Biomedical Signal Processing and Control最新文献_第8页

Research on flexible assisted control strategies of lower limb rehabilitation robot based on fuzzy admittance 基于模糊导纳的下肢康复机器人柔性辅助控制策略研究

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-02-05 DOI: 10.1016/j.bspc.2026.109782

Chao Gao , Chang Wang , Hui Li , Yongliang Cao , Shilong Wu , Jianhua Zhang

While human-robot interaction compliance has been advanced in studies on lower limb exoskeleton rehabilitation robots, a satisfactory solution to balancing the standardization of rehabilitation goals and the autonomy of user motion remains lacking. This paper proposes a long short-term memory neural network model based on an improved sparrow search algorithm (ISSA-LSTM) and a bilayer adaptive admittance framework based on fuzzy logic. Key electromyographic (EMG) signals of the lower limb muscles, human-robot interaction forces, and motion signals from the exoskeleton joint actuators are fused into an advanced controller integrated with a fuzzy adaptive admittance model, which is used to adjust the human-robot synergy performance during rehabilitation training. A low-order controller is designed using the backstepping method to ensure position control accuracy, and systematic uncertainties are compensated for through a state observer. The effectiveness of the proposed method was validated through trajectory tracking experiments and wearing experiments conducted on healthy participants. Results demonstrate that the proposed control method enables the rehabilitation robot to exhibit superior compliance and human-robot synergy effects, laying a solid foundation for its clinical application in stroke patients.

虽然在下肢外骨骼康复机器人的研究中，人机交互依从性已经取得了进展，但仍然缺乏一个令人满意的解决方案来平衡康复目标的标准化和用户运动的自主性。提出了一种基于改进的麻雀搜索算法（ISSA-LSTM）和基于模糊逻辑的双层自适应导纳框架的长短期记忆神经网络模型。将下肢肌肉的关键肌电信号、人机交互力和外骨骼关节执行器的运动信号融合到一个先进的控制器中，并结合模糊自适应导纳模型，用于调节康复训练过程中人机协同性能。采用反推法设计了低阶控制器，保证了系统的位置控制精度，并通过状态观测器对系统的不确定性进行了补偿。通过对健康受试者进行轨迹跟踪实验和佩戴实验，验证了该方法的有效性。结果表明，所提出的控制方法使康复机器人表现出优越的顺应性和人机协同效应，为其在脑卒中患者中的临床应用奠定了坚实的基础。

{"title":"Research on flexible assisted control strategies of lower limb rehabilitation robot based on fuzzy admittance","authors":"Chao Gao , Chang Wang , Hui Li , Yongliang Cao , Shilong Wu , Jianhua Zhang","doi":"10.1016/j.bspc.2026.109782","DOIUrl":"10.1016/j.bspc.2026.109782","url":null,"abstract":"<div><div>While human-robot interaction compliance has been advanced in studies on lower limb exoskeleton rehabilitation robots, a satisfactory solution to balancing the standardization of rehabilitation goals and the autonomy of user motion remains lacking. This paper proposes a long short-term memory neural network model based on an improved sparrow search algorithm (ISSA-LSTM) and a bilayer adaptive admittance framework based on fuzzy logic. Key electromyographic (EMG) signals of the lower limb muscles, human-robot interaction forces, and motion signals from the exoskeleton joint actuators are fused into an advanced controller integrated with a fuzzy adaptive admittance model, which is used to adjust the human-robot synergy performance during rehabilitation training. A low-order controller is designed using the backstepping method to ensure position control accuracy, and systematic uncertainties are compensated for through a state observer. The effectiveness of the proposed method was validated through trajectory tracking experiments and wearing experiments conducted on healthy participants. Results demonstrate that the proposed control method enables the rehabilitation robot to exhibit superior compliance and human-robot synergy effects, laying a solid foundation for its clinical application in stroke patients.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109782"},"PeriodicalIF":4.9,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174337","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

TAVC: Utilizing a variational clustering-based transformation-enhanced attention mechanism for spatial domain recognition 利用基于变分聚类的变换增强注意机制进行空间域识别

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-02-05 DOI: 10.1016/j.bspc.2026.109794

Yifan Dai, Xiangzhen Kong, Lingyun Dai, Shengjun Li, Juan Wang

Spatial transcriptomics (ST) technology enables the analysis of gene expression profiles, the exploration of cellular spatial distribution, and the decoding of tissue spatial architecture. However, effectively utilizing spatially transcribed datasets with similar gene expression and histological characteristics to more precisely identify spatial domains remains a significant challenge. The key lies in integrating gene expression data with its spatial context and histological imagery. This paper introduces a novel multimodal clustering framework: Transformer-Augmented Variational Clustering (TAVC). This approach achieves precise spatial domain identification by synergistically modeling histological image features, gene expression profiles, and spatial location information. Specifically, TAVC employs a pre-trained ResNet50 model to extract histological morphological features. It utilizes a symmetric encoder–decoder architecture combined with multi-head attention mechanisms to generate low-dimensional embeddings, enabling efficient spatial recognition. Within the Variational Autoencoder (VAE) framework, TAVC introduces a hybrid Transformer-GAT module. This module not only captures spatial adjacency relationships but also integrates graph structure priors with global attention mechanisms. By incorporating self-expression mechanisms, TAVC enhances representational self-consistency and stability, improves the model’s robustness to input perturbations, and participates as a regularization term in total loss optimization, effectively preventing overfitting. Experiments across multiple datasets demonstrate that TAVC outperforms existing methods in spatial domain recognition accuracy, clustering quality, and biological consistency, offering a novel technical pathway for the in-depth analysis of spatial transcriptomics data.

