Biomedical Signal Processing and Control最新文献_第3页

DIMER: EEG-based DNA-inspired multi-dimension feature emotion quantization recognition method DIMER：基于脑电图的dna启发的多维特征情绪量化识别方法

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-15 Epub Date: 2026-02-13 DOI: 10.1016/j.bspc.2026.109830

Lingyue Hou , Xiaoxia Li , Donglei Liu , Yingyue Zhou

A novel DNA-Inspired Multi-dimensional Emotion Recognition Network (DIMER) is proposed to address critical challenges in affective computing. Firstly, addressing the issue of incomplete emotional state quantification in the field of emotion recognition, a 16-level emotion quantification methodology based on valence, arousal, dominance, and liking is proposed, enabling fine-grained classification of emotional states. Secondly, to overcome the complexity of EEG feature extraction and poor multi-dimensional feature fusion performance, a dynamic complementary fusion approach for multi-dimensional features based on DNA interaction mechanisms is introduced. This approach conceptualizes EEG differential entropy features and their visualized representations as DNA double-strand carriers, quantifying the complementary degree between features by simulating hydrogen bond strength through feature affinity calculations. Finally, main chain-complementary chain feature associations are established based on DNA base-pairing principles, thereby achieving adaptive feature fusion. Experimental results on the DEAP emotion dataset demonstrate that subject-dependent experiments achieve 95.07% accuracy, outperforming MLP and LSTM by 1.82% and 18.16%, respectively, while subject-independent experiments attain 94.82% accuracy, surpassing MLP and LSTM by 15.66% and 33.33%, respectively. The proposed method effectively advances the development of high-precision multi-dimensional emotion recognition.

提出了一种新的dna启发的多维情感识别网络（DIMER），以解决情感计算中的关键挑战。首先，针对情绪识别领域中情绪状态量化不完整的问题，提出了一种基于效价、唤醒、优势和喜欢的16级情绪量化方法，实现了情绪状态的细粒度分类。其次，针对脑电特征提取复杂、多维特征融合性能差的问题，提出了一种基于DNA相互作用机制的多维特征动态互补融合方法。该方法将脑电微分熵特征及其可视化表示概念化为DNA双链载体，通过特征亲和度计算模拟氢键强度，量化特征之间的互补程度。最后，基于DNA碱基配对原理建立主链-互补链特征关联，实现自适应特征融合。在DEAP情绪数据集上的实验结果表明，受试者依赖实验的准确率为95.07%，分别比MLP和LSTM高1.82%和18.16%；受试者独立实验的准确率为94.82%，分别比MLP和LSTM高15.66%和33.33%。该方法有效地推进了高精度多维情感识别的发展。

{"title":"DIMER: EEG-based DNA-inspired multi-dimension feature emotion quantization recognition method","authors":"Lingyue Hou , Xiaoxia Li , Donglei Liu , Yingyue Zhou","doi":"10.1016/j.bspc.2026.109830","DOIUrl":"10.1016/j.bspc.2026.109830","url":null,"abstract":"<div><div>A novel DNA-Inspired Multi-dimensional Emotion Recognition Network (DIMER) is proposed to address critical challenges in affective computing. Firstly, addressing the issue of incomplete emotional state quantification in the field of emotion recognition, a 16-level emotion quantification methodology based on valence, arousal, dominance, and liking is proposed, enabling fine-grained classification of emotional states. Secondly, to overcome the complexity of EEG feature extraction and poor multi-dimensional feature fusion performance, a dynamic complementary fusion approach for multi-dimensional features based on DNA interaction mechanisms is introduced. This approach conceptualizes EEG differential entropy features and their visualized representations as DNA double-strand carriers, quantifying the complementary degree between features by simulating hydrogen bond strength through feature affinity calculations. Finally, main chain-complementary chain feature associations are established based on DNA base-pairing principles, thereby achieving adaptive feature fusion. Experimental results on the DEAP emotion dataset demonstrate that subject-dependent experiments achieve 95.07% accuracy, outperforming MLP and LSTM by 1.82% and 18.16%, respectively, while subject-independent experiments attain 94.82% accuracy, surpassing MLP and LSTM by 15.66% and 33.33%, respectively. The proposed method effectively advances the development of high-precision multi-dimensional emotion recognition.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"119 ","pages":"Article 109830"},"PeriodicalIF":4.9,"publicationDate":"2026-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192612","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

CSA-Net: A lightweight channel split attention network with residual feature fusion for retinal vessel segmentation CSA-Net：基于残差特征融合的轻型通道分割注意网络

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-15 Epub Date: 2026-02-13 DOI: 10.1016/j.bspc.2026.109834

MinShan Jiang, Hongkai Liu, Shuai Huang, Jihui Mao, Yongfei Zhu, Xuedian Zhang

Automatic retinal vessel segmentation is vital for clinical assessment and therapeutic intervention. Extracting global and local features from fundus images remains a significant challenge for current methods. To address this, we propose a lightweight channel split attention network (CSA-Net), which integrates channel split attention and residual feature fusion, and can effectively capture global context information and fine-grained vascular details. In our model, we first suggest a channel split attention (CSA) module to facilitate multiscale feature aggregation and acquisition of global information. Then, we introduce a residual feature fusion (RFF) module to reduce information loss by incorporating residuals and enhancing feature maps during the multiscale fusion process. In addition, we setup a lightweight design using adaptive inverted residual encoders with varied kernel sizes to increase the computational efficiency. Five publicly available fundus datasets (DRIVE, CHASEDB1, STARE, HRF, LES-AV) were used to test our model. Experimental results demonstrate that CSA-Net achieves state-of-the-art performance, with ACC values up to 0.9830 and AUC values of 0.9948 with only 2.39 M parameters. Ablation studies validate the effectiveness of individual modules. The proposed CSA-Net achieves a good balance between segmentation accuracy and model complexity. In multiple retinal vascular segmentation benchmark tests, it achieves competitive or better performance with fewer parameters.

