Pub Date : 2026-01-13DOI: 10.1016/j.bspc.2026.109612
Eryang Huan , Hui Dun , Junbao Li
Electrocardiograms (ECG) face challenges in practical applications, including multimodal feature fusion, limited representation capabilities and multi-center data privacy protection as an important tool for cardiovascular disease (CVD) diagnosis. To address these challenges, this paper proposes FedMCF-xLSTM, a federated contrastive xLSTM framework for multimodal multi-label ECG classification. First, we design a multimodal fusion backbone (MF-xLSTM) that jointly encodes raw 12-lead ECG signals via an xLSTM encoder and structured clinical attributes (e.g., age and sex) via a multilayer perceptron, and then fuses the resulting embeddings for multi-label prediction. Second, we introduce the contrastive representation enhancement module (MCF-xLSTM), which applies the random masking and the contrastive loss to encourage compact intra-class clustering and enlarged inter-class margins in the latent space. Finally, we embed the MCF-xLSTM into the federated learning framework, enabling collaborative optimization across multiple clients without sharing raw ECG data and thus preserving patient privacy. Comprehensive experiments on PTB-XL dataset show that our model achieves an accuracy and AUC of 89.28 % and 92.07 %, respectively. Additional experiments on SPH dataset further confirm the robustness of our approach, achieving an accuracy and AUC of 95.16 % and 87.83 %, respectively.
{"title":"FedMCF-xLSTM: Federated contrastive xLSTM for multimodal multi-label ECG classification","authors":"Eryang Huan , Hui Dun , Junbao Li","doi":"10.1016/j.bspc.2026.109612","DOIUrl":"10.1016/j.bspc.2026.109612","url":null,"abstract":"<div><div>Electrocardiograms (ECG) face challenges in practical applications, including multimodal feature fusion, limited representation capabilities and multi-center data privacy protection as an important tool for cardiovascular disease (CVD) diagnosis. To address these challenges, this paper proposes FedMCF-xLSTM, a federated contrastive xLSTM framework for multimodal multi-label ECG classification. First, we design a multimodal fusion backbone (MF-xLSTM) that jointly encodes raw 12-lead ECG signals via an xLSTM encoder and structured clinical attributes (e.g., age and sex) via a multilayer perceptron, and then fuses the resulting embeddings for multi-label prediction. Second, we introduce the contrastive representation enhancement module (MCF-xLSTM), which applies the random masking and the contrastive loss to encourage compact intra-class clustering and enlarged inter-class margins in the latent space. Finally, we embed the MCF-xLSTM into the federated learning framework, enabling collaborative optimization across multiple clients without sharing raw ECG data and thus preserving patient privacy. Comprehensive experiments on PTB-XL dataset show that our model achieves an accuracy and AUC of 89.28 % and 92.07 %, respectively. Additional experiments on SPH dataset further confirm the robustness of our approach, achieving an accuracy and AUC of 95.16 % and 87.83 %, respectively.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"116 ","pages":"Article 109612"},"PeriodicalIF":4.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979987","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}
Pub Date : 2026-01-13DOI: 10.1016/j.bspc.2026.109624
Changkai Ji , Yusheng Liu , Sheng Wang , Yuxian Jiang , Huayan Guo , Wen Xiao , Zhengzhan Lv , Miri Chung , Xingnan Lin , Xueyan Xiong , Lingyong Jiang , Lisheng Wang
Tooth alignment is crucial for oral health and aesthetics, requiring dentists to consider complex spatial relationships among teeth, including local, regional, and global spatial relationships. Manual planning is time-consuming and highly dependent on clinician experience. Therefore, researchers have developed AI techniques for tooth alignment. However, while AI techniques improve planning efficiency, their performance in tooth alignment applications is usually limited. They typically model local tooth neighborhoods and the global dental arch, neglecting the hierarchical multiscale spatial constraints among teeth across ever-expanding regions. To address this issue and effectively use clinical priors, we propose a tooth alignment framework that incorporates a Prior-guided Hierarchical Window Attention (PHWA) module and a Continuous Guidance Diffusion Model (CGDM) to enhance alignment accuracy and efficiency. Specifically, the PHWA module utilizes multilevel window divisions and skip connections to progressively model spatial relationships from local tooth environments to global arch structures, capturing both fine-grained details and comprehensive spatial dependencies of teeth in alignment planning and improving planning accuracy. Furthermore, we apply an improved diffusion model to predict orthodontic transformation matrices. The CGDM module leverages historical estimation information as prior guidance to accelerate convergence toward high-quality alignment schemes. This approach reduces sampling steps while improving the quality of generated samples. Experimental results demonstrate that our method outperforms six state-of-the-art approaches, achieving a superior alignment accuracy of 1.513 mm in ADD and enhanced spatial modeling. Our framework thus presents a promising solution for orthodontic treatment planning.