空间转录组学（ST）技术能够分析基因表达谱，探索细胞空间分布，以及解码组织空间结构。然而，如何有效利用具有相似基因表达和组织学特征的空间转录数据集来更精确地识别空间域仍然是一个重大挑战。关键在于整合基因表达数据及其空间背景和组织学图像。介绍了一种新的多模态聚类框架：变分聚类（TAVC）。该方法通过协同建模组织学图像特征、基因表达谱和空间位置信息来实现精确的空间域识别。具体而言，TAVC采用预训练的ResNet50模型提取组织学形态学特征。它利用对称编码器-解码器架构结合多头注意机制来生成低维嵌入，从而实现高效的空间识别。在变分自编码器（VAE）框架中，TAVC引入了混合变压器- gat模块。该模块不仅捕获空间邻接关系，而且将图结构先验与全局注意机制相结合。通过引入自表达机制，TAVC增强了表征自一致性和稳定性，提高了模型对输入扰动的鲁棒性，并作为正则化项参与总损失优化，有效防止过拟合。跨多个数据集的实验表明，TAVC在空间域识别精度、聚类质量和生物一致性方面优于现有方法，为空间转录组学数据的深入分析提供了新的技术途径。

{"title":"TAVC: Utilizing a variational clustering-based transformation-enhanced attention mechanism for spatial domain recognition","authors":"Yifan Dai, Xiangzhen Kong, Lingyun Dai, Shengjun Li, Juan Wang","doi":"10.1016/j.bspc.2026.109794","DOIUrl":"10.1016/j.bspc.2026.109794","url":null,"abstract":"<div><div>Spatial transcriptomics (ST) technology enables the analysis of gene expression profiles, the exploration of cellular spatial distribution, and the decoding of tissue spatial architecture. However, effectively utilizing spatially transcribed datasets with similar gene expression and histological characteristics to more precisely identify spatial domains remains a significant challenge. The key lies in integrating gene expression data with its spatial context and histological imagery. This paper introduces a novel multimodal clustering framework: Transformer-Augmented Variational Clustering (TAVC). This approach achieves precise spatial domain identification by synergistically modeling histological image features, gene expression profiles, and spatial location information. Specifically, TAVC employs a pre-trained ResNet50 model to extract histological morphological features. It utilizes a symmetric encoder–decoder architecture combined with multi-head attention mechanisms to generate low-dimensional embeddings, enabling efficient spatial recognition. Within the Variational Autoencoder (VAE) framework, TAVC introduces a hybrid Transformer-GAT module. This module not only captures spatial adjacency relationships but also integrates graph structure priors with global attention mechanisms. By incorporating self-expression mechanisms, TAVC enhances representational self-consistency and stability, improves the model’s robustness to input perturbations, and participates as a regularization term in total loss optimization, effectively preventing overfitting. Experiments across multiple datasets demonstrate that TAVC outperforms existing methods in spatial domain recognition accuracy, clustering quality, and biological consistency, offering a novel technical pathway for the in-depth analysis of spatial transcriptomics data.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109794"},"PeriodicalIF":4.9,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174478","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 multilabel ECG classification under class imbalance using augmentation and adaptive iterative training 基于增强和自适应迭代训练的类不平衡下稳健多标签心电分类

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-02-05 DOI: 10.1016/j.bspc.2026.109767

Joshua Tiffany, Qi Zhang

Deep learning has become an important approach for automated electrocardiogram (ECG) interpretation, with the potential to improve diagnostic consistency and reduce clinical workload. Despite these advances, multilabel ECG classification remains difficult in practice because publicly available datasets are highly imbalanced and limited in size. Clinically important arrhythmias often appear infrequently, which restricts effective learning and leads to reduced sensitivity for rare conditions. These challenges motivate the development of training strategies that can improve robustness under realistic data constraints. This study presents two complementary approaches designed to mitigate the effects of class imbalance. The first is a targeted data augmentation framework that increases the representation of minority classes using controlled, morphology-preserving signal transformations, including noise injection, amplitude scaling, filtering, and localized temporal warping. These transformations expand training diversity while maintaining diagnostically relevant waveform structure. The second approach introduces an adaptive iterative training scheme in which class weights and synthetic sample counts are updated at each iteration based on per-class validation F1 scores. This feedback mechanism directs additional training emphasis toward persistently underperforming classes across successive training stages. Experimental results show that targeted augmentation leads to consistent improvements across most diagnostic categories when compared with standard imbalance-handling baselines. Adaptive iterative training further improves recognition of rare and challenging arrhythmias, although with some trade-offs in overall discrimination. Together, these results demonstrate that carefully designed augmentation and adaptive training strategies can substantially improve multilabel ECG classification in data-limited and imbalanced settings, and they provide practical guidance for future work in medical time-series analysis.