视网膜血管自动分割对临床评估和治疗干预至关重要。从眼底图像中提取全局和局部特征仍然是当前方法面临的重大挑战。为了解决这个问题，我们提出了一种轻量级的通道分裂注意网络（CSA-Net），该网络集成了通道分裂注意和残差特征融合，可以有效地捕获全局上下文信息和细粒度血管细节。在我们的模型中，我们首先提出了一个通道分裂注意（CSA）模块，以促进多尺度特征聚合和全局信息的获取。然后，我们引入残差特征融合（RFF）模块，在多尺度融合过程中通过残差融合和增强特征映射来减少信息丢失。此外，我们建立了一个轻量级的设计，使用自适应倒置残差编码器与不同的核大小，以提高计算效率。使用5个公开的眼底数据集（DRIVE、CHASEDB1、STARE、HRF、LES-AV）来测试我们的模型。实验结果表明，CSA-Net在2.39 M参数下，ACC值可达0.9830，AUC值可达0.9948，达到了最先进的性能。消融研究验证了单个模块的有效性。本文提出的CSA-Net在分割精度和模型复杂度之间取得了很好的平衡。在多次视网膜血管分割基准测试中，以较少的参数获得了具有竞争力或更好的性能。

{"title":"CSA-Net: A lightweight channel split attention network with residual feature fusion for retinal vessel segmentation","authors":"MinShan Jiang, Hongkai Liu, Shuai Huang, Jihui Mao, Yongfei Zhu, Xuedian Zhang","doi":"10.1016/j.bspc.2026.109834","DOIUrl":"10.1016/j.bspc.2026.109834","url":null,"abstract":"<div><div>Automatic retinal vessel segmentation is vital for clinical assessment and therapeutic intervention. Extracting global and local features from fundus images remains a significant challenge for current methods. To address this, we propose a lightweight channel split attention network (CSA-Net), which integrates channel split attention and residual feature fusion, and can effectively capture global context information and fine-grained vascular details. In our model, we first suggest a channel split attention (CSA) module to facilitate multiscale feature aggregation and acquisition of global information. Then, we introduce a residual feature fusion (RFF) module to reduce information loss by incorporating residuals and enhancing feature maps during the multiscale fusion process. In addition, we setup a lightweight design using adaptive inverted residual encoders with varied kernel sizes to increase the computational efficiency. Five publicly available fundus datasets (DRIVE, CHASEDB1, STARE, HRF, LES-AV) were used to test our model. Experimental results demonstrate that CSA-Net achieves state-of-the-art performance, with ACC values up to 0.9830 and AUC values of 0.9948 with only 2.39 M parameters. Ablation studies validate the effectiveness of individual modules. The proposed CSA-Net achieves a good balance between segmentation accuracy and model complexity. In multiple retinal vascular segmentation benchmark tests, it achieves competitive or better performance with fewer parameters.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"119 ","pages":"Article 109834"},"PeriodicalIF":4.9,"publicationDate":"2026-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192613","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

PEEG-HAR: A novel pain-evoked EEG extraction method guided by the adaptive localization of high-activation-rate pain-related EEG sources peg - har：一种基于高激活率疼痛相关脑电图源自适应定位的疼痛诱发脑电图提取新方法

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-15 Epub Date: 2026-02-12 DOI: 10.1016/j.bspc.2026.109725

Wenjia Gao, Dan Liu, Qisong Wang, Yongping Zhao, Jinwei Sun

Laser-evoked potentials (LEPs) are widely recognized as optimal for pain assessment, but raw EEG signals are often contaminated by noise and background activity, making LEPs extraction challenging. Existing wavelet-based methods construct templates using all trials. However, since LEPs exhibit variations across different pain intensities, the use of all trials may result in the loss of discriminative features essential for distinguishing between varying levels of pain intensity. In this study, EEG source signals are obtained using electrophysiological source imaging, and high-activation-rate pain-related EEG sources are identified based on their activation characteristics. A within-subject pain intensity model based on a support vector machine is then developed to link recorded EEG signals with historical EEG samples. The model guides the selection of suitable EEG segments to construct a pain-specific template, which is subsequently used to reconstruct the recorded EEG signals by exploiting the time–frequency distribution of pain-related wavelet coefficients, thereby facilitating more effective extraction of LEPs. Experiments on real EEG recordings confirm that the proposed method can extract signals that more closely reflect genuine pain-evoked EEG activity, thereby enhancing the representation of pain and significantly improving subsequent classification performance, with binary accuracy increasing from 59.61% to 81.13% and three-class accuracy from 39.22% to 64.23%. The method addresses the challenge of insufficient pain expression in raw signals and provides a data foundation for developing objective pain biomarkers.

