shot and 5-shot settings, yielding + 12.54 pp and + 16.23 pp improvements. On BraTS 2021, the gains reach + 14.87 pp and + 16.65 pp, while on BraTS Africa, SHCNet obtains + 8.27 pp and + 10.93 pp. Moreover, SHCNet delivers the lowest HD95 (down to 6.33 mm), demonstrating strong boundary awareness and cross-domain robustness. These results verify that SHCNet provides an effective solution for source-free few-shot tumor segmentation under clinically realistic constraints.

在医学图像分析中，Cross-domain few-shot segmentation （CD-FSS）在缓解注释稀缺性和缓解域差异方面显示出巨大的潜力。然而，现有的CD-FSS方法通常需要访问源域数据，这与临床隐私法规和数据隔离约束相冲突。为了解决这个问题，我们提出了SHCNet，一个无源的MRI小片段脑肿瘤分割框架。SHCNet由三个关键组件组成：(i)一个特征级风格扰动模块（enhancyle），它提高了对特定领域外观变化的鲁棒性；（ii）以热图为导向的结构对齐机制（HGSA），以加强支持和查询特征之间的空间显著性一致性；（iii）语义一致性对齐（SCA）模块，通过前景-背景距离约束和三元组丢失增强类内紧凑性和类间可分离性。大量的实验表明，在不同的数据集上，SHCNet显著优于最先进的无源方法ABCDFSS。在BraTS 2020上，SHCNet （ResNet-50）在1针和5针设置下的平均dsc分别为76.22%和80.36%，分别提高了12.54和16.23个pp。在BraTS 2021上，增益达到+ 14.87 pp和+ 16.65 pp，而在BraTS Africa上，SHCNet获得+ 8.27 pp和+ 10.93 pp。此外，SHCNet提供最低的HD95（降至6.33 mm），表现出强大的边界感知和跨域鲁棒性。这些结果验证了SHCNet在临床现实约束下为无源少次肿瘤分割提供了有效的解决方案。

{"title":"Robust source-free few-shot brain tumor segmentation via style perturbation and heatmap-guided consistency","authors":"Li Liu , Khairunnisa Hasikin , Kaijian Xia , Khin Wee Lai","doi":"10.1016/j.bspc.2026.109790","DOIUrl":"10.1016/j.bspc.2026.109790","url":null,"abstract":"<div><div>Cross-domain few-shot segmentation (CD-FSS) has shown great potential for alleviating annotation scarcity and mitigating domain discrepancies in medical image analysis. However, existing CD-FSS approaches typically require access to source-domain data, which conflicts with clinical privacy regulations and data-isolation constraints. To address this issue, we propose SHCNet, a source-free few-shot brain tumor segmentation framework for MRI. SHCNet consists of three key components: (i) a feature-level style perturbation module (EnhanceStyle) that improves robustness to domain-specific appearance variations; (ii) a heatmap-guided structure alignment mechanism (HGSA) for enforcing spatial saliency consistency between support and query features; and (iii) a semantic consistency alignment (SCA) module that enhances intra-class compactness and inter-class separability via foreground–background distance constraints and triplet loss. Extensive experiments show that SHCNet significantly outperforms the state-of-the-art source-free method ABCDFSS across diverse datasets. On BraTS 2020, SHCNet (ResNet-50) achieves mean DSCs of 76.22% and 80.36% under 1-shot and 5-shot settings, yielding + 12.54 pp and + 16.23 pp improvements. On BraTS 2021, the gains reach + 14.87 pp and + 16.65 pp, while on BraTS Africa, SHCNet obtains + 8.27 pp and + 10.93 pp. Moreover, SHCNet delivers the lowest HD95 (down to 6.33 mm), demonstrating strong boundary awareness and cross-domain robustness. These results verify that SHCNet provides an effective solution for source-free few-shot tumor segmentation under clinically realistic constraints.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109790"},"PeriodicalIF":4.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174336","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

AFET: Adaptive Frequency-Enhanced Transformer for X-ray image compression 用于x射线图像压缩的自适应频率增强变压器

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-01 Epub Date: 2026-02-05 DOI: 10.1016/j.bspc.2026.109727

Tiansong Li , Qingsong Yang , Guofen Wang , Shaoguo Cui , Hongkui Wang , Li Yu

With the rapid development of modern medical imaging equipment, X-ray image resolution continues to increase, leading to an exponential growth in data volume. To alleviate the storage and transmission burden of massive X-ray image data, efficient compression has become a key requirement for contemporary medical information systems. An adaptive frequency-enhancement transformer (AFET) is proposed for X-ray image compression by leveraging adaptive enhanced window attention (AEWA) and a novel deep feedforward network (DFFN) to build a multi-frequency domain interaction mechanism. Firstly, the AEWA module is designed to dynamically enhances key frequency components through adaptive weighting across windows, effectively eliminating redundancy and preserving subtle details. Secondly, the DFFN module is introduced to capture spatial correlations between different frequency components, improving structural feature extraction. Experiments on the ChestX-ray8 and CheXpert datasets demonstrate that AFET outperforms state-of-the-art learning-based compressors and traditional codecs (such as JPEG and BPG) in terms of PSNR and MS-SSIM metrics, achieving a BD-Rate reduction of 19.51% on ChestX-ray8. Furthermore, the clinical feasibility of our AFET was verified in a downstream fine-grained classification task on compressed images, achieving superior AUC scores in chest disease classification, confirming the effectiveness of our AFET in clinical applications. Code: https://github.com/TiansongLi/AFET.