深度学习已成为自动心电图（ECG）解释的重要方法，具有提高诊断一致性和减少临床工作量的潜力。尽管取得了这些进展，但由于公开可用的数据集高度不平衡且规模有限，因此多标签ECG分类在实践中仍然很困难。临床上重要的心律失常往往很少出现，这限制了有效的学习，并导致对罕见情况的敏感性降低。这些挑战激发了训练策略的发展，这些策略可以提高现实数据约束下的鲁棒性。本研究提出了两种互补的方法来减轻阶级失衡的影响。第一个是目标数据增强框架，它使用可控的、保持形态的信号变换（包括噪声注入、幅度缩放、滤波和局部时间扭曲）来增加少数类的表示。这些转换扩展了训练的多样性，同时保持了诊断相关的波形结构。第二种方法引入了一种自适应迭代训练方案，其中在每次迭代中基于每个类验证F1分数更新类权重和合成样本计数。这种反馈机制将额外的训练重点指向连续训练阶段中持续表现不佳的班级。实验结果表明，与标准的不平衡处理基线相比，有针对性的增强在大多数诊断类别中导致一致的改进。自适应迭代训练进一步提高了对罕见和具有挑战性的心律失常的识别，尽管在总体歧视方面存在一些权衡。总之，这些结果表明，精心设计的增强和自适应训练策略可以大大改善数据有限和不平衡设置下的多标签心电分类，并为未来的医疗时间序列分析工作提供实用指导。

{"title":"Robust multilabel ECG classification under class imbalance using augmentation and adaptive iterative training","authors":"Joshua Tiffany, Qi Zhang","doi":"10.1016/j.bspc.2026.109767","DOIUrl":"10.1016/j.bspc.2026.109767","url":null,"abstract":"<div><div>Deep learning has become an important approach for automated electrocardiogram (ECG) interpretation, with the potential to improve diagnostic consistency and reduce clinical workload. Despite these advances, multilabel ECG classification remains difficult in practice because publicly available datasets are highly imbalanced and limited in size. Clinically important arrhythmias often appear infrequently, which restricts effective learning and leads to reduced sensitivity for rare conditions. These challenges motivate the development of training strategies that can improve robustness under realistic data constraints. This study presents two complementary approaches designed to mitigate the effects of class imbalance. The first is a targeted data augmentation framework that increases the representation of minority classes using controlled, morphology-preserving signal transformations, including noise injection, amplitude scaling, filtering, and localized temporal warping. These transformations expand training diversity while maintaining diagnostically relevant waveform structure. The second approach introduces an adaptive iterative training scheme in which class weights and synthetic sample counts are updated at each iteration based on per-class validation F1 scores. This feedback mechanism directs additional training emphasis toward persistently underperforming classes across successive training stages. Experimental results show that targeted augmentation leads to consistent improvements across most diagnostic categories when compared with standard imbalance-handling baselines. Adaptive iterative training further improves recognition of rare and challenging arrhythmias, although with some trade-offs in overall discrimination. Together, these results demonstrate that carefully designed augmentation and adaptive training strategies can substantially improve multilabel ECG classification in data-limited and imbalanced settings, and they provide practical guidance for future work in medical time-series analysis.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109767"},"PeriodicalIF":4.9,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174403","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

DiverseCER-Net: A compound emotion recognition framework based on residual Anti-Aliasing Attention network and diverse data augmentation DiverseCER-Net：一种基于残差抗混叠注意网络和多样化数据增强的复合情绪识别框架

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-02-05 DOI: 10.1016/j.bspc.2026.109735

Baiyi Zhang , Feng Wang , Lingguang Hao , Fan Feng , Yufeng Guo

This study addresses the critical challenges in compound facial emotion recognition, including high-frequency feature aliasing, multi-scale semantic fragmentation, and feature confusion under long-tailed distributions. Existing methods, which typically rely on static single-scale feature modeling and fixed optimization strategies, often suffer from insufficient representation capability in complex emotional regions and reduced discriminative performance. To overcome these limitations, we propose a novel compound emotion recognition framework named DiverseCER-Net. This framework constructs a feature extraction backbone, ResA3EmoNet, and comprehensively reformulates the feature representation and optimization processes through three core technologies. First, a Dynamic Anti-Aliasing Attention (DAAA) mechanism is introduced to effectively preserve micro-expression details. Second, a Cross-Scale Semantic Collaborative Residual Feature Pyramid (CSCR-FP) is designed to facilitate explicit synergy between global contours and local micro-expression features, thereby solving the problem of semantic dilution during feature fusion. Finally, a Staged Weighted Triple-Order Joint Loss (SWT-Loss) is proposed, which adopts a progressive objective strategy of “Class Balance — Feature Separation — Distribution Compactness” to effectively mitigate the conflict between long-tailed data distributions and high inter-class similarity. Furthermore, the DiverseMix data augmentation strategy is incorporated to further enhance the model’s generalization ability on scarce samples. Extensive experiments conducted on the RAF-DB Compound and RAF-ML datasets demonstrate that DiverseCER-Net achieves state-of-the-art performance in multi-label compound emotion recognition tasks.

