Biomedical Signal Processing and Control最新文献

{"title":"Semi-supervised medical image segmentation with confidence-aware learning","authors":"Huanqing Xu ,&nbsp;Bo Peng ,&nbsp;Jie Zhu ,&nbsp;Yi Cao ,&nbsp;Jiahui Song ,&nbsp;Jianjun Lei","doi":"10.1016/j.bspc.2025.108439","DOIUrl":"10.1016/j.bspc.2025.108439","url":null,"abstract":"<div><div>Medical image segmentation is emerging as a hot research topic due to its promising application prospects in disease diagnosis and treatment. While deep learning-based medical image segmentation has made significant progress, most of them rely on fully supervised learning, requiring extensive expert annotations. To mitigate this limitation, there has been a growing interest in the study of semi-supervised medical image segmentation, which leverages a small amount of labeled data and abundant unlabeled data to achieve accurate segmentation. In this paper, a novel confidence-aware semi-supervised medical image segmentation method is proposed to construct precise and reliable supervision by designing an adaptive pixel-wise weighting approach for different regions. Specifically, for uncertain regions, an entropy-weighted consistency optimization mechanism is proposed to enable more focus on challenging pixels, thereby making the network reduce prediction errors. Besides, for certain regions, a dynamic weighting pseudo supervision strategy is proposed to emphasize reliable pixels within the pseudo labels, thus obtaining more effective pseudo supervision. Experimental results on benchmark datasets demonstrate the superiority of the proposed method, with a Dice score of 86.33% on the BraTS 2019 dataset, under the 20% labeled data setting.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"112 ","pages":"Article 108439"},"PeriodicalIF":4.9,"publicationDate":"2025-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144932699","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}

{"title":"Stenosis detection of coronary arteries in X-ray angiography images using Swin UNETR with self-supervised pre-training approach and PEFT strategy","authors":"Mehrshad Lalinia,&nbsp;Farshad Almasganj,&nbsp;Seyyed Ali Seyyedsalehi","doi":"10.1016/j.bspc.2025.108617","DOIUrl":"10.1016/j.bspc.2025.108617","url":null,"abstract":"<div><div>Coronary angiography remains the main diagnostic tool for coronary artery disease (CAD), the world’s leading cause of death. The severity of CAD is determined by the site, extent of narrowing (stenosis), and the number of affected arteries. This paper presents a 2D Swin UNETR architecture for automated coronary artery stenosis detection in XCA images, employing fine-tuning and adapter-tuning strategies to optimize accuracy and computational efficiency. The model integrates binary segmentation as a pre-processing step and small segment elimination as post-processing to enhance precision, with label smoothing applied during training to reduce overfitting and improve generalization. The 2D Swin UNETR model demonstrated superior performance compared to existing models ,achieving a Dice score of 0.5375 with fine-tuning and 0.5519 with adapter-tuning, offering greater computational efficiency and accuracy in detecting stenoses and segmenting coronary arteries, making it suitable for clinical use.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"112 ","pages":"Article 108617"},"PeriodicalIF":4.9,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144922104","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}

{"title":"Securing deep learning models with differential privacy for cardiovascular disease prediction","authors":"Zoher Orabe,&nbsp;Antti Vasankari,&nbsp;Tapio Pahikkala,&nbsp;Matti Kaisti,&nbsp;Antti Airola","doi":"10.1016/j.bspc.2025.108502","DOIUrl":"10.1016/j.bspc.2025.108502","url":null,"abstract":"<div><div>This study investigates how differential privacy (DP) can enhance data confidentiality in deep learning models for predicting cardiovascular diseases (CVDs) using electrocardiography (ECG) data collected from various hospitals. We evaluated the privacy–utility trade-off by analyzing model performance under different privacy budgets (<span><math><mi>ϵ</mi></math></span>) across different model architectures, including the high-capacity ResNet with squeeze-and-excitation (ResNet-SE), transformer-based model, and two simple baselines: logistic regression (LR) and multilayer perceptrons (MLP). The original ResNet-SE model, with 8.81 million parameters, showed substantial performance degradation under DP with macro- and micro-average AUCs decreasing from 0.90 and 0.92 to 0.79 and 0.82 at <span><math><mrow><mi>ϵ</mi><mo>=</mo><mn>10</mn></mrow></math></span>. By reducing the model size by 98.4% to 142,934 parameters, we achieved a better balance between accuracy and privacy, with macro- and micro-average AUCs of 0.87 and 0.89, only 0.03 lower than its non-private performance. The transformer-based model showed weaker robustness to DP, with a macro- and micro-average AUCs dropping from 0.88 and 0.91 to 0.64 and 0.73, while LR and MLP baselines trained on ECG handcrafted features achieved low performance even without privacy. The effect of training with DP varied across classes, having only minimal impact on the four largest classes (AUC reduction <span><math><mo>≤</mo></math></span> 0.01), but more substantial performance decreases were observed for many of the smaller classes (e.g. 0.10 drop for a condition with a 1.19% class size, and a drop of 0.28 for condition with class size of 3.10%). Overall, our study demonstrates the positive effect of reducing model complexity for improving privacy-utility trade-off for predicting CVDs.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"112 ","pages":"Article 108502"},"PeriodicalIF":4.9,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144922105","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}

{"title":"A dynamic Bi-CSAM-Net for brain tumor segmentation","authors":"Ernest Asimeng ,&nbsp;Zhe Liu ,&nbsp;Liang Yin ,&nbsp;Kai Han ,&nbsp;Issahaku Fadilul-lah Yassaanah ,&nbsp;Qiaoying Teng","doi":"10.1016/j.bspc.2025.108615","DOIUrl":"10.1016/j.bspc.2025.108615","url":null,"abstract":"<div><div>The accurate segmentation of brain tumors from MRI scans is a critical task in medical image analysis and is essential for clinical diagnosis, treatment planning, and monitoring. Despite advancements, many existing segmentation methods focus primarily on extracting global semantic features from 3D volumetric data, often neglecting finer voxel-specific details, inter-layer connections, and subtle anatomical variations. To address these limitations, we propose a novel deep learning architecture, Bi-CSAM-Net, that integrates advanced data preprocessing techniques, attention mechanism, and the CNN model as a benchmark for segmentation tasks. The preprocessing techniques use dynamic contrast enhancement, multi-resolution data integration, and patient-specific normalization (DCA-TR) to focus on the finer voxel-specific features to enhance model performance. The attention mechanism called the BiC-AM employs a Bi-LSTM to enhance the Channel Attention Mechanism within a Convolutional Block Attention Module (CBAM) framework to capture spatial and sequential MRI data dependencies better. Our extensive evaluation of two benchmark datasets; ASNR-MICCAI-BraTs 2021 and BraTs Adult-Giloma 2023, reveals that the proposed model achieves a high segmentation Dice score for all the modalities, consistently outperforming existing methods, specifically the whole tumor (WT) and tumor core (TC) for the BraTs 2021 and 2023 datasets. The results also demonstrate significant improvements in sensitivity and specificity across various tumor sub-regions, underscoring the model’s robustness. Ablation studies further validate the contributions of our proposed modules, highlighting their effectiveness in enhancing segmentation performance. These findings underscore the potential clinical applicability of Deep Bi-CSAM-Net for accurate and reliable brain tumor segmentation.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"112 ","pages":"Article 108615"},"PeriodicalIF