激光诱发电位（LEPs）被广泛认为是疼痛评估的最佳方法，但原始脑电图信号经常受到噪声和背景活动的污染，使得LEPs的提取具有挑战性。现有的基于小波的方法使用所有试验来构建模板。然而，由于lep在不同的疼痛强度中表现出差异，因此使用所有试验可能会导致丧失区分不同疼痛强度水平所必需的判别特征。本研究采用电生理源成像技术获取脑电源信号，并根据其激活特征对高激活率疼痛相关脑电源进行识别。然后建立了基于支持向量机的受试者疼痛强度模型，将记录的脑电信号与历史脑电信号样本联系起来。该模型引导选择合适的脑电信号片段构建痛觉特异性模板，然后利用痛觉相关小波系数的时频分布对记录的脑电信号进行重构，从而更有效地提取lep。真实脑电记录实验证实，该方法能够提取更贴近真实疼痛诱发脑电活动的信号，从而增强疼痛表征，显著提高后续分类性能，二值准确率从59.61%提高到81.13%，三级准确率从39.22%提高到64.23%。该方法解决了原始信号中疼痛表达不足的挑战，为开发客观的疼痛生物标志物提供了数据基础。

{"title":"PEEG-HAR: A novel pain-evoked EEG extraction method guided by the adaptive localization of high-activation-rate pain-related EEG sources","authors":"Wenjia Gao, Dan Liu, Qisong Wang, Yongping Zhao, Jinwei Sun","doi":"10.1016/j.bspc.2026.109725","DOIUrl":"10.1016/j.bspc.2026.109725","url":null,"abstract":"<div><div>Laser-evoked potentials (LEPs) are widely recognized as optimal for pain assessment, but raw EEG signals are often contaminated by noise and background activity, making LEPs extraction challenging. Existing wavelet-based methods construct templates using all trials. However, since LEPs exhibit variations across different pain intensities, the use of all trials may result in the loss of discriminative features essential for distinguishing between varying levels of pain intensity. In this study, EEG source signals are obtained using electrophysiological source imaging, and high-activation-rate pain-related EEG sources are identified based on their activation characteristics. A within-subject pain intensity model based on a support vector machine is then developed to link recorded EEG signals with historical EEG samples. The model guides the selection of suitable EEG segments to construct a pain-specific template, which is subsequently used to reconstruct the recorded EEG signals by exploiting the time–frequency distribution of pain-related wavelet coefficients, thereby facilitating more effective extraction of LEPs. Experiments on real EEG recordings confirm that the proposed method can extract signals that more closely reflect genuine pain-evoked EEG activity, thereby enhancing the representation of pain and significantly improving subsequent classification performance, with binary accuracy increasing from 59.61% to 81.13% and three-class accuracy from 39.22% to 64.23%. The method addresses the challenge of insufficient pain expression in raw signals and provides a data foundation for developing objective pain biomarkers.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"119 ","pages":"Article 109725"},"PeriodicalIF":4.9,"publicationDate":"2026-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192618","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

Enhancing deep learning reliability in MRI images of Alzheimer’s disease using pairwise probability differential reliability quantification 利用两两概率差分可靠性量化增强阿尔茨海默病MRI图像的深度学习可靠性

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-15 Epub Date: 2026-02-12 DOI: 10.1016/j.bspc.2026.109717

Junaidul Islam , Isack Farady , Chia-Chen Kuo , Fu-Yu Lin , Chih-Yang Lin

In domains where safety is crucial, such as medical imaging, achieving a high degree of classification accuracy is inadequate unless it is complemented by predictions that are both reliable and interpretable. Traditional measures of uncertainty frequently yield global assessments that obscure localized fluctuations in model confidence, a factor that can be pivotal when differentiating between closely related clinical conditions. This paper presents Pairwise Probability Differential Reliability Quantification (PPDRQ), an innovative framework that assesses the reliability of predictions made by deep neural networks through the evaluation of pairwise discrepancies among class probability estimates. By incorporating PPDRQ into a custom loss function enhanced with a triplet loss to refine the feature space. The proposed approach incentivizes the network to generate outputs that are both more distinct and trustworthy. Comprehensive experiments conducted on MRI-based Alzheimer’s disease diagnosis indicate that models with PPDRQ exhibit enhanced reliability, improved interpretability, and superior performance, thereby offering significant insights for clinical decision-making.

在安全至关重要的领域，如医学成像，除非辅以既可靠又可解释的预测，否则实现高度的分类准确性是不够的。传统的不确定性测量方法经常产生的全球评估模糊了模型置信度的局部波动，而这一因素在区分密切相关的临床条件时可能是关键因素。本文提出了一种创新的框架——两两概率差分可靠性量化（PPDRQ），它通过评估类别概率估计之间的两两差异来评估深度神经网络预测的可靠性。通过将PPDRQ合并到自定义损失函数中，增强了三重损失，以改进特征空间。所提出的方法激励网络产生更独特和更值得信赖的输出。对基于mri的阿尔茨海默病诊断的综合实验表明，PPDRQ模型具有更高的可靠性、更好的可解释性和更好的性能，从而为临床决策提供了重要的见解。