该研究解决了复合面部情感识别中的关键挑战，包括高频特征混叠、多尺度语义碎片和长尾分布下的特征混淆。现有的方法通常依赖于静态的单尺度特征建模和固定的优化策略，在复杂的情感区域中往往存在表征能力不足和判别性能下降的问题。为了克服这些限制，我们提出了一种新的复合情绪识别框架——DiverseCER-Net。该框架构建了特征提取骨干ResA3EmoNet，并通过三大核心技术全面重新制定了特征表示和优化流程。首先，引入动态抗混叠注意（DAAA）机制，有效保存微表情细节；其次，设计跨尺度语义协同残差特征金字塔（CSCR-FP），促进全局轮廓和局部微表达特征之间的显式协同，从而解决特征融合过程中的语义稀释问题。最后，提出了一种分阶段加权三阶联合损失算法（SWT-Loss），该算法采用“类平衡-特征分离-分布紧密”的渐进目标策略，有效缓解了长尾数据分布与高类间相似度之间的冲突。此外，还引入了DiverseMix数据增强策略，进一步增强了模型在稀缺样本上的泛化能力。在RAF-DB Compound和RAF-ML数据集上进行的大量实验表明，DiverseCER-Net在多标签复合情绪识别任务中达到了最先进的性能。

{"title":"DiverseCER-Net: A compound emotion recognition framework based on residual Anti-Aliasing Attention network and diverse data augmentation","authors":"Baiyi Zhang , Feng Wang , Lingguang Hao , Fan Feng , Yufeng Guo","doi":"10.1016/j.bspc.2026.109735","DOIUrl":"10.1016/j.bspc.2026.109735","url":null,"abstract":"<div><div>This study addresses the critical challenges in compound facial emotion recognition, including high-frequency feature aliasing, multi-scale semantic fragmentation, and feature confusion under long-tailed distributions. Existing methods, which typically rely on static single-scale feature modeling and fixed optimization strategies, often suffer from insufficient representation capability in complex emotional regions and reduced discriminative performance. To overcome these limitations, we propose a novel compound emotion recognition framework named DiverseCER-Net. This framework constructs a feature extraction backbone, ResA3EmoNet, and comprehensively reformulates the feature representation and optimization processes through three core technologies. First, a Dynamic Anti-Aliasing Attention (DAAA) mechanism is introduced to effectively preserve micro-expression details. Second, a Cross-Scale Semantic Collaborative Residual Feature Pyramid (CSCR-FP) is designed to facilitate explicit synergy between global contours and local micro-expression features, thereby solving the problem of semantic dilution during feature fusion. Finally, a Staged Weighted Triple-Order Joint Loss (SWT-Loss) is proposed, which adopts a progressive objective strategy of “Class Balance — Feature Separation — Distribution Compactness” to effectively mitigate the conflict between long-tailed data distributions and high inter-class similarity. Furthermore, the DiverseMix data augmentation strategy is incorporated to further enhance the model’s generalization ability on scarce samples. Extensive experiments conducted on the RAF-DB Compound and RAF-ML datasets demonstrate that DiverseCER-Net achieves state-of-the-art performance in multi-label compound emotion recognition tasks.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109735"},"PeriodicalIF":4.9,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174480","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

Progressive semi-supervised learning for multimodal pneumonia diagnosis 渐进式半监督学习在多模式肺炎诊断中的应用

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-02-05 DOI: 10.1016/j.bspc.2026.109750

Md Jobayer , Md. Mehedi Hasan Shawon , Mirza Rasheduzzaman , Mohammad Walidur Rahman , A.B.M.S.U. Doulah

To develop a multimodal machine learning framework for pneumonia detection from lung sound recordings to address the challenge of timely and affordable diagnosis in resource-limited settings. We designed a progressive semi-supervised learning model that processes lung sound signals in three forms: time-series data, spectrogram representations, and wavelet transforms. Contrastive and diversity losses were introduced during progressive training to improve generalization and reduce overfitting with limited labeled data. The proposed framework achieved state-of-the-art performance with an accuracy of 97.85% and an F1-score of 97.80%, outperforming existing unimodal and multimodal benchmarks. This approach shows strong potential as a reliable and efficient noninvasive screening tool for pneumonia, offering robust performance with a minimal computational footprint.

开发一个多模态机器学习框架，用于从肺录音中检测肺炎，以应对在资源有限的情况下及时和负担得起的诊断的挑战。我们设计了一个渐进式半监督学习模型，以三种形式处理肺声信号：时间序列数据、谱图表示和小波变换。在渐进式训练中引入了对比损失和多样性损失，以提高泛化并减少有限标记数据的过拟合。提出的框架实现了最先进的性能，准确率为97.85%，f1得分为97.80%，优于现有的单模态和多模态基准。这种方法显示出作为一种可靠和有效的非侵入性肺炎筛查工具的强大潜力，以最小的计算足迹提供强大的性能。

引用次数: 0

Automatic video-based monitoring of physically demanding tasks for biomechanical risk assessment 基于视频的生物力学风险评估任务的自动监控

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-02-05 DOI: 10.1016/j.bspc.2026.109810

F. Milano , I.M. Di Somma , L. Di Lorenzo , T. D’Orazio , M. Rizzi , C. Guaragnella

Musculoskeletal disorders (MSDs) represent a major concern in occupational health, impacting millions of workers globally. These conditions arise from repetitive movements, awkward postures, and excessive physical strain. As manual handling tasks continue to be widespread across various industries, a deeper understanding of the risk factors contributing to MSDs has become increasingly important. The use of systems for direct measurements of human movements that do not rely on wearable devices enables operators to carry out tasks in their regular work attire, making them convenient for real-world applications. This study explores the implementation of low-cost, video-based monitoring systems for the assessment of manual handling tasks, with the objective of detecting biomechanically suboptimal postures associated with an elevated risk of musculoskeletal disorders (MSDs). The validity of the measurements obtained using a standard RGB camera was verified through comparison with reference data provided by the Vicon gold standard system. The automatic analysis of postures — whether correct or incorrect — observed in a real experimental dataset was performed using unsupervised clustering techniques, following validation by domain experts. The models derived from this analysis were subsequently employed to classify new task executions and automatically detect postural anomalies. The proposed system demonstrated sufficient accuracy for the automatic detection of risk-related behaviors, highlighting its potential to enhance workplace safety.