随着现代医学成像设备的快速发展，x射线图像分辨率不断提高，导致数据量呈指数级增长。为了减轻海量x射线图像数据的存储和传输负担，高效压缩已成为当代医疗信息系统的关键要求。提出了一种用于x射线图像压缩的自适应频率增强变压器（AFET），该变压器利用自适应增强窗口注意（AEWA）和一种新型的深度前馈网络（DFFN）来构建多频域交互机制。首先，AEWA模块通过自适应跨窗口加权来动态增强关键频率分量，有效消除冗余并保留细微细节。其次，引入DFFN模块捕获不同频率分量之间的空间相关性，提高结构特征提取的效率；在ChestX-ray8和CheXpert数据集上的实验表明，在PSNR和MS-SSIM指标方面，AFET优于最先进的基于学习的压缩器和传统的编码器（如JPEG和BPG），在ChestX-ray8上实现了19.51%的BD-Rate降低。此外，在压缩图像的下游细粒度分类任务中验证了我们的AFET的临床可行性，在胸部疾病分类中获得了优异的AUC评分，证实了我们的AFET在临床应用中的有效性。代码:https://github.com/TiansongLi/AFET。

{"title":"AFET: Adaptive Frequency-Enhanced Transformer for X-ray image compression","authors":"Tiansong Li , Qingsong Yang , Guofen Wang , Shaoguo Cui , Hongkui Wang , Li Yu","doi":"10.1016/j.bspc.2026.109727","DOIUrl":"10.1016/j.bspc.2026.109727","url":null,"abstract":"<div><div>With the rapid development of modern medical imaging equipment, X-ray image resolution continues to increase, leading to an exponential growth in data volume. To alleviate the storage and transmission burden of massive X-ray image data, efficient compression has become a key requirement for contemporary medical information systems. An adaptive frequency-enhancement transformer (AFET) is proposed for X-ray image compression by leveraging adaptive enhanced window attention (AEWA) and a novel deep feedforward network (DFFN) to build a multi-frequency domain interaction mechanism. Firstly, the AEWA module is designed to dynamically enhances key frequency components through adaptive weighting across windows, effectively eliminating redundancy and preserving subtle details. Secondly, the DFFN module is introduced to capture spatial correlations between different frequency components, improving structural feature extraction. Experiments on the ChestX-ray8 and CheXpert datasets demonstrate that AFET outperforms state-of-the-art learning-based compressors and traditional codecs (such as JPEG and BPG) in terms of PSNR and MS-SSIM metrics, achieving a BD-Rate reduction of 19.51% on ChestX-ray8. Furthermore, the clinical feasibility of our AFET was verified in a downstream fine-grained classification task on compressed images, achieving superior AUC scores in chest disease classification, confirming the effectiveness of our AFET in clinical applications. Code: <span><span>https://github.com/TiansongLi/AFET</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109727"},"PeriodicalIF":4.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174338","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

Box-guided class-contextual representation learning for self-visual lung nodule detection 盒引导类-上下文表征学习用于自视肺结节检测

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-01 Epub Date: 2026-02-04 DOI: 10.1016/j.bspc.2026.109731

Kefu Zhao, Lei Zhang, Le Yi, Xiuyuan Xu

Detecting lung nodules from computed tomography images has become the routine practice for early lung cancer screening. Conventional methods have made progress in discriminating the high similarity between nodules and non-nodules by extracting contextual information through attention-based or multi-scale mechanisms. However, these methods fail to guide models to focus on learning robust contextual representations in contextual regions as they lack explicit mechanisms to perceive these ambiguous regions, leading to a suboptimal discriminability for such similarity. Moreover, by mimicking clinicians’ diagnostic processes, providing visual evidence of learned regions can enhance the interpretability of detection tools and guide models to generate reliable decisions. In this paper, we propose class-contextual representation learning with ante-hoc visual interpretations to enhance lung nodule detection. It leverages a query-based mechanism via class-contextual vectors to explicitly perceive ambiguous contextual regions for guiding the learning of contextual representations. According to the manifold assumption, this query compresses the contextual information embedded in the feature manifold into a low-dimensional, class-contextual latent space, thereby filtering redundancy for learning discriminative representations. To ensure the efficacy of this query, we propose a box-guided instance-level method that enables class-contextual vectors align with discriminative representations of their respective classes. Our method exhibits a reliable intrinsic visualization effect, which enhances the transparency of the model’s decision-making process and provides learning guidance for further performance gains. Experiments demonstrate that this method achieves competition performance metrics of 93.38% and 65.58% on the LUNA16 and PN9 datasets, respectively.