{"title":"Enhancing deep learning reliability in MRI images of Alzheimer’s disease using pairwise probability differential reliability quantification","authors":"Junaidul Islam , Isack Farady , Chia-Chen Kuo , Fu-Yu Lin , Chih-Yang Lin","doi":"10.1016/j.bspc.2026.109717","DOIUrl":"10.1016/j.bspc.2026.109717","url":null,"abstract":"<div><div>In domains where safety is crucial, such as medical imaging, achieving a high degree of classification accuracy is inadequate unless it is complemented by predictions that are both reliable and interpretable. Traditional measures of uncertainty frequently yield global assessments that obscure localized fluctuations in model confidence, a factor that can be pivotal when differentiating between closely related clinical conditions. This paper presents Pairwise Probability Differential Reliability Quantification (PPDRQ), an innovative framework that assesses the reliability of predictions made by deep neural networks through the evaluation of pairwise discrepancies among class probability estimates. By incorporating PPDRQ into a custom loss function enhanced with a triplet loss to refine the feature space. The proposed approach incentivizes the network to generate outputs that are both more distinct and trustworthy. Comprehensive experiments conducted on MRI-based Alzheimer’s disease diagnosis indicate that models with PPDRQ exhibit enhanced reliability, improved interpretability, and superior performance, thereby offering significant insights for clinical decision-making.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"119 ","pages":"Article 109717"},"PeriodicalIF":4.9,"publicationDate":"2026-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192775","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

Design, modeling, and experimental evaluation of a fuzzy-controlled 2DOF knee exoskeleton for gait rehabilitation 用于步态康复的模糊控制二自由度膝关节外骨骼的设计、建模和实验评估

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-15 Epub Date: 2026-02-09 DOI: 10.1016/j.bspc.2026.109730

Mohammad Amin Iravani Rad , Ali Mokhtarian , Mohammad Taghi Karimi , Davood Toghraie

Wearable exoskeletons, equipped with intelligent control algorithms, have emerged as promising tools for restoring mobility in patients with neuromuscular impairments. This study presents the modeling, simulation, and experimental validation of a two-degree-of-freedom (2DOF) knee exoskeleton, specifically designed for adaptive gait rehabilitation. A dynamic link-segment model of the swing phase is developed using Lagrangian formulation, allowing for accurate estimation of knee torque (∼25 Nm) and power demands (>100 W), which guide actuator selection. A fuzzy logic controller (FLC) is designed and compared with a conventional PID controller to regulate knee joint motion. Unlike standard encoder-based setups, this system employs a potentiometer as an angular sensor, offering a low-cost yet effective solution for joint angle tracking. The exoskeleton also features a conical gearbox and mechanical joint limits (5°–55°) to ensure structural compactness and biomechanical safety. Experimental validation was conducted on six healthy subjects across three gait conditions (natural, PID-assisted, and FLC-assisted) using optical motion capture and kinematic analysis. Results show that the fuzzy controller achieved smoother transitions and reduced trajectory error (RMS error reduced by ∼40%) compared to PID control, closely approximating natural knee motion. These findings highlight the feasibility of signal-driven, adaptive control frameworks for future clinical applications in personalized neurorehabilitation.

配备智能控制算法的可穿戴外骨骼已经成为恢复神经肌肉损伤患者活动能力的有前途的工具。本研究介绍了专门为适应性步态康复设计的二自由度（2DOF）膝关节外骨骼的建模、仿真和实验验证。使用拉格朗日公式开发了摆动相位的动态连杆段模型，可以准确估计膝关节扭矩（~ 25 Nm）和功率需求（>100 W），从而指导执行器的选择。设计了一种模糊控制器（FLC），并与传统的PID控制器进行了比较。与标准的基于编码器的设置不同，该系统采用电位器作为角度传感器，为关节角度跟踪提供了低成本但有效的解决方案。外骨骼还具有锥形齿轮箱和机械关节极限（5°-55°），以确保结构紧凑性和生物力学安全性。通过光学运动捕捉和运动学分析，对六名健康受试者进行了三种步态状态（自然、pid辅助和flc辅助）的实验验证。结果表明，与PID控制相比，模糊控制器实现了更平滑的过渡，减少了轨迹误差（RMS误差减少了约40%），非常接近膝关节的自然运动。这些发现强调了信号驱动、自适应控制框架在个性化神经康复中未来临床应用的可行性。

{"title":"Design, modeling, and experimental evaluation of a fuzzy-controlled 2DOF knee exoskeleton for gait rehabilitation","authors":"Mohammad Amin Iravani Rad , Ali Mokhtarian , Mohammad Taghi Karimi , Davood Toghraie","doi":"10.1016/j.bspc.2026.109730","DOIUrl":"10.1016/j.bspc.2026.109730","url":null,"abstract":"<div><div>Wearable exoskeletons, equipped with intelligent control algorithms, have emerged as promising tools for restoring mobility in patients with neuromuscular impairments. This study presents the modeling, simulation, and experimental validation of a two-degree-of-freedom (2DOF) knee exoskeleton, specifically designed for adaptive gait rehabilitation. A dynamic link-segment model of the swing phase is developed using Lagrangian formulation, allowing for accurate estimation of knee torque (∼25 Nm) and power demands (>100 W), which guide actuator selection. A fuzzy logic controller (FLC) is designed and compared with a conventional PID controller to regulate knee joint motion. Unlike standard encoder-based setups, this system employs a potentiometer as an angular sensor, offering a low-cost yet effective solution for joint angle tracking. The exoskeleton also features a conical gearbox and mechanical joint limits (5°–55°) to ensure structural compactness and biomechanical safety. Experimental validation was conducted on six healthy subjects across three gait conditions (natural, PID-assisted, and FLC-assisted) using optical motion capture and kinematic analysis. Results show that the fuzzy controller achieved smoother transitions and reduced trajectory error (RMS error reduced by ∼40%) compared to PID control, closely approximating natural knee motion. These findings highlight the feasibility of signal-driven, adaptive control frameworks for future clinical applications in personalized neurorehabilitation.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"119 ","pages":"Article 109730"},"PeriodicalIF":4.9,"publicationDate":"2026-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192860","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