肌肉骨骼疾病（MSDs）是职业健康中的一个主要问题，影响着全球数百万工人。这些情况是由重复的动作、笨拙的姿势和过度的身体紧张引起的。随着人工处理任务在各个行业的不断普及，对msd风险因素的深入了解变得越来越重要。使用系统直接测量人体运动，不依赖于可穿戴设备，使操作员能够在常规工作服中执行任务，使其便于实际应用。本研究探索了低成本、基于视频的监测系统的实施，用于评估手动处理任务，目的是检测与肌肉骨骼疾病（MSDs）风险升高相关的生物力学次优姿势。通过与Vicon金标准系统提供的参考数据进行比较，验证了使用标准RGB相机获得的测量结果的有效性。在经过领域专家的验证后，使用无监督聚类技术对真实实验数据集中观察到的姿势（无论正确与否）进行自动分析。从这一分析中得出的模型随后被用于分类新的任务执行和自动检测姿势异常。拟议的系统显示出足够的准确性，可以自动检测与风险有关的行为，突出了其提高工作场所安全的潜力。

{"title":"Automatic video-based monitoring of physically demanding tasks for biomechanical risk assessment","authors":"F. Milano , I.M. Di Somma , L. Di Lorenzo , T. D’Orazio , M. Rizzi , C. Guaragnella","doi":"10.1016/j.bspc.2026.109810","DOIUrl":"10.1016/j.bspc.2026.109810","url":null,"abstract":"<div><div>Musculoskeletal disorders (MSDs) represent a major concern in occupational health, impacting millions of workers globally. These conditions arise from repetitive movements, awkward postures, and excessive physical strain. As manual handling tasks continue to be widespread across various industries, a deeper understanding of the risk factors contributing to MSDs has become increasingly important. The use of systems for direct measurements of human movements that do not rely on wearable devices enables operators to carry out tasks in their regular work attire, making them convenient for real-world applications. This study explores the implementation of low-cost, video-based monitoring systems for the assessment of manual handling tasks, with the objective of detecting biomechanically suboptimal postures associated with an elevated risk of musculoskeletal disorders (MSDs). The validity of the measurements obtained using a standard RGB camera was verified through comparison with reference data provided by the Vicon gold standard system. The automatic analysis of postures — whether correct or incorrect — observed in a real experimental dataset was performed using unsupervised clustering techniques, following validation by domain experts. The models derived from this analysis were subsequently employed to classify new task executions and automatically detect postural anomalies. The proposed system demonstrated sufficient accuracy for the automatic detection of risk-related behaviors, highlighting its potential to enhance workplace safety.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109810"},"PeriodicalIF":4.9,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174529","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

Early recurrence prediction model for DLBCL based on Gaussian mixture model bidirectional clustering resampling and improved deep forest 基于高斯混合模型双向聚类重采样和改进深度森林的DLBCL早期递归预测模型

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-02-05 DOI: 10.1016/j.bspc.2026.109743

Yanhong Luo , Mengyuan Li , Yan Zhang , Tingting Zheng , Junxia Wang , Yanbo Zhang , Hongmei Yu , Hongyan Cao , Yongao Li , Nan Dai , Aiai Zhang , Tianyi Wang , Zhiqiang Zhao , Na Yang , Xiaoyan Li

Background

Patients with diffuse large B-cell lymphoma (DLBCL) are prone to relapse within two years of achieving complete remission with first-line treatment. The between-class and within-class imbalance that exists between relapsed and non-relapsed patients negatively affects the performance of traditional relapse prediction models.

Methods

A predictive model is constructed using the electronic medical record information of 544 DLBCL patients (2011–2021) who achieved complete remission in a tertiary care cancer hospital. Bidirectional clustering resampling based on a Gaussian mixture model is combined with nine classifiers in turn, and its performance is compared with that of eight other resampling methods. The optimal predictive model is determined by combining the model constructed using the optimal resampling method with the different classifiers.

Results

Bidirectional clustering resampling with the Gaussian mixture model outperforms other resampling techniques when applied to DLBCL data, and achieves optimal performance when combined with the improved deep forest classifier (AUC = 0.8549, MSE = 0.1468). This combination also demonstrates superior classification performance on three imbalanced datasets from KEEL.

Conclusion

Bidirectional clustering resampling using the Gaussian mixture model effectively addresses between-class and within-class imbalance issues, and its combination with the improved deep forest classifier significantly improves the predictive performance for early relapse in DLBCL patients.