从计算机断层图像中检测肺结节已成为早期肺癌筛查的常规做法。传统方法通过基于注意或多尺度的机制提取上下文信息，在判别结节与非结节的高度相似性方面取得了进展。然而，这些方法无法引导模型专注于学习上下文区域中的鲁棒上下文表示，因为它们缺乏明确的机制来感知这些模糊区域，导致这种相似性的次优可判别性。此外，通过模仿临床医生的诊断过程，提供学习区域的视觉证据可以增强检测工具的可解释性，并指导模型生成可靠的决策。在本文中，我们提出了类上下文表征学习与临时视觉解释，以提高肺结节检测。它利用基于查询的机制，通过类上下文向量显式地感知模糊的上下文区域，以指导上下文表示的学习。根据流形假设，该查询将嵌入在特征流形中的上下文信息压缩到一个低维的类上下文潜在空间中，从而过滤冗余以学习判别表示。为了确保这个查询的有效性，我们提出了一个盒引导的实例级方法，使类上下文向量与它们各自类的区别表示保持一致。我们的方法具有可靠的内在可视化效果，增强了模型决策过程的透明度，并为进一步的性能提升提供了学习指导。实验表明，该方法在LUNA16和PN9数据集上的竞争性能指标分别达到了93.38%和65.58%。

{"title":"Box-guided class-contextual representation learning for self-visual lung nodule detection","authors":"Kefu Zhao, Lei Zhang, Le Yi, Xiuyuan Xu","doi":"10.1016/j.bspc.2026.109731","DOIUrl":"10.1016/j.bspc.2026.109731","url":null,"abstract":"<div><div>Detecting lung nodules from computed tomography images has become the routine practice for early lung cancer screening. Conventional methods have made progress in discriminating the high similarity between nodules and non-nodules by extracting contextual information through attention-based or multi-scale mechanisms. However, these methods fail to guide models to focus on learning robust contextual representations in contextual regions as they lack explicit mechanisms to perceive these ambiguous regions, leading to a suboptimal discriminability for such similarity. Moreover, by mimicking clinicians’ diagnostic processes, providing visual evidence of learned regions can enhance the interpretability of detection tools and guide models to generate reliable decisions. In this paper, we propose class-contextual representation learning with ante-hoc visual interpretations to enhance lung nodule detection. It leverages a query-based mechanism via class-contextual vectors to explicitly perceive ambiguous contextual regions for guiding the learning of contextual representations. According to the manifold assumption, this query compresses the contextual information embedded in the feature manifold into a low-dimensional, class-contextual latent space, thereby filtering redundancy for learning discriminative representations. To ensure the efficacy of this query, we propose a box-guided instance-level method that enables class-contextual vectors align with discriminative representations of their respective classes. Our method exhibits a reliable intrinsic visualization effect, which enhances the transparency of the model’s decision-making process and provides learning guidance for further performance gains. Experiments demonstrate that this method achieves competition performance metrics of 93.38% and 65.58% on the LUNA16 and PN9 datasets, respectively.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109731"},"PeriodicalIF":4.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174417","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

From CNNs to diffusion models: a decade of advances in brain tumor and stroke lesion segmentation across BraTS, ATLAS, and ISLES benchmarks 从cnn到扩散模型：跨越BraTS、ATLAS和ISLES基准的脑肿瘤和中风病灶分割的十年进展

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-01 Epub Date: 2026-02-05 DOI: 10.1016/j.bspc.2026.109780

Manzoor Mohammad, Burra Vijaya Babu

Accurate segmentation of brain tumors and ischemic stroke lesions from magnetic resonance imaging (MRI) is a fundamental step in diagnosis, prognosis, and treatment planning. Over the past decade, segmentation research has advanced rapidly with the development of deep learning architectures, large annotated datasets, and diverse evaluation protocols. This survey provides a unified overview of segmentation techniques spanning both brain tumors and ischemic stroke lesions, offering a comprehensive perspective on their evolution, strengths, and limitations. We examine key benchmark datasets, including BraTS (2015–2023), ISLES (2015–2022), and ATLAS v2.0, and summarize their imaging modalities, annotations, and clinical contexts. Major model families such as U-Net variants, CNN-Transformer hybrids, transformer-only models, ensemble frameworks, and emerging diffusion-based approaches are systematically analysed with respect to design principles and reported performance across subregions and lesion types. The survey further compiles leaderboard results, state-of-the-art comparisons, and over 70 influential studies, highlighting region-wise trends and performance variability. Visual analyses, including histograms and box plots, offer additional insight into how segmentation accuracy differs across datasets and methods. We also review persistent challenges such as data scarcity, domain shift, boundary ambiguity, and clinical integration barriers, along with research trends including foundation models, semi- and self-supervised learning, and multimodal fusion. By consolidating datasets, methodologies, evaluation metrics, and future directions, this survey serves as a reference point for researchers and clinicians, outlining both the progress made and the opportunities that remain in developing robust and clinically relevant segmentation systems.