Triplet-branch diffusion model with conditional guidance and boundary enhancement for cervical nucleus segmentation 有条件引导和边界增强的三分支扩散模型用于颈核分割

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-15 Epub Date: 2026-02-10 DOI: 10.1016/j.bspc.2026.109753

Tianyu Shi , Jia Tang , Yantao Sun , Zhimin Liu

As the fourth most common cancer among women worldwide, cervical cancer diagnosis relies heavily on accurate and automated segmentation of cell nuclei in pathological images for early screening. To address challenges such as blurred boundaries, overlapping cells, and complex background interference in this task, we propose a novel triplet-branch diffusion model which is constructed in three key stages: First, a diffusion backbone network is developed to progressively reconstruct the target structures from noisy masks via a denoising process, integrating a frequency-domain attention mechanism to suppress high-frequency noise. Second, a semantic condition branch based on the U-Net architecture is designed to extract multi-scale image features, which injects anatomical priors into the diffusion backbone through cross-layer connections. Third, an edge guided branch is introduced, which employs a Boundary Attention Module to fuse explicit edge features extracted by the Canny operator into the diffusion backbone, enabling multi-level boundary guidance during the decoding phase. We validate the proposed model on two public datasets and one internal private dataset, achieving Dice coefficients of 94.36%, 95.04%, and 93.16%, respectively—representing improvements of 1.2% to 2.1% over state-of-the-art models in the field. Ablation studies on the proposed modules and loss functions, as well as visual analyses of the reverse diffusion process, further demonstrate the effectiveness of our approach. This method effectively reduces boundary errors in the segmentation of cervical cell nuclei while maintaining high interpretability. It provides potential intelligent diagnostic support for large-scale early screening of cervical cancer. However, further validation of its reliability on multi-center or clinical datasets is necessary.

作为全球第四大女性常见癌症，宫颈癌的诊断在很大程度上依赖于病理图像中细胞核的准确和自动分割，以进行早期筛查。为了解决该任务中存在的边界模糊、细胞重叠和背景干扰复杂等问题，我们提出了一种新的三分支扩散模型，该模型分为三个关键阶段：首先，建立扩散骨干网络，通过去噪过程从噪声掩模中逐步重建目标结构，并集成频域注意机制来抑制高频噪声；其次，设计基于U-Net架构的语义条件分支提取多尺度图像特征，通过跨层连接将解剖先验注入扩散主干；第三，引入边缘引导分支，利用边界注意模块将Canny算子提取的显式边缘特征融合到扩散主干中，实现解码阶段的多级边界引导。我们在两个公共数据集和一个内部私有数据集上验证了所提出的模型，分别获得了94.36%、95.04%和93.16%的Dice系数，比该领域最先进的模型提高了1.2%到2.1%。对所提出的模块和损失函数的烧蚀研究，以及反向扩散过程的可视化分析，进一步证明了我们方法的有效性。该方法在保持较高的可解释性的同时，有效地减少了宫颈细胞核分割的边界误差。它为大规模宫颈癌早期筛查提供了潜在的智能诊断支持。然而，在多中心或临床数据集上进一步验证其可靠性是必要的。

{"title":"Triplet-branch diffusion model with conditional guidance and boundary enhancement for cervical nucleus segmentation","authors":"Tianyu Shi , Jia Tang , Yantao Sun , Zhimin Liu","doi":"10.1016/j.bspc.2026.109753","DOIUrl":"10.1016/j.bspc.2026.109753","url":null,"abstract":"<div><div>As the fourth most common cancer among women worldwide, cervical cancer diagnosis relies heavily on accurate and automated segmentation of cell nuclei in pathological images for early screening. To address challenges such as blurred boundaries, overlapping cells, and complex background interference in this task, we propose a novel triplet-branch diffusion model which is constructed in three key stages: First, a diffusion backbone network is developed to progressively reconstruct the target structures from noisy masks via a denoising process, integrating a frequency-domain attention mechanism to suppress high-frequency noise. Second, a semantic condition branch based on the U-Net architecture is designed to extract multi-scale image features, which injects anatomical priors into the diffusion backbone through cross-layer connections. Third, an edge guided branch is introduced, which employs a Boundary Attention Module to fuse explicit edge features extracted by the Canny operator into the diffusion backbone, enabling multi-level boundary guidance during the decoding phase. We validate the proposed model on two public datasets and one internal private dataset, achieving Dice coefficients of 94.36%, 95.04%, and 93.16%, respectively—representing improvements of 1.2% to 2.1% over state-of-the-art models in the field. Ablation studies on the proposed modules and loss functions, as well as visual analyses of the reverse diffusion process, further demonstrate the effectiveness of our approach. This method effectively reduces boundary errors in the segmentation of cervical cell nuclei while maintaining high interpretability. It provides potential intelligent diagnostic support for large-scale early screening of cervical cancer. However, further validation of its reliability on multi-center or clinical datasets is necessary.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"119 ","pages":"Article 109753"},"PeriodicalIF":4.9,"publicationDate":"2026-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191958","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