{"title":"Tooth alignment by diffusion model with hierarchical spatial relationship learning","authors":"Changkai Ji , Yusheng Liu , Sheng Wang , Yuxian Jiang , Huayan Guo , Wen Xiao , Zhengzhan Lv , Miri Chung , Xingnan Lin , Xueyan Xiong , Lingyong Jiang , Lisheng Wang","doi":"10.1016/j.bspc.2026.109624","DOIUrl":"10.1016/j.bspc.2026.109624","url":null,"abstract":"<div><div>Tooth alignment is crucial for oral health and aesthetics, requiring dentists to consider complex spatial relationships among teeth, including local, regional, and global spatial relationships. Manual planning is time-consuming and highly dependent on clinician experience. Therefore, researchers have developed AI techniques for tooth alignment. However, while AI techniques improve planning efficiency, their performance in tooth alignment applications is usually limited. They typically model local tooth neighborhoods and the global dental arch, neglecting the hierarchical multiscale spatial constraints among teeth across ever-expanding regions. To address this issue and effectively use clinical priors, we propose a tooth alignment framework that incorporates a Prior-guided Hierarchical Window Attention (PHWA) module and a Continuous Guidance Diffusion Model (CGDM) to enhance alignment accuracy and efficiency. Specifically, the PHWA module utilizes multilevel window divisions and skip connections to progressively model spatial relationships from local tooth environments to global arch structures, capturing both fine-grained details and comprehensive spatial dependencies of teeth in alignment planning and improving planning accuracy. Furthermore, we apply an improved diffusion model to predict orthodontic transformation matrices. The CGDM module leverages historical estimation information as prior guidance to accelerate convergence toward high-quality alignment schemes. This approach reduces sampling steps while improving the quality of generated samples. Experimental results demonstrate that our method outperforms six state-of-the-art approaches, achieving a superior alignment accuracy of 1.513 mm in ADD and enhanced spatial modeling. Our framework thus presents a promising solution for orthodontic treatment planning.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"116 ","pages":"Article 109624"},"PeriodicalIF":4.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980110","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}
Pub Date : 2026-01-13DOI: 10.1016/j.bspc.2026.109503
Xinwen Yi , Dongyu He , Jiachang Liu , Xiaoling Zhu , Zhifang Pan
Electronic fetal monitoring is widely employed during pregnancy and labor periods to detect fetal hypoxia. Due to internal observer differences, visual inspection of cardiotocography based on clinical guidelines exhibits a high false-positive rate. Therefore, AI-based cardiotocography classification using fetal heart rate signals is essential and challenging, as it assists clinicians in objectively and accurately assessing fetal health status. Most existing methods either focus on single-modal cardiotocography classification based on signals or employ multimodal modeling with signals combined with natural language or other statistical features. However, these methods do not consider the health status of the gravida or fetus. This study is the first to incorporate maternal and fetal health status into the fetal heart rate classification task. The fetal heart rate is transformed into a two-dimensional image, and health status is extracted from the database. Both of them are fed to the network we propose for classification. To address the lack of generalization and robustness in existing methods, we propose a fetal hypoxia diagnosis model based on the vision transformer. Compared to the related works, the proposed method demonstrates strong generalization, achieving 95.582% accuracy with a 97.222% AUC on the public database, and 97.658% accuracy with an 79.910% AUC on the private database. Compared to related works, our proposed model demonstrates the most balanced performance across all metrics. Moreover, experiments conducted in various scenarios demonstrate our model’s strong robustness.