从磁共振成像（MRI）中准确分割脑肿瘤和缺血性脑卒中病变是诊断、预后和治疗计划的基本步骤。在过去的十年中，随着深度学习架构、大型注释数据集和各种评估协议的发展，分割研究取得了快速进展。这项调查提供了一个统一的概述分割技术跨越脑肿瘤和缺血性中风病变，提供了一个全面的观点，他们的发展，优势和局限性。我们研究了关键的基准数据集，包括BraTS（2015-2023）、ISLES（2015-2022）和ATLAS v2.0，并总结了它们的成像方式、注释和临床背景。主要的模型家族，如U-Net变体、CNN-Transformer混合模型、仅变压器模型、集成框架和新兴的基于扩散的方法，系统地分析了设计原则和跨子区域和病变类型的报告性能。该调查进一步汇总了排行榜结果、最先进的比较和70多项有影响力的研究，突出了地区趋势和绩效差异。可视化分析，包括直方图和箱形图，提供了更多的洞察如何分割精度不同的数据集和方法。我们还回顾了持续的挑战，如数据稀缺，领域转移，边界模糊，临床整合障碍，以及研究趋势，包括基础模型，半和自我监督学习，以及多模态融合。通过整合数据集、方法、评估指标和未来方向，该调查为研究人员和临床医生提供了参考点，概述了在开发稳健的临床相关分割系统方面取得的进展和仍然存在的机会。

{"title":"From CNNs to diffusion models: a decade of advances in brain tumor and stroke lesion segmentation across BraTS, ATLAS, and ISLES benchmarks","authors":"Manzoor Mohammad, Burra Vijaya Babu","doi":"10.1016/j.bspc.2026.109780","DOIUrl":"10.1016/j.bspc.2026.109780","url":null,"abstract":"<div><div>Accurate segmentation of brain tumors and ischemic stroke lesions from magnetic resonance imaging (MRI) is a fundamental step in diagnosis, prognosis, and treatment planning. Over the past decade, segmentation research has advanced rapidly with the development of deep learning architectures, large annotated datasets, and diverse evaluation protocols. This survey provides a unified overview of segmentation techniques spanning both brain tumors and ischemic stroke lesions, offering a comprehensive perspective on their evolution, strengths, and limitations. We examine key benchmark datasets, including BraTS (2015–2023), ISLES (2015–2022), and ATLAS v2.0, and summarize their imaging modalities, annotations, and clinical contexts. Major model families such as U-Net variants, CNN-Transformer hybrids, transformer-only models, ensemble frameworks, and emerging diffusion-based approaches are systematically analysed with respect to design principles and reported performance across subregions and lesion types. The survey further compiles leaderboard results, state-of-the-art comparisons, and over 70 influential studies, highlighting region-wise trends and performance variability. Visual analyses, including histograms and box plots, offer additional insight into how segmentation accuracy differs across datasets and methods. We also review persistent challenges such as data scarcity, domain shift, boundary ambiguity, and clinical integration barriers, along with research trends including foundation models, semi- and self-supervised learning, and multimodal fusion. By consolidating datasets, methodologies, evaluation metrics, and future directions, this survey serves as a reference point for researchers and clinicians, outlining both the progress made and the opportunities that remain in developing robust and clinically relevant segmentation systems.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109780"},"PeriodicalIF":4.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174484","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

Lung-SOS model: multi-stage lung tumour classification via secretary bird optimized features guided stacked ensemble learning model lung - sos模型：基于秘书鸟优化特征引导的多阶段肺部肿瘤分类堆叠集成学习模型

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-01 Epub Date: 2026-02-03 DOI: 10.1016/j.bspc.2026.109582