Normalized mutual information centrality-based BCI channels selecting enhanced by refined multiple frequency bands 基于归一化互信息中心性的BCI信道选择，改进了多个频段

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-15 Epub Date: 2026-02-09 DOI: 10.1016/j.bspc.2026.109734

Yu Wang , Guorui Li , Xin Zhang , Shengpu Xu , Bo Yao , Jiangbo Pu

Brain-computer interface (BCI) applications are significantly influenced by electroencephalography (EEG) signal acquisition. The requirement for numerous channels not only increases the preparation time but also introduces redundant information, which negatively impacts BCI performance. To address this issue, a multi-frequency channel selection method, termed Multi-Frequency-normalized Mutual Information and Betweenness Centrality-based Channel Selection (MF-MIBCCS), was proposed. This approach independently selected optimal channels across multiple frequency bands to effectively mitigate cross-frequency interference. To evaluate the proposed method, studies were conducted on two datasets (BCI Competition III-IIIa and a self-collected dataset). The results demonstrated that the MF-MIBCCS outperformed conventional all-channel usage, achieving superior classification performance with fewer than 20 channels on average. The MF-MIBCCS method shows significant potential for reducing the number of required EEG channels, thereby facilitating more efficient and personalized BCI system design.

脑机接口（BCI）的应用受到脑电图信号采集的显著影响。对众多通道的需求不仅增加了准备时间，而且引入了冗余信息，对BCI性能产生了负面影响。为了解决这一问题，提出了一种基于多频率归一化互信息和中间度中心性的多频率信道选择方法（MF-MIBCCS）。该方法在多个频带中独立选择最优信道，有效地减轻了交叉频率干扰。为了评估所提出的方法，研究人员在两个数据集（BCI Competition III-IIIa和一个自收集数据集）上进行了研究。结果表明，MF-MIBCCS优于传统的全通道使用，在平均少于20个通道的情况下实现了优越的分类性能。MF-MIBCCS方法在减少所需脑电通道数量方面显示出巨大的潜力，从而促进更高效和个性化的脑机接口系统设计。

{"title":"Normalized mutual information centrality-based BCI channels selecting enhanced by refined multiple frequency bands","authors":"Yu Wang , Guorui Li , Xin Zhang , Shengpu Xu , Bo Yao , Jiangbo Pu","doi":"10.1016/j.bspc.2026.109734","DOIUrl":"10.1016/j.bspc.2026.109734","url":null,"abstract":"<div><div>Brain-computer interface (BCI) applications are significantly influenced by electroencephalography (EEG) signal acquisition. The requirement for numerous channels not only increases the preparation time but also introduces redundant information, which negatively impacts BCI performance. To address this issue, a multi-frequency channel selection method, termed Multi-Frequency-normalized Mutual Information and Betweenness Centrality-based Channel Selection (MF-MIBCCS), was proposed. This approach independently selected optimal channels across multiple frequency bands to effectively mitigate cross-frequency interference. To evaluate the proposed method, studies were conducted on two datasets (BCI Competition III-IIIa and a self-collected dataset). The results demonstrated that the MF-MIBCCS outperformed conventional all-channel usage, achieving superior classification performance with fewer than 20 channels on average. The MF-MIBCCS method shows significant potential for reducing the number of required EEG channels, thereby facilitating more efficient and personalized BCI system design.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"119 ","pages":"Article 109734"},"PeriodicalIF":4.9,"publicationDate":"2026-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191957","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

Time frequency transform kernel enhanced ShallowConvNet for auditory selective attention decoding with steady state motion auditory evoked potential 基于时频变换核增强的浅卷积神经网络的稳态运动听觉诱发电位听觉选择性注意解码

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-15 Epub Date: 2026-02-06 DOI: 10.1016/j.bspc.2026.109736

Huanqing Zhang , Jun Xie , Kaixuan Liu , Yan Liu , Wenxiang Dong , Guanghua Xu

Steady state motion auditory evoked potential (SSMAEP) is neural responses elicited by rhythmic auditory stimuli with periodic spatial motion. SSMAEP brain computer interface (BCI) relies on auditory selective attention to decode user intent in multi-source environments. However, the complex temporal and spectral structure of SSMAEP presents challenges for effective feature extraction from electroencephalogram (EEG). Time-frequency transforms are suited for extracting the joint time–frequency features of SSMAEP. Notably, these transforms share structural similarities with convolution operations in convolution neural networks. In this study, we propose a novel time frequency convolutional layer that incorporates structured kernels based on the S transform, continuous wavelet transform (CWT), and short-time Fourier transform (STFT). These time frequency kernels are embedded as learnable filters and replace the conventional first convolutional layer of ShallowConvNet. This design enables the model to more effectively capture SSMAEP signal dynamics across both time and frequency domains. The proposed method was evaluated on two SSMAEP-BCI datasets with two and three auditory targets. Experimental results demonstrate consistent improvements in classification accuracy and robustness compared to baseline models. Furthermore, analysis of the learned kernels revealed that the time frequency filters retained their interpretable structure after training, with task-relevant shifts in center frequency and bandwidth. These findings highlight not only the performance advantage but also the improved interpretability of the proposed model, offering insights into the spectral encoding of SSMAEP-BCI.