弥漫性大b细胞淋巴瘤（DLBCL）患者在接受一线治疗后完全缓解后两年内容易复发。复发和非复发患者之间存在的类间和类内失衡对传统复发预测模型的性能产生了负面影响。方法利用某三级肿瘤医院2011-2021年544例完全缓解的DLBCL患者的电子病历信息构建预测模型。将基于高斯混合模型的双向聚类重采样与9种分类器相结合，并与其他8种重采样方法进行了性能比较。将最优重采样方法构建的模型与不同分类器相结合，确定最优预测模型。结果基于高斯混合模型的双向聚类重采样在DLBCL数据上优于其他重采样技术，并与改进的深度森林分类器结合使用时达到最佳性能（AUC = 0.8549, MSE = 0.1468）。这种组合在来自KEEL的三个不平衡数据集上也显示了优越的分类性能。结论基于高斯混合模型的双向聚类重采样有效地解决了类间和类内不平衡问题，并与改进的深度森林分类器相结合，显著提高了DLBCL患者早期复发的预测性能。

{"title":"Early recurrence prediction model for DLBCL based on Gaussian mixture model bidirectional clustering resampling and improved deep forest","authors":"Yanhong Luo , Mengyuan Li , Yan Zhang , Tingting Zheng , Junxia Wang , Yanbo Zhang , Hongmei Yu , Hongyan Cao , Yongao Li , Nan Dai , Aiai Zhang , Tianyi Wang , Zhiqiang Zhao , Na Yang , Xiaoyan Li","doi":"10.1016/j.bspc.2026.109743","DOIUrl":"10.1016/j.bspc.2026.109743","url":null,"abstract":"<div><h3>Background</h3><div>Patients with diffuse large B-cell lymphoma (DLBCL) are prone to relapse within two years of achieving complete remission with first-line treatment. The between-class and within-class imbalance that exists between relapsed and non-relapsed patients negatively affects the performance of traditional relapse prediction models.</div></div><div><h3>Methods</h3><div>A predictive model is constructed using the electronic medical record information of 544 DLBCL patients (2011–2021) who achieved complete remission in a tertiary care cancer hospital. Bidirectional clustering resampling based on a Gaussian mixture model is combined with nine classifiers in turn, and its performance is compared with that of eight other resampling methods. The optimal predictive model is determined by combining the model constructed using the optimal resampling method with the different classifiers.</div></div><div><h3>Results</h3><div>Bidirectional clustering resampling with the Gaussian mixture model outperforms other resampling techniques when applied to DLBCL data, and achieves optimal performance when combined with the improved deep forest classifier (AUC = 0.8549, MSE = 0.1468). This combination also demonstrates superior classification performance on three imbalanced datasets from KEEL.</div></div><div><h3>Conclusion</h3><div>Bidirectional clustering resampling using the Gaussian mixture model effectively addresses between-class and within-class imbalance issues, and its combination with the improved deep forest classifier significantly improves the predictive performance for early relapse in DLBCL patients.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109743"},"PeriodicalIF":4.9,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174479","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

Temporally coherent self-supervised apical trajectory learning for apical rocking classification in echocardiographic videos 超声心动图视频中用于心尖摇摆分类的时间相干自监督心尖轨迹学习

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-02-05 DOI: 10.1016/j.bspc.2026.109728

Mingshan Li , Fangyan Tian , Qin Wang , Zeju Li , Xianhong Shu , Yi Guo , Yuanyuan Wang

Apical Rocking (AR) is an echocardiography-observable marker of the interventricular dyssynchrony. In clinical routine, it serves as a critical indicator to determine whether the treatment via cardiac resynchronization therapy would be effective for heart failure. However, AR classification based on visual assessment by cardiologists requires clinical expertise and is subject to inter-observer variability, and we find that existing end-to-end machine learning methods are suboptimal on AR classification as they fail to model apical motion – a local but non-trivial key factor for AR. In this study, we propose ARNet for automatic AR classification by explicitly learning apical trajectories from echocardiographic videos. The architecture incorporates a branch that autonomously generates labels for self-supervised apical trajectory learning, eliminating dependency on costly manual annotations. In parallel, ARNet contains another branch which processes echocardiographic frame sequences to capture apical coordinate regression information, thereby enhancing temporal consistency in trajectory estimation. ARNet facilitates knowledge transfer between its dual branches and performs cross-fusion on their features to generate the final predictions. Extensive experiments demonstrate that ARNet achieves superior performance (area under the receiver operating characteristic curve (AUC) = 0.981) compared to state-of-the-art action recognition algorithms (AUC = 0.927) and even cardiologists (AUC = 0.918), highlighting its potential to enhance clinical decision-making and resource allocation in heart failure management.

心尖摇晃（AR）是超声心动图可观察到的室间非同步化的标志。在临床常规中，它是判断心脏再同步化治疗对心力衰竭是否有效的关键指标。然而，基于心脏病专家视觉评估的AR分类需要临床专业知识，并且受观察者之间的可变性的影响，我们发现现有的端到端机器学习方法在AR分类上是次优的，因为它们无法模拟根尖运动——一个局部但重要的AR关键因素。在本研究中，我们提出了通过明确学习超声心动图视频的根尖轨迹来实现自动AR分类的ARNet。该体系结构包含了一个分支，该分支可以自主生成自监督顶点轨迹学习的标签，从而消除了对昂贵的手动注释的依赖。同时，ARNet包含另一个分支，该分支处理超声心动图帧序列以捕获顶点坐标回归信息，从而增强轨迹估计的时间一致性。ARNet促进其双分支之间的知识转移，并对其特征进行交叉融合以生成最终预测。大量实验表明，与最先进的动作识别算法（AUC = 0.927）甚至心脏病专家（AUC = 0.918）相比，ARNet的性能（受试者工作特征曲线下面积(AUC) = 0.981）更优越，凸显了其在心力衰竭管理中增强临床决策和资源分配的潜力。