S. Vijaya Bhargavi , R. Praveena

Lung cancer (LC) is one of the most prevalent and deadliest cancers, accounting for a great number of cancer-related deaths. Traditional segmentation networks often fail to accurately delineate tumor boundaries in low-contrast or noisy computed tomography (CT) that lead to imprecise localization and classification. To overcome these issues, a novel Lung-SOS model has been proposed for LC detection from CT images utilizing Ensemble classification techniques. The enhanced images are then segmented using DeepLabv3 + with Shuffle Atrous Spatial Pyramid Pooling (SASPP) that effectively isolates LC regions by capturing multi-scale contextual information. The Convolutional-Attention Network (CoAtNet) used to extract the deep features by effectively leveraging the benefits of both convolution and attention mechanisms. The Secretary Bird Optimization Algorithm (SBOA) selects the most relevant features to create optimal feature subset for enhancing classification efficiency. Finally, the classification through a stacked ensemble classifier categorizing into three types such as Non-Small Cell Lung Cancer (NSCLC), Small Cell Lung Cancer (SCLC), and Lung Carcinoid Tumour (LCT). From the experimental results, the proposed Lung-SOS model yields an overall accuracy of 99.25 %, and the overall F1 Score of 96.93 % for efficient LC detection. The proposed DeepLabv3 + with SASPP increases the overall DI by 15.23 %, 14.09 %, 5.86 %, 3.49 %, 5.21 %, 4.69 % and 1.47 % better than U-Net, V-net, SegNet, and DeepLabv3+, Swin-UNet, TransUNet and SegFormer, respectively. The proposed Lung-SOS Model enhances the total accuracy range of 8.41 %, 4.47 %, 3.38 %, 2.26 %, 5.24 % and 0.06 % better than Gradient Boosting, CNN, GAME, SVM, ResNet50 and FusionLungNet respectively.

肺癌（LC）是最常见和最致命的癌症之一，占癌症相关死亡人数的很大一部分。传统的分割网络往往不能准确地描绘肿瘤边界在低对比度或噪声的计算机断层扫描（CT），导致不精确的定位和分类。为了克服这些问题，我们提出了一种新的Lung-SOS模型，利用集成分类技术从CT图像中检测LC。然后使用DeepLabv3 +与Shuffle Atrous空间金字塔池（SASPP）进行分割，该池通过捕获多尺度上下文信息有效地隔离LC区域。卷积注意网络（CoAtNet）通过有效地利用卷积和注意机制的优点来提取深度特征。秘书鸟优化算法（SBOA）选择最相关的特征来创建最优特征子集，以提高分类效率。最后，通过堆叠集成分类器将分类分为非小细胞肺癌（NSCLC）、小细胞肺癌（SCLC）和类肺癌（LCT）三种类型。实验结果表明，本文提出的Lung-SOS模型总体准确率为99.25%，总体F1分数为96.93%，能够有效检测LC。与U-Net、V-net、SegNet和DeepLabv3+、swwin - unet、TransUNet和SegFormer相比，采用SASPP的DeepLabv3+的DI分别提高了15.23%、14.09%、5.86%、3.49%、5.21%、4.69%和1.47%。该模型比Gradient Boosting、CNN、GAME、SVM、ResNet50和FusionLungNet分别提高了8.41%、4.47%、3.38%、2.26%、5.24%和0.06%的总准确率。

{"title":"Lung-SOS model: multi-stage lung tumour classification via secretary bird optimized features guided stacked ensemble learning model","authors":"S. Vijaya Bhargavi , R. Praveena","doi":"10.1016/j.bspc.2026.109582","DOIUrl":"10.1016/j.bspc.2026.109582","url":null,"abstract":"<div><div>Lung cancer (LC) is one of the most prevalent and deadliest cancers, accounting for a great number of cancer-related deaths. Traditional segmentation networks often fail to accurately delineate tumor boundaries in low-contrast or noisy computed tomography (CT) that lead to imprecise localization and classification. To overcome these issues, a novel Lung-SOS model has been proposed for LC detection from CT images utilizing Ensemble classification techniques. The enhanced images are then segmented using DeepLabv3 + with Shuffle Atrous Spatial Pyramid Pooling (SASPP) that effectively isolates LC regions by capturing multi-scale contextual information. The Convolutional-Attention Network (CoAtNet) used to extract the deep features by effectively leveraging the benefits of both convolution and attention mechanisms. The Secretary Bird Optimization Algorithm (SBOA) selects the most relevant features to create optimal feature subset for enhancing classification efficiency. Finally, the classification through a stacked ensemble classifier categorizing into three types such as Non-Small Cell Lung Cancer (NSCLC), Small Cell Lung Cancer (SCLC), and Lung Carcinoid Tumour (LCT). From the experimental results, the proposed Lung-SOS model yields an overall accuracy of 99.25 %, and the overall F1 Score of 96.93 % for efficient LC detection. The proposed DeepLabv3 + with SASPP increases the overall DI by 15.23 %, 14.09 %, 5.86 %, 3.49 %, 5.21 %, 4.69 % and 1.47 % better than U-Net, V-net, SegNet, and DeepLabv3+, Swin-UNet, TransUNet and SegFormer, respectively. The proposed Lung-SOS Model enhances the total accuracy range of 8.41 %, 4.47 %, 3.38 %, 2.26 %, 5.24 % and 0.06 % better than Gradient Boosting, CNN, GAME, SVM, ResNet50 and FusionLungNet respectively.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109582"},"PeriodicalIF":4.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174521","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