{"title":"HaFeiT: A fetal hypoxia diagnosis model using health status and fetal heart rate based on vision transformer","authors":"Xinwen Yi , Dongyu He , Jiachang Liu , Xiaoling Zhu , Zhifang Pan","doi":"10.1016/j.bspc.2026.109503","DOIUrl":"10.1016/j.bspc.2026.109503","url":null,"abstract":"<div><div>Electronic fetal monitoring is widely employed during pregnancy and labor periods to detect fetal hypoxia. Due to internal observer differences, visual inspection of cardiotocography based on clinical guidelines exhibits a high false-positive rate. Therefore, AI-based cardiotocography classification using fetal heart rate signals is essential and challenging, as it assists clinicians in objectively and accurately assessing fetal health status. Most existing methods either focus on single-modal cardiotocography classification based on signals or employ multimodal modeling with signals combined with natural language or other statistical features. However, these methods do not consider the health status of the gravida or fetus. This study is the first to incorporate maternal and fetal health status into the fetal heart rate classification task. The fetal heart rate is transformed into a two-dimensional image, and health status is extracted from the database. Both of them are fed to the network we propose for classification. To address the lack of generalization and robustness in existing methods, we propose a fetal hypoxia diagnosis model based on the vision transformer. Compared to the related works, the proposed method demonstrates strong generalization, achieving 95.582% accuracy with a 97.222% AUC on the public database, and 97.658% accuracy with an 79.910% AUC on the private database. Compared to related works, our proposed model demonstrates the most balanced performance across all metrics. Moreover, experiments conducted in various scenarios demonstrate our model’s strong robustness.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"116 ","pages":"Article 109503"},"PeriodicalIF":4.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980034","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}
Pub Date : 2026-01-13DOI: 10.1016/j.bspc.2026.109532
Yuqi Cao , Xiaolei Guo , Wentao Liang , Xinao Jin , Yining Zhao , Jiayuan Zhang , Weiting Ge , Pingjie Huang , Dibo Hou , Guangxin Zhang
Objectives
Challenging tasks require larger and deeper model architectures and specialized mechanisms. In the field of medical image segmentation, the scarcity of high-quality annotated datasets limits the ability of complex models to achieve optimal training performance. Therefore, this paper proposes the Minimum Full-Depth link feature fusion U-shaped network (MFD-UNet) for the Immunohistochemical (IHC) image segmentation with limited training data.
Methods
To Address the redundancy issues in UNet++ and UNet3, MFD-UNet strategically reduces depth-link features while preserving essential network depth, thereby enhancing segmentation accuracy and mitigating overfitting risks. Furthermore, MFD-UNet synergistically integrates three complementary mechanisms: self-attention modules for long-range contextual modeling, residual connections for stable gradient propagation, and channel-wise attention for dynamic feature refinement. The proposed method exhibits notable effectiveness in the segmentation for thegland and differentially stained tissue regions.
Results
MFD-UNet achieved DICE accuracies of 92.52% on the public dataset CRAG and 90.69% on the internal dataset CRC, which outperforms the current state-of-the-art U-Net-based methods.
Conclusion
This work promotes the intelligence and generalization of professional medical image diagnosis and is expected to play an important role in areas with limited medical resources.