{"title":"Temporally coherent self-supervised apical trajectory learning for apical rocking classification in echocardiographic videos","authors":"Mingshan Li , Fangyan Tian , Qin Wang , Zeju Li , Xianhong Shu , Yi Guo , Yuanyuan Wang","doi":"10.1016/j.bspc.2026.109728","DOIUrl":"10.1016/j.bspc.2026.109728","url":null,"abstract":"<div><div>Apical Rocking (AR) is an echocardiography-observable marker of the interventricular dyssynchrony. In clinical routine, it serves as a critical indicator to determine whether the treatment via cardiac resynchronization therapy would be effective for heart failure. However, AR classification based on visual assessment by cardiologists requires clinical expertise and is subject to inter-observer variability, and we find that existing end-to-end machine learning methods are suboptimal on AR classification as they fail to model apical motion – a local but non-trivial key factor for AR. In this study, we propose ARNet for automatic AR classification by explicitly learning apical trajectories from echocardiographic videos. The architecture incorporates a branch that autonomously generates labels for self-supervised apical trajectory learning, eliminating dependency on costly manual annotations. In parallel, ARNet contains another branch which processes echocardiographic frame sequences to capture apical coordinate regression information, thereby enhancing temporal consistency in trajectory estimation. ARNet facilitates knowledge transfer between its dual branches and performs cross-fusion on their features to generate the final predictions. Extensive experiments demonstrate that ARNet achieves superior performance (area under the receiver operating characteristic curve (AUC) = 0.981) compared to state-of-the-art action recognition algorithms (AUC = 0.927) and even cardiologists (AUC = 0.918), highlighting its potential to enhance clinical decision-making and resource allocation in heart failure management.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109728"},"PeriodicalIF":4.9,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174404","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

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-02-05 DOI: 10.1016/j.bspc.2026.109769

Yue Han, Shaohua Chen, Fang Gao, Lijia Zhao

Group analysis of multi-subject fMRI data is vital for decoding brain functions and detecting brain disorders. Tucker decomposition, a multidimensional factorization technique, efficiently extracts shared spatiotemporal components and core tensors from fMRI datasets, making it well-suited for dynamic analyses. However, existing dynamic Tucker decomposition-based approaches overlook complex-valued information and underutilize the structural richness of core tensors. To address these gaps, we propose a sliding-window-based complex-valued Tucker decomposition framework with one-time-point overlap to capture fine-grained features. Our method incorporates spatial sparsity constraints, solved via the alternating direction method of multipliers combined with gradient descent approach, and resolves spatiotemporal ambiguity through core tensor uncompressing strategy. Using the extracted complex-valued spatial components for all the time windows, we propose to extract spatial voxel (SV) features and spatial functional network connectivity (sFNC) features. More importantly, we leverage the complex-valued core tensor to extract dynamic phase eigenvalue (PEV) features. Using resting-state fMRI data from schizophrenia patients (SZs) and healthy controls (HCs), the proposed method obtains SV features that captures group- and individual-level discrimination via one/two-sample t-tests on spatial components. By using PEV features along with sFNC features, our approach achieves 92.5% classification accuracy for SZ vs. HC, a 14.7% improvement over real-valued Tucker decomposition. This work highlights the utility of complex-valued Tucker decomposition methods in enhancing fMRI-based disease diagnosis by leveraging full tensor structural information and dynamic neural signatures.

多主体功能磁共振成像数据的分组分析对于解码脑功能和检测脑疾病至关重要。Tucker分解是一种多维因子分解技术，它能有效地从fMRI数据集中提取共享的时空分量和核心张量，使其非常适合于动态分析。然而，现有的基于动态Tucker分解的方法忽略了复杂值信息，并且没有充分利用核心张量的结构丰富性。为了解决这些差距，我们提出了一种基于滑动窗口的复值Tucker分解框架，该框架具有单时间点重叠，以捕获细粒度特征。该方法结合了空间稀疏性约束，通过乘法器交替方向法结合梯度下降法求解，并通过核心张量解压缩策略解决时空模糊问题。利用提取的所有时间窗的复值空间分量，我们提出提取空间体素（SV）特征和空间功能网络连通性（sFNC）特征。更重要的是，我们利用复值核心张量提取动态相位特征值（PEV）特征。该方法利用来自精神分裂症患者（SZs）和健康对照（hc）的静息状态fMRI数据，通过对空间分量的单/双样本t检验，获得捕捉群体和个体水平歧视的SV特征。通过使用PEV特征和sFNC特征，我们的方法对SZ和HC的分类准确率达到了92.5%，比实值Tucker分解提高了14.7%。这项工作强调了复杂值Tucker分解方法在利用全张量结构信息和动态神经特征增强基于fmri的疾病诊断中的效用。