Pre-filtering ECG signals using the parabolic moving average to boost denoising performance 利用抛物线移动平均线对心电信号进行预滤波，提高去噪性能

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-01 Epub Date: 2026-02-02 DOI: 10.1016/j.bspc.2026.109788

F.A. Costabile, M.I. Gualtieri, A. Napoli

Electrocardiogram (ECG) signals are frequently corrupted by different types of noise, including baseline wander, muscle artifacts, and additive Gaussian noise, particularly relevant in wearable and ambulatory monitoring systems. In this paper, we propose a simple and effective pre-smoothing step to enhance the robustness of classical ECG denoising techniques. Specifically, we introduce a Parabolic Moving Average (PMA), that is, a symmetric three-point weighted average with weights

(1, 4, 1) / 6

, applied as a preliminary smoothing stage before standard filters such as Savitzky–Golay, Median, Butterworth and Second Order Smoothing. The PMA attenuates high-frequency fluctuations while preserving morphological features, with negligible computational cost. Unlike traditional moving averages, the PMA assigns nonuniform parabolic weights, yielding better smoothing with reduced distortion. We validate the approach on three noisy ECG signals, two synthetic and one real, measuring its effectiveness using standard metrics: Mean Squared Error, Signal-to-Noise Ratio, and Percentage Root-Mean-Square Difference. Results in all tested datasets demonstrate that the PMA improves the performance of all denoising filters considered without modifying their internal structure. These findings suggest that a simple pre-smoothing step can significantly improve traditional ECG denoising pipelines.

心电图（ECG）信号经常被不同类型的噪声所破坏，包括基线漂移、肌肉伪影和加性高斯噪声，这在可穿戴和动态监测系统中尤为重要。在本文中，我们提出了一种简单有效的预平滑步骤来增强经典心电去噪技术的鲁棒性。具体来说，我们引入了抛物移动平均（PMA），即权重为(1,4,1)/6的对称三点加权平均，作为标准滤波器（如Savitzky-Golay， Median， Butterworth和二阶平滑）之前的初步平滑阶段。PMA在保留形态学特征的同时衰减高频波动，计算成本可以忽略不计。与传统的移动平均线不同，PMA分配非均匀抛物线权重，产生更好的平滑和减少失真。我们在三个有噪声的心电信号（两个合成信号和一个真实信号）上验证了该方法，并使用标准指标：均方误差、信噪比和百分均方根差来衡量其有效性。所有测试数据集的结果表明，PMA在不改变其内部结构的情况下提高了所有考虑的去噪滤波器的性能。这些结果表明，一个简单的预平滑步骤可以显著改善传统的心电去噪管道。

{"title":"Pre-filtering ECG signals using the parabolic moving average to boost denoising performance","authors":"F.A. Costabile, M.I. Gualtieri, A. Napoli","doi":"10.1016/j.bspc.2026.109788","DOIUrl":"10.1016/j.bspc.2026.109788","url":null,"abstract":"<div><div>Electrocardiogram (ECG) signals are frequently corrupted by different types of noise, including baseline wander, muscle artifacts, and additive Gaussian noise, particularly relevant in wearable and ambulatory monitoring systems. In this paper, we propose a simple and effective pre-smoothing step to enhance the robustness of classical ECG denoising techniques. Specifically, we introduce a Parabolic Moving Average (PMA), that is, a symmetric three-point weighted average with weights <span><math><mrow><mrow><mo>(</mo><mn>1</mn><mo>,</mo><mn>4</mn><mo>,</mo><mn>1</mn><mo>)</mo></mrow><mo>/</mo><mn>6</mn></mrow></math></span>, applied as a preliminary smoothing stage before standard filters such as Savitzky–Golay, Median, Butterworth and Second Order Smoothing. The PMA attenuates high-frequency fluctuations while preserving morphological features, with negligible computational cost. Unlike traditional moving averages, the PMA assigns nonuniform parabolic weights, yielding better smoothing with reduced distortion. We validate the approach on three noisy ECG signals, two synthetic and one real, measuring its effectiveness using standard metrics: Mean Squared Error, Signal-to-Noise Ratio, and Percentage Root-Mean-Square Difference. Results in all tested datasets demonstrate that the PMA improves the performance of all denoising filters considered without modifying their internal structure. These findings suggest that a simple pre-smoothing step can significantly improve traditional ECG denoising pipelines.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109788"},"PeriodicalIF":4.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174525","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

TiMoS-Net: a multi-temporal image and morphological spatial feature fusion network for predicting neoadjuvant chemotherapy response in breast cancer TiMoS-Net：用于预测乳腺癌新辅助化疗反应的多时间图像和形态空间特征融合网络