{"title":"MFD-UNet: minimum full-depth connected U-Net for accurate glandular segmentation in IHC images","authors":"Yuqi Cao , Xiaolei Guo , Wentao Liang , Xinao Jin , Yining Zhao , Jiayuan Zhang , Weiting Ge , Pingjie Huang , Dibo Hou , Guangxin Zhang","doi":"10.1016/j.bspc.2026.109532","DOIUrl":"10.1016/j.bspc.2026.109532","url":null,"abstract":"<div><h3>Objectives</h3><div>Challenging tasks require larger and deeper model architectures and specialized mechanisms. In the field of medical image segmentation, the scarcity of high-quality annotated datasets limits the ability of complex models to achieve optimal training performance. Therefore, this paper proposes the Minimum Full-Depth link feature fusion U-shaped network (MFD-UNet) for the Immunohistochemical (IHC) image segmentation with limited training data.</div></div><div><h3>Methods</h3><div>To Address the redundancy issues in UNet++ and UNet3, MFD-UNet strategically reduces depth-link features while preserving essential network depth, thereby enhancing segmentation accuracy and mitigating overfitting risks. Furthermore, MFD-UNet synergistically integrates three complementary mechanisms: self-attention modules for long-range contextual modeling, residual connections for stable gradient propagation, and channel-wise attention for dynamic feature refinement. The proposed method exhibits notable effectiveness in the segmentation for thegland and differentially stained tissue regions.</div></div><div><h3>Results</h3><div>MFD-UNet achieved DICE accuracies of 92.52% on the public dataset CRAG and 90.69% on the internal dataset CRC, which outperforms the current state-of-the-art U-Net-based methods.</div></div><div><h3>Conclusion</h3><div>This work promotes the intelligence and generalization of professional medical image diagnosis and is expected to play an important role in areas with limited medical resources.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"116 ","pages":"Article 109532"},"PeriodicalIF":4.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980036","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}
Pub Date : 2026-01-13DOI: 10.1016/j.bspc.2026.109533
Zhengheng Yi , Peng Wu , Yiru Wang , Aining Sun , Xinsheng Lai , Zhizhou Zhou
Facial paralysis refers to impaired motor function in the facial muscles, resulting in limited or complete loss of facial expressions and muscle control. Effective evaluation of facial paralysis is essential for patients as it provides objective, accurate, and consistent assessment results. This evaluation facilitates early detection, personalized treatment, and self-monitoring, significantly improving facial functional recovery and quality of life. However, the subjective judgment inconsistency caused by doctors’ personal experiences poses a challenge in facial paralysis evaluation. To address this issue, this study proposes a method that combines handcrafted features and features extracted from neural network to analyze and evaluate the degree of facial paralysis, categorizing it into healthy, mild facial paralysis and severe facial paralysis. We first process the facial images from the YFP (YouTube Facial Paralysis) and Extended Cohn-Kanade (CK+) datasets using the Dlib algorithm, extracting facial landmarks and calculating facial symmetry metrics as handcrafted features. Different neural networks are then employed to extract features from both global and local facial regions. Experimental results show that our method achieves an accuracy of 94.20%, improving by up to 6.75% compared to other methods, demonstrating its significant superiority.
{"title":"A multi-feature fusion approach for intelligent facial paralysis evaluation","authors":"Zhengheng Yi , Peng Wu , Yiru Wang , Aining Sun , Xinsheng Lai , Zhizhou Zhou","doi":"10.1016/j.bspc.2026.109533","DOIUrl":"10.1016/j.bspc.2026.109533","url":null,"abstract":"<div><div>Facial paralysis refers to impaired motor function in the facial muscles, resulting in limited or complete loss of facial expressions and muscle control. Effective evaluation of facial paralysis is essential for patients as it provides objective, accurate, and consistent assessment results. This evaluation facilitates early detection, personalized treatment, and self-monitoring, significantly improving facial functional recovery and quality of life. However, the subjective judgment inconsistency caused by doctors’ personal experiences poses a challenge in facial paralysis evaluation. To address this issue, this study proposes a method that combines handcrafted features and features extracted from neural network to analyze and evaluate the degree of facial paralysis, categorizing it into healthy, mild facial paralysis and severe facial paralysis. We first process the facial images from the YFP (YouTube Facial Paralysis) and Extended Cohn-Kanade (CK+) datasets using the Dlib algorithm, extracting facial landmarks and calculating facial symmetry metrics as handcrafted features. Different neural networks are then employed to extract features from both global and local facial regions. Experimental results show that our method achieves an accuracy of 94.20%, improving by up to 6.75% compared to other methods, demonstrating its significant superiority.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"116 ","pages":"Article 109533"},"PeriodicalIF":4.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979936","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}
Pub Date : 2026-01-13DOI: 10.1016/j.bspc.2026.109467
Smita Samrat Mande, Dhanashri Wategaonkar
Heart disease is a major life-threatening health condition, and the leading cause of death globally. Timely and appropriate detection is essential for effective management and improved patient outcomes. In recent decades, have seen the development of machine learning (ML) and Deep learning (DL) methods in the healthcare industry. This review presents an analysis of 45 research studies based on ML, DL, and ensemble methods, with the significance on the approaches, strengths, challenges, and performance achievements. This comprehensive analysis explores diverse classifiers and their efficacy in heart disease prediction based on the dataset, performance metrics, methodology, and preprocessing techniques. The analysis shows that the deep learning models, specifically the CNN and ANN achieve more accurate outcome than the other detection methods, due to their ability in learning intricate or complex patterns from the data. This systematic review identifies the research gaps and offering valuable insight for developing the robust predictive framework in the future.