稳态运动听觉诱发电位（SSMAEP）是由周期性空间运动的节奏性听觉刺激引起的神经反应。SSMAEP脑机接口（BCI）在多源环境下依靠听觉选择性注意解码用户意图。然而，SSMAEP复杂的时间和频谱结构给有效提取脑电图特征带来了挑战。时频变换适合提取SSMAEP的联合时频特征。值得注意的是，这些变换与卷积神经网络中的卷积操作具有结构相似性。在这项研究中，我们提出了一种新的时频卷积层，该层结合了基于S变换、连续小波变换（CWT）和短时傅立叶变换（STFT）的结构化核。这些时频核被嵌入为可学习滤波器，并取代了传统的ShallowConvNet的第一卷积层。这种设计使模型能够更有效地捕获跨时域和频域的SSMAEP信号动态。在两个具有两个和三个听觉目标的SSMAEP-BCI数据集上对所提出的方法进行了评估。实验结果表明，与基线模型相比，分类精度和鲁棒性得到了一致的提高。此外，对学习到的核的分析表明，训练后的时频滤波器保留了其可解释的结构，中心频率和带宽发生了与任务相关的偏移。这些发现不仅突出了性能优势，而且还提高了所提出模型的可解释性，为SSMAEP-BCI的频谱编码提供了见解。

{"title":"Time frequency transform kernel enhanced ShallowConvNet for auditory selective attention decoding with steady state motion auditory evoked potential","authors":"Huanqing Zhang , Jun Xie , Kaixuan Liu , Yan Liu , Wenxiang Dong , Guanghua Xu","doi":"10.1016/j.bspc.2026.109736","DOIUrl":"10.1016/j.bspc.2026.109736","url":null,"abstract":"<div><div>Steady state motion auditory evoked potential (SSMAEP) is neural responses elicited by rhythmic auditory stimuli with periodic spatial motion. SSMAEP brain computer interface (BCI) relies on auditory selective attention to decode user intent in multi-source environments. However, the complex temporal and spectral structure of SSMAEP presents challenges for effective feature extraction from electroencephalogram (EEG). Time-frequency transforms are suited for extracting the joint time–frequency features of SSMAEP. Notably, these transforms share structural similarities with convolution operations in convolution neural networks. In this study, we propose a novel time frequency convolutional layer that incorporates structured kernels based on the S transform, continuous wavelet transform (CWT), and short-time Fourier transform (STFT). These time frequency kernels are embedded as learnable filters and replace the conventional first convolutional layer of ShallowConvNet. This design enables the model to more effectively capture SSMAEP signal dynamics across both time and frequency domains. The proposed method was evaluated on two SSMAEP-BCI datasets with two and three auditory targets. Experimental results demonstrate consistent improvements in classification accuracy and robustness compared to baseline models. Furthermore, analysis of the learned kernels revealed that the time frequency filters retained their interpretable structure after training, with task-relevant shifts in center frequency and bandwidth. These findings highlight not only the performance advantage but also the improved interpretability of the proposed model, offering insights into the spectral encoding of SSMAEP-BCI.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"119 ","pages":"Article 109736"},"PeriodicalIF":4.9,"publicationDate":"2026-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146193101","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

Medical image fusion for enhanced edge adaptive Level Set 基于增强边缘自适应水平集的医学图像融合

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-15 Epub Date: 2026-02-12 DOI: 10.1016/j.bspc.2026.109525

Jiao Du , Xiaoyu Yu , Chengxin Su , Qun Zhao

Brain tumors are serious disease, and lesion areas detected by medical imaging typically exhibit distinct edge and contrast information. The aim of medical image fusion method is to synthesize multiple-image information. However, existing methods, while effective in preserving rich information, often struggle to enhance edge and contrast, resulting in artifacts or noise that influencing image quality. In this paper, the input images are first enhanced, and then, based on the model of level set segmentation, an additive bias correction (ABC) level set method is used to adaptively decompose the image into a base layer, a strong edge layer, and a weak edge layer. The sum of the eigenvalues of the covariance matrices (COV) in different directions is used to obtain the weight map for the strong edge layer, while the covariance matrix between image channels is utilized to evaluate the correlation among channels, thereby calculating the weight for fusing the weak edge layer. The experimental results demonstrate that the proposed method achieves an approximately 5% increase in the objective evaluation metrics of Standard Deviation (SD) and Visual Information Fidelity (VIF).To help doctors better observe the characteristics of the lesions.

脑肿瘤是一种严重的疾病，医学成像检测到的病变区域通常具有明显的边缘和对比信息。医学图像融合方法的目的是综合多幅图像信息。然而，现有的方法虽然可以有效地保留丰富的信息，但往往难以增强边缘和对比度，从而导致影响图像质量的伪影或噪声。本文首先对输入图像进行增强，然后在水平集分割模型的基础上，采用加性偏置校正（ABC）水平集方法自适应地将图像分解为基层、强边缘层和弱边缘层。利用不同方向的协方差矩阵（COV）特征值和得到强边缘层的权值映射，利用图像通道间的协方差矩阵评价通道间的相关性，从而计算融合弱边缘层的权值。实验结果表明，该方法在标准偏差（SD）和视觉信息保真度（VIF）的客观评价指标上提高了约5%。帮助医生更好地观察病变特征。