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-01 Epub Date: 2026-02-01 DOI: 10.1016/j.bspc.2026.109665

Chen Jin , Siyi Chen , Haixin Shen , Wenjie Tang , Yongxin Chen , Jie Xiao , Yuan Guo , Qianwei Zhou

Artificial intelligence-based diagnostic systems are increasingly vital in clinical decision support. This study introduces TiMoS-Net, a novel multi-temporal image and spatial morphological feature fusion network designed for early prediction of neoadjuvant chemotherapy (NAC) response in breast cancer. TiMoS-Net integrates longitudinal MRI data with advanced spatial-morphological features, leveraging a novel a pre-training strategy combining self-supervision and label guidance, a temporal attention mechanism to capture dynamic tumor changes, and sophisticated topological and fractal morphological feature extraction, followed by a genetic algorithm-based feature fusion. Validated across multiple patient cohorts totaling 452 individuals, the model demonstrated superior predictive performance for pathological complete response (pCR), achieving areas under the curve (AUCs) of 0.898 and 0.922 on the external validation cohorts, significantly outperforming conventional single-timepoint approaches and models relying solely on temporal image data. Ablation studies confirmed the significant contributions of both the temporal attention mechanism and the advanced morphological features to the model’s efficacy. Furthermore, radiogenomic analysis linked the image-derived predictions to distinct biological pathways and immune cell infiltration patterns, and survival analysis indicated a significant association between the predicted pCR status and improved patient outcomes. TiMoS-Net offers a robust and interpretable approach for enhancing NAC response prediction, providing valuable insights for personalized treatment strategies in breast cancer.

基于人工智能的诊断系统在临床决策支持中越来越重要。本研究介绍了TiMoS-Net，一种新型的多时间图像和空间形态特征融合网络，旨在早期预测乳腺癌新辅助化疗（NAC）的反应。TiMoS-Net将纵向MRI数据与先进的空间形态特征相结合，利用一种结合自我监督和标签引导的新颖预训练策略，一种捕捉动态肿瘤变化的时间注意机制，以及复杂的拓扑和分形形态特征提取，然后是基于遗传算法的特征融合。在共452个患者队列中验证，该模型在病理完全缓解（pCR）方面表现出卓越的预测性能，在外部验证队列中实现了0.898和0.922的曲线下面积（auc），显著优于传统的单时间点方法和仅依赖时间图像数据的模型。消融研究证实了时间注意机制和高级形态学特征对模型有效性的重要贡献。此外，放射基因组学分析将图像衍生的预测与不同的生物学途径和免疫细胞浸润模式联系起来，生存分析表明预测的pCR状态与改善的患者预后之间存在显著关联。TiMoS-Net为增强NAC反应预测提供了一种可靠且可解释的方法，为乳腺癌的个性化治疗策略提供了有价值的见解。

{"title":"TiMoS-Net: a multi-temporal image and morphological spatial feature fusion network for predicting neoadjuvant chemotherapy response in breast cancer","authors":"Chen Jin , Siyi Chen , Haixin Shen , Wenjie Tang , Yongxin Chen , Jie Xiao , Yuan Guo , Qianwei Zhou","doi":"10.1016/j.bspc.2026.109665","DOIUrl":"10.1016/j.bspc.2026.109665","url":null,"abstract":"<div><div>Artificial intelligence-based diagnostic systems are increasingly vital in clinical decision support. This study introduces TiMoS-Net, a novel multi-temporal image and spatial morphological feature fusion network designed for early prediction of neoadjuvant chemotherapy (NAC) response in breast cancer. TiMoS-Net integrates longitudinal MRI data with advanced spatial-morphological features, leveraging a novel a pre-training strategy combining self-supervision and label guidance, a temporal attention mechanism to capture dynamic tumor changes, and sophisticated topological and fractal morphological feature extraction, followed by a genetic algorithm-based feature fusion. Validated across multiple patient cohorts totaling 452 individuals, the model demonstrated superior predictive performance for pathological complete response (pCR), achieving areas under the curve (AUCs) of 0.898 and 0.922 on the external validation cohorts, significantly outperforming conventional single-timepoint approaches and models relying solely on temporal image data. Ablation studies confirmed the significant contributions of both the temporal attention mechanism and the advanced morphological features to the model’s efficacy. Furthermore, radiogenomic analysis linked the image-derived predictions to distinct biological pathways and immune cell infiltration patterns, and survival analysis indicated a significant association between the predicted pCR status and improved patient outcomes. TiMoS-Net offers a robust and interpretable approach for enhancing NAC response prediction, providing valuable insights for personalized treatment strategies in breast cancer.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109665"},"PeriodicalIF":4.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174530","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

AMF-MedIT: An efficient align-modulation-fusion framework for medical image–tabular data AMF-MedIT：一个有效的医学图像表格数据对齐-调制-融合框架