{"title":"Comprehensive study and analysis of machine learning and deep learning methods used for heart disease prediction","authors":"Smita Samrat Mande, Dhanashri Wategaonkar","doi":"10.1016/j.bspc.2026.109467","DOIUrl":"10.1016/j.bspc.2026.109467","url":null,"abstract":"<div><div>Heart disease is a major life-threatening health condition, and the leading cause of death globally. Timely and appropriate detection is essential for effective management and improved patient outcomes. In recent decades, have seen the development of machine learning (ML) and Deep learning (DL) methods in the healthcare industry. This review presents an analysis of 45 research studies based on ML, DL, and ensemble methods, with the significance on the approaches, strengths, challenges, and performance achievements. This comprehensive analysis explores diverse classifiers and their efficacy in heart disease prediction based on the dataset, performance metrics, methodology, and preprocessing techniques. The analysis shows that the deep learning models, specifically the CNN and ANN achieve more accurate outcome than the other detection methods, due to their ability in learning intricate or complex patterns from the data. This systematic review identifies the research gaps and offering valuable insight for developing the robust predictive framework in the future.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"116 ","pages":"Article 109467"},"PeriodicalIF":4.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979989","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}
Pub Date : 2026-01-13DOI: 10.1016/j.bspc.2026.109572
Yu Wang , Xiaoqiong Jiang , Yuqi Wang , Xiaoxiao Han , Guangyao Xi , Fuqian Shi , Nilanjan Dey , Fuman Cai
Pressure injuries are prevalent in hospital and residential care settings, imposing a significant economic burden and causing considerable patient distress, especially among paralyzed individuals. The development of a non-invasive and rapid method for early PIs prevention and diagnosis is crucial. This study leverages a novel deep transfer learning framework employed the modern ConvNeXt architecture to identify early pressure injuries (Stage one) and deep tissue pressure injuries from infrared thermal images. This approach enables detection prior to visual manifestation, providing a critical window for intervention. An early pressure injuries infrared thermal image dataset was established for the first time, comprising over 3000 images labelled with three classes: deep tissue pressure injuries, normal, and pressure injuries stage one. Image enhancement and data augmentation techniques were used to address issues of image quality and data imbalance. Several fine-tuned models based on the ResNet, ResNeXt, MobileNet, and ConvNeXt architectures were trained, and a comparative analysis of these models was conducted. The refined ConvNeXt-based model achieved an overall accuracy of 92.97% and an AUC of 0.98 on the test dataset, outperforming others based on state-of-the-art convolutional neural networks. This novel framework provides invaluable assistance to nurses and home care staff in clinical settings, particularly with the smartphone and portable infrared camera, enabling the early diagnosis of pressure injuries and facilitating the optimal use of medical resources while reducing the burden on inpatients.