{"title":"Phase-related spatial feature extraction via dynamic complex-valued tucker decomposition in multi-subject fMRI data","authors":"Yue Han, Shaohua Chen, Fang Gao, Lijia Zhao","doi":"10.1016/j.bspc.2026.109769","DOIUrl":"10.1016/j.bspc.2026.109769","url":null,"abstract":"<div><div>Group analysis of multi-subject fMRI data is vital for decoding brain functions and detecting brain disorders. Tucker decomposition, a multidimensional factorization technique, efficiently extracts shared spatiotemporal components and core tensors from fMRI datasets, making it well-suited for dynamic analyses. However, existing dynamic Tucker decomposition-based approaches overlook complex-valued information and underutilize the structural richness of core tensors. To address these gaps, we propose a sliding-window-based complex-valued Tucker decomposition framework with one-time-point overlap to capture fine-grained features. Our method incorporates spatial sparsity constraints, solved via the alternating direction method of multipliers combined with gradient descent approach, and resolves spatiotemporal ambiguity through core tensor uncompressing strategy. Using the extracted complex-valued spatial components for all the time windows, we propose to extract spatial voxel (SV) features and spatial functional network connectivity (sFNC) features. More importantly, we leverage the complex-valued core tensor to extract dynamic phase eigenvalue (PEV) features. Using resting-state fMRI data from schizophrenia patients (SZs) and healthy controls (HCs), the proposed method obtains SV features that captures group- and individual-level discrimination via one/two-sample t-tests on spatial components. By using PEV features along with sFNC features, our approach achieves 92.5% classification accuracy for SZ vs. HC, a 14.7% improvement over real-valued Tucker decomposition. This work highlights the utility of complex-valued Tucker decomposition methods in enhancing fMRI-based disease diagnosis by leveraging full tensor structural information and dynamic neural signatures.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109769"},"PeriodicalIF":4.9,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174405","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

An automated optical flow-mediated dilation method for fast screening of endothelial function 一种用于内皮功能快速筛选的自动光流介导扩张方法

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-02-05 DOI: 10.1016/j.bspc.2026.109785

Yi Qi , Ying Jie Chee , Congwen Miao , Songhua Zheng , Tristan Wen Jie Choo , Ruochong Zhang , Quan Wang , Michael Yu Qi Zhou , Malini Olivo , Rinkoo Dalan , Renzhe Bi

Flow-mediated dilation (FMD) is the non-invasive gold standard for assessing endothelial dysfunction, an early and reversible marker of atherosclerosis, yet its uptake is limited by the cost, complexity, and operator dependence of ultrasound. This study presents an optical method that quantifies endothelial dysfunction using a compact, low-cost, fully automated diffuse speckle pulsatile flowmetry (DSPF) device. The system offers a new Reactive-Hyperemia–Flow-Volume (RHFV) index that achieves a strong correlation (r = 0.87) with clinical ultrasound FMD index and high discriminative performance for endothelial dysfunction (AUC = 0.8475), demonstrating accuracy comparable to Doppler ultrasound. By enabling convenient, reliable evaluation of endothelial dysfunction at the point of care, this optical technology holds substantial promise as a primary-care screening tool for cardiovascular risk stratification and longitudinal monitoring, with the potential to improve prevention pathways and broaden access to vascular health assessment. Trial registration: NHG DSRB, 2022/00223. Registered 28 April 2022.

血流介导扩张（FMD）是评估内皮功能障碍的非侵入性金标准，是动脉粥样硬化的早期和可逆标志物，但其应用受到超声的成本、复杂性和操作者依赖性的限制。本研究提出了一种光学方法，使用紧凑，低成本，全自动漫射散斑脉动血流仪（DSPF）设备量化内皮功能障碍。该系统提供了一种新的反应-充血-流量-体积（RHFV）指数，该指数与临床超声FMD指数具有很强的相关性（r = 0.87），对内皮功能障碍具有很高的鉴别性能（AUC = 0.8475），其准确性可与多普勒超声媲美。通过在护理点方便、可靠地评估内皮功能障碍，这种光学技术作为心血管风险分层和纵向监测的初级保健筛查工具具有很大的前景，具有改善预防途径和拓宽血管健康评估途径的潜力。试验注册：NHG DSRB， 2022/00223。注册于2022年4月28日。

{"title":"An automated optical flow-mediated dilation method for fast screening of endothelial function","authors":"Yi Qi , Ying Jie Chee , Congwen Miao , Songhua Zheng , Tristan Wen Jie Choo , Ruochong Zhang , Quan Wang , Michael Yu Qi Zhou , Malini Olivo , Rinkoo Dalan , Renzhe Bi","doi":"10.1016/j.bspc.2026.109785","DOIUrl":"10.1016/j.bspc.2026.109785","url":null,"abstract":"<div><div>Flow-mediated dilation (FMD) is the non-invasive gold standard for assessing endothelial dysfunction, an early and reversible marker of atherosclerosis, yet its uptake is limited by the cost, complexity, and operator dependence of ultrasound. This study presents an optical method that quantifies endothelial dysfunction using a compact, low-cost, fully automated diffuse speckle pulsatile flowmetry (DSPF) device. The system offers a new Reactive-Hyperemia–Flow-Volume (RHFV) index that achieves a strong correlation (r = 0.87) with clinical ultrasound FMD index and high discriminative performance for endothelial dysfunction (AUC = 0.8475), demonstrating accuracy comparable to Doppler ultrasound. By enabling convenient, reliable evaluation of endothelial dysfunction at the point of care, this optical technology holds substantial promise as a primary-care screening tool for cardiovascular risk stratification and longitudinal monitoring, with the potential to improve prevention pathways and broaden access to vascular health assessment. Trial registration: NHG DSRB, 2022/00223. Registered 28 April 2022.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"118 ","pages":"Article 109785"},"PeriodicalIF":4.9,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174476","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