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-01 Epub Date: 2026-02-04 DOI: 10.1016/j.bspc.2026.109772

Congjing Yu , Jing Ye , Yang Liu , Xiaodong Zhang , Zhiyong Zhang

Multimodal medical analysis combining image and tabular data has gained increasing attention. However, effective fusion remains challenging due to cross-modal discrepancies in feature dimensions and modality contributions, as well as the noise from high-dimensional tabular inputs. To address these problems, we present AMF-MedIT, an efficient Align-Modulation-Fusion framework for medical image and tabular data integration, particularly under data-scarce conditions. Built upon a self-supervised learning strategy, we introduce the Adaptive Modulation and Fusion (AMF) module, a novel, streamlined fusion paradigm that explicitly regulates the mismatch between modality confidence and representation dominance. Instead of relying on data-hungry learning, the AMF module incorporates modality confidence priors to guide adaptive contribution allocation, while harmonizing unimodal feature dimensionality and magnitude through feature modulation. To enhance tabular representation learning under noisy and limited data conditions, we adopt FT-Mamba as the tabular encoder within the proposed framework, leveraging its selective mechanism to extract discriminative features. Extensive experiments on both clean and clinically noisy datasets demonstrate that AMF-MedIT achieves superior accuracy, robustness, and data efficiency across multimodal classification tasks. Interpretability analyses further reveal how FT-Mamba shapes multimodal pretraining and enhances the image encoder’s attention, highlighting the practical value of our framework for reliable and efficient clinical artificial intelligence applications. The code is available at https://github.com/Jasmine-ycj/AMF-MedIT.git.

结合图像和表格数据的多模态医学分析越来越受到人们的关注。然而，由于特征维度和模态贡献的跨模态差异，以及来自高维表格输入的噪声，有效的融合仍然具有挑战性。为了解决这些问题，我们提出了AMF-MedIT，这是一个有效的对齐-调制-融合框架，用于医学图像和表格数据集成，特别是在数据稀缺的条件下。在自监督学习策略的基础上，我们引入了自适应调制和融合（AMF）模块，这是一种新颖的流线型融合范式，可以明确调节模态置信度和表征优势之间的不匹配。AMF模块不依赖于数据饥渴型学习，而是采用模态置信度先验来指导自适应贡献分配，同时通过特征调制协调单模态特征维度和大小。为了增强在噪声和有限数据条件下的表格表示学习，我们在提出的框架中采用FT-Mamba作为表格编码器，利用其选择机制提取判别特征。在干净和临床噪声数据集上进行的大量实验表明，AMF-MedIT在多模态分类任务中实现了卓越的准确性、鲁棒性和数据效率。可解释性分析进一步揭示了FT-Mamba如何塑造多模态预训练并增强图像编码器的注意力，突出了我们的框架在可靠和高效的临床人工智能应用中的实用价值。代码可在https://github.com/Jasmine-ycj/AMF-MedIT.git上获得。

{"title":"AMF-MedIT: An efficient align-modulation-fusion framework for medical image–tabular data","authors":"Congjing Yu , Jing Ye , Yang Liu , Xiaodong Zhang , Zhiyong Zhang","doi":"10.1016/j.bspc.2026.109772","DOIUrl":"10.1016/j.bspc.2026.109772","url":null,"abstract":"<div><div>Multimodal medical analysis combining image and tabular data has gained increasing attention. However, effective fusion remains challenging due to cross-modal discrepancies in feature dimensions and modality contributions, as well as the noise from high-dimensional tabular inputs. To address these problems, we present AMF-MedIT, an efficient Align-Modulation-Fusion framework for medical image and tabular data integration, particularly under data-scarce conditions. Built upon a self-supervised learning strategy, we introduce the Adaptive Modulation and Fusion (AMF) module, a novel, streamlined fusion paradigm that explicitly regulates the mismatch between modality confidence and representation dominance. Instead of relying on data-hungry learning, the AMF module incorporates modality confidence priors to guide adaptive contribution allocation, while harmonizing unimodal feature dimensionality and magnitude through feature modulation. To enhance tabular representation learning under noisy and limited data conditions, we adopt FT-Mamba as the tabular encoder within the proposed framework, leveraging its selective mechanism to extract discriminative features. Extensive experiments on both clean and clinically noisy datasets demonstrate that AMF-MedIT achieves superior accuracy, robustness, and data efficiency across multimodal classification tasks. Interpretability analyses further reveal how FT-Mamba shapes multimodal pretraining and enhances the image encoder’s attention, highlighting the practical value of our framework for reliable and efficient clinical artificial intelligence applications. The code is available at <span><span>https://github.com/Jasmine-ycj/AMF-MedIT.git</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109772"},"PeriodicalIF":4.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174644","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