{"title":"Early detection of pressure injuries via infrared thermography and ConvNeXt","authors":"Yu Wang , Xiaoqiong Jiang , Yuqi Wang , Xiaoxiao Han , Guangyao Xi , Fuqian Shi , Nilanjan Dey , Fuman Cai","doi":"10.1016/j.bspc.2026.109572","DOIUrl":"10.1016/j.bspc.2026.109572","url":null,"abstract":"<div><div>Pressure injuries are prevalent in hospital and residential care settings, imposing a significant economic burden and causing considerable patient distress, especially among paralyzed individuals. The development of a non-invasive and rapid method for early PIs prevention and diagnosis is crucial. This study leverages a novel deep transfer learning framework employed the modern ConvNeXt architecture to identify early pressure injuries (Stage one) and deep tissue pressure injuries from infrared thermal images. This approach enables detection prior to visual manifestation, providing a critical window for intervention. An early pressure injuries infrared thermal image dataset was established for the first time, comprising over 3000 images labelled with three classes: deep tissue pressure injuries, normal, and pressure injuries stage one. Image enhancement and data augmentation techniques were used to address issues of image quality and data imbalance. Several fine-tuned models based on the ResNet, ResNeXt, MobileNet, and ConvNeXt architectures were trained, and a comparative analysis of these models was conducted. The refined ConvNeXt-based model achieved an overall accuracy of 92.97% and an AUC of 0.98 on the test dataset, outperforming others based on state-of-the-art convolutional neural networks. This novel framework provides invaluable assistance to nurses and home care staff in clinical settings, particularly with the smartphone and portable infrared camera, enabling the early diagnosis of pressure injuries and facilitating the optimal use of medical resources while reducing the burden on inpatients.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"116 ","pages":"Article 109572"},"PeriodicalIF":4.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980094","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}
Pub Date : 2026-01-13DOI: 10.1016/j.bspc.2026.109619
Jun Wang, Xianglei Li, Yuxue Gao, Xiwen Qin
Dynamic modelling of infectious diseases is a central tool for disease control and early warning, but there are many challenges to its construction and application, including the widespread presence of longitudinal and missing data. Missing data on infectious diseases, incomplete data collection, the secrecy of patients with asymptomatic infections, and many other factors contribute to the lack of data, which poses a great challenge to the prediction of infectious disease trends. A new method-TPENN (Time-Varying Parameter Estimation Neural Network) is proposed for solving parameter identification and dynamic prediction of infectious disease models. In order to deal the above challenges, the method combines the powerful computational ability of PINN(Physics-informed Neural Network) for differential equations with the powerful ability of GRU (Gated Recurrent Unit) to handle time series data and can better deal missing data. A robust estimation framework is constructed by adding a mask to the GRU input layer to directly exploit the intrinsic structure of the data for dynamic estimation without relying on traditional numerical filling techniques. The method maintains an accurate description of the infectious disease dynamics model while coping with challenges such as missing data and time-varying parameters. In the numerical simulation section, we compare TPENN with PINN, Extended Kalman Filter and Maximum Likelihood-based Extended Kalman Filter(MLE-EKF), and the results show that TPENN has a significant advantage in terms of fitting performance in the case of time-varying parameters. In the empirical analysis section, we validate it based on infectious disease data from the Brazilian state of Amazonas and data from the US state of Tennessee. The experimental results show that the method cannot only accurately fit and predict the real data, but also effectively estimate the time-varying parameters in the infectious disease compartmental model. TPENN provides an accurate and effective solution for modelling the dynamics of infectious diseases, which contributes to in-depth research on the transmission mechanism of infectious diseases.