{"title":"Medical image fusion for enhanced edge adaptive Level Set","authors":"Jiao Du , Xiaoyu Yu , Chengxin Su , Qun Zhao","doi":"10.1016/j.bspc.2026.109525","DOIUrl":"10.1016/j.bspc.2026.109525","url":null,"abstract":"<div><div>Brain tumors are serious disease, and lesion areas detected by medical imaging typically exhibit distinct edge and contrast information. The aim of medical image fusion method is to synthesize multiple-image information. However, existing methods, while effective in preserving rich information, often struggle to enhance edge and contrast, resulting in artifacts or noise that influencing image quality. In this paper, the input images are first enhanced, and then, based on the model of level set segmentation, an additive bias correction (ABC) level set method is used to adaptively decompose the image into a base layer, a strong edge layer, and a weak edge layer. The sum of the eigenvalues of the covariance matrices (COV) in different directions is used to obtain the weight map for the strong edge layer, while the covariance matrix between image channels is utilized to evaluate the correlation among channels, thereby calculating the weight for fusing the weak edge layer. The experimental results demonstrate that the proposed method achieves an approximately 5% increase in the objective evaluation metrics of Standard Deviation (SD) and Visual Information Fidelity (VIF).To help doctors better observe the characteristics of the lesions.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"119 ","pages":"Article 109525"},"PeriodicalIF":4.9,"publicationDate":"2026-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146192793","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

Federated learning for prenatal detection of interrupted aortic arch using fetal ultrasound imaging 联合学习用于胎儿超声成像的主动脉弓中断产前检测

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2026-06-15 Epub Date: 2026-02-06 DOI: 10.1016/j.bspc.2026.109795

Jiancheng Han , Heqing Wang , Yifan Feng , Qi Yang , Jingtan Li , Haojie Zhang , Yihua He , Jiang Liu , Toru Nakamura , Yang Cao , Naidi Sun , Kun Qian , Bin Hu , Xinru Gao , Yan Xia , Zongjie Weng , Björn W. Schuller , Yoshiharu Yamamoto

This study presents the first application of federated learning (FL) for prenatal detection of Interrupted Aortic Arch (IAA) using fetal ultrasound images. To address the challenges of data scarcity, privacy constraints, and inter-institutional variability, we develop a federated learning IAA detection method and systematically evaluate three representative strategies (FedAvg, FedProx, and FedBABU) across five clinical centres. Results show that FL improves model performance over local training in recall and F1-score in data-scarce centres. Among FL algorithms, FedAvg and FedProx consistently outperform FedBABU in stability and generalisation. Among the three CNN architectures compared — ResNet-50, EfficientNet-B3, and DenseNet-121 — DenseNet-121 demonstrates superior overall performance, particularly in non-independent and identically distributed (Non-IID) scenarios. Our framework demonstrates the feasibility of collaborative AI for rare disease detection without data sharing, laying the foundation for scalable, real-world prenatal screening of congenital heart defects.

本研究首次应用联邦学习（FL）胎儿超声图像检测主动脉弓中断（IAA）。为了应对数据稀缺、隐私约束和机构间可变性的挑战，我们开发了一种联邦学习IAA检测方法，并系统地评估了五个临床中心的三种代表性策略（FedAvg、FedProx和FedBABU）。结果表明，在数据稀缺的中心，FL比局部训练在召回和f1得分方面提高了模型的性能。在FL算法中，fedag和FedProx在稳定性和泛化方面始终优于FedBABU。在ResNet-50、EfficientNet-B3和DenseNet-121这三种CNN架构中，DenseNet-121表现出了卓越的整体性能，特别是在非独立和同分布（Non-IID）场景中。我们的框架证明了协作人工智能在没有数据共享的情况下进行罕见疾病检测的可行性，为可扩展的、现实世界的先天性心脏缺陷产前筛查奠定了基础。

{"title":"Federated learning for prenatal detection of interrupted aortic arch using fetal ultrasound imaging","authors":"Jiancheng Han , Heqing Wang , Yifan Feng , Qi Yang , Jingtan Li , Haojie Zhang , Yihua He , Jiang Liu , Toru Nakamura , Yang Cao , Naidi Sun , Kun Qian , Bin Hu , Xinru Gao , Yan Xia , Zongjie Weng , Björn W. Schuller , Yoshiharu Yamamoto","doi":"10.1016/j.bspc.2026.109795","DOIUrl":"10.1016/j.bspc.2026.109795","url":null,"abstract":"<div><div>This study presents the first application of federated learning (FL) for prenatal detection of Interrupted Aortic Arch (IAA) using fetal ultrasound images. To address the challenges of data scarcity, privacy constraints, and inter-institutional variability, we develop a federated learning IAA detection method and systematically evaluate three representative strategies (FedAvg, FedProx, and FedBABU) across five clinical centres. Results show that FL improves model performance over local training in recall and F1-score in data-scarce centres. Among FL algorithms, FedAvg and FedProx consistently outperform FedBABU in stability and generalisation. Among the three CNN architectures compared — ResNet-50, EfficientNet-B3, and DenseNet-121 — DenseNet-121 demonstrates superior overall performance, particularly in non-independent and identically distributed (Non-IID) scenarios. Our framework demonstrates the feasibility of collaborative AI for rare disease detection without data sharing, laying the foundation for scalable, real-world prenatal screening of congenital heart defects.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"119 ","pages":"Article 109795"},"PeriodicalIF":4.9,"publicationDate":"2026-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146193096","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