{"title":"Identification and prediction of time-varying parameters in the SIRD model: A TPENN approach for missing longitudinal data","authors":"Jun Wang, Xianglei Li, Yuxue Gao, Xiwen Qin","doi":"10.1016/j.bspc.2026.109619","DOIUrl":"10.1016/j.bspc.2026.109619","url":null,"abstract":"<div><div>Dynamic modelling of infectious diseases is a central tool for disease control and early warning, but there are many challenges to its construction and application, including the widespread presence of longitudinal and missing data. Missing data on infectious diseases, incomplete data collection, the secrecy of patients with asymptomatic infections, and many other factors contribute to the lack of data, which poses a great challenge to the prediction of infectious disease trends. A new method-TPENN (Time-Varying Parameter Estimation Neural Network) is proposed for solving parameter identification and dynamic prediction of infectious disease models. In order to deal the above challenges, the method combines the powerful computational ability of PINN(Physics-informed Neural Network) for differential equations with the powerful ability of GRU (Gated Recurrent Unit) to handle time series data and can better deal missing data. A robust estimation framework is constructed by adding a mask to the GRU input layer to directly exploit the intrinsic structure of the data for dynamic estimation without relying on traditional numerical filling techniques. The method maintains an accurate description of the infectious disease dynamics model while coping with challenges such as missing data and time-varying parameters. In the numerical simulation section, we compare TPENN with PINN, Extended Kalman Filter and Maximum Likelihood-based Extended Kalman Filter(MLE-EKF), and the results show that TPENN has a significant advantage in terms of fitting performance in the case of time-varying parameters. In the empirical analysis section, we validate it based on infectious disease data from the Brazilian state of Amazonas and data from the US state of Tennessee. The experimental results show that the method cannot only accurately fit and predict the real data, but also effectively estimate the time-varying parameters in the infectious disease compartmental model. TPENN provides an accurate and effective solution for modelling the dynamics of infectious diseases, which contributes to in-depth research on the transmission mechanism of infectious diseases.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"116 ","pages":"Article 109619"},"PeriodicalIF":4.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979986","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}
Diabetic retinopathy (DR) is a leading cause of vision loss, making accurate, early screening essential. We present a lightweight, self-supervised hybrid CNN for automated DR detection from retinal fundus images that fuses MobileNet and EfficientNet with a residual attention mechanism for multi-scale, DR-specific feature extraction, and combines predictions in an ensemble with ResNet for robustness. The training pipeline includes a two-stage self-supervised warm start (general EyePACS pretraining followed by domain-specific refinement), a binary-to-five-class curriculum schedule to stabilize optimization, Grad-CAM for visual interpretability, and an uncertainty-aware referral module based on Monte Carlo Dropout. Evaluated on public (APTOS, Messidor) and private datasets under stratified 5-fold cross-validation with strict patient-level separation and external held-out testing on an independent private dataset, the framework achieves state-of-the-art performance with a macro F1-score of 98.1% and an AUC of 99.2% (mean across test folds), representing a 2.1 percentage-point gain over an ImageNet-initialized baseline. The uncertainty module flags 75% of misclassified cases for expert review, while a fairness audit stratified by image quality shows stable performance with a worst-case AUC drop of only 1.2 percentage points. Inference is efficient (approximately 45 ms on a consumer GPU for the full ensemble and approximately 180 ms on a mobile-class CPU for a single forward pass of the hybrid backbone), supporting deployment from clinic to edge. These results indicate that a self-supervised hybrid CNN with explainability and uncertainty-aware referral can deliver accurate, reliable, and equitable DR screening with promising potential for real-world clinical workflows.
{"title":"A self-supervised hybrid CNN with uncertainty-aware referral for diabetic retinopathy screening","authors":"Neelapala Anil Kumar , D. Madhusudan , Iacovos Ioannou , G.S. Pradeep Ghantasala , Vasos Vassiliou","doi":"10.1016/j.bspc.2026.109482","DOIUrl":"10.1016/j.bspc.2026.109482","url":null,"abstract":"<div><div>Diabetic retinopathy (DR) is a leading cause of vision loss, making accurate, early screening essential. We present a lightweight, self-supervised hybrid CNN for automated DR detection from retinal fundus images that fuses MobileNet and EfficientNet with a residual attention mechanism for multi-scale, DR-specific feature extraction, and combines predictions in an ensemble with ResNet for robustness. The training pipeline includes a two-stage self-supervised warm start (general EyePACS pretraining followed by domain-specific refinement), a binary-to-five-class curriculum schedule to stabilize optimization, Grad-CAM for visual interpretability, and an uncertainty-aware referral module based on Monte Carlo Dropout. Evaluated on public (APTOS, Messidor) and private datasets under stratified 5-fold cross-validation with strict patient-level separation and external held-out testing on an independent private dataset, the framework achieves state-of-the-art performance with a macro F1-score of 98.1% and an AUC of 99.2% (mean across test folds), representing a 2.1 percentage-point gain over an ImageNet-initialized baseline. The uncertainty module flags 75% of misclassified cases for expert review, while a fairness audit stratified by image quality shows stable performance with a worst-case AUC drop of only 1.2 percentage points. Inference is efficient (approximately 45 ms on a consumer GPU for the full ensemble and approximately 180 ms on a mobile-class CPU for a single forward pass of the hybrid backbone), supporting deployment from clinic to edge. These results indicate that a self-supervised hybrid CNN with explainability and uncertainty-aware referral can deliver accurate, reliable, and equitable DR screening with promising potential for real-world clinical workflows.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"116 ","pages":"Article 109482"},"PeriodicalIF":4.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979983","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}
Pub Date : 2026-01-13DOI: 10.1016/j.bspc.2026.109526
Monserrat Pallares Di Nunzio , Mauro Granado , Federico Miceli , Santiago Collavini , Fernando Montani
Refractory epilepsy poses a significant clinical challenge, as some patients do not respond to pharmacological treatments targeting the characteristic seizures of the disease. While surgical resection of the affected regions can be an effective intervention, it is not always feasible, forcing patients to cope with a diminished quality of life. The use of intracranial electrodes () provides signals with high spatial and temporal resolution, allowing the identification of three principal brain states: basal, postictal, and preictal. Early detection of the preictal state is critical, as it facilitates seizure prediction several minutes in advance.
In this study, mutual information (MI) was utilized to analyze both linear and nonlinear statistical dependencies in multichannel time series. By quantifying information transmission across neural rhythms associated with state epilepticus, MI demonstrated robustness as a biomarker with significant potential for early seizure detection. Furthermore, MI-based metrics derived from different frequency bands were incorporated into supervised learning models, enabling accurate classification of the preictal state with 81% accuracy.
This preliminary study, conducted on data from four patients, suggests that MI-based analysis may represent a promising biomarker for the detection of the preictal state. Nevertheless, the generalization of these findings requires further validation in future studies involving larger and more heterogeneous cohorts. By facilitating seizure prediction at an early stage, this approach holds promise as a tool to improve the quality of life for patients with refractory epilepsy.
{"title":"Mutual information analysis of intracranial EEG for the detection of preictal brain state in refractory epilepsy","authors":"Monserrat Pallares Di Nunzio , Mauro Granado , Federico Miceli , Santiago Collavini , Fernando Montani","doi":"10.1016/j.bspc.2026.109526","DOIUrl":"10.1016/j.bspc.2026.109526","url":null,"abstract":"<div><div>Refractory epilepsy poses a significant clinical challenge, as some patients do not respond to pharmacological treatments targeting the characteristic seizures of the disease. While surgical resection of the affected regions can be an effective intervention, it is not always feasible, forcing patients to cope with a diminished quality of life. The use of intracranial electrodes (<span><math><mrow><mi>i</mi><mi>E</mi><mi>E</mi><mi>G</mi></mrow></math></span>) provides signals with high spatial and temporal resolution, allowing the identification of three principal brain states: basal, postictal, and preictal. Early detection of the preictal state is critical, as it facilitates seizure prediction several minutes in advance.</div><div>In this study, mutual information (MI) was utilized to analyze both linear and nonlinear statistical dependencies in multichannel <span><math><mrow><mi>i</mi><mi>E</mi><mi>E</mi><mi>G</mi></mrow></math></span> time series. By quantifying information transmission across neural rhythms associated with state epilepticus, MI demonstrated robustness as a biomarker with significant potential for early seizure detection. Furthermore, MI-based metrics derived from different frequency bands were incorporated into supervised learning models, enabling accurate classification of the preictal state with 81% accuracy.</div><div>This preliminary study, conducted on data from four patients, suggests that MI-based analysis may represent a promising biomarker for the detection of the preictal state. Nevertheless, the generalization of these findings requires further validation in future studies involving larger and more heterogeneous cohorts. By facilitating seizure prediction at an early stage, this approach holds promise as a tool to improve the quality of life for patients with refractory epilepsy.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"116 ","pages":"Article 109526"},"PeriodicalIF":4.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980037","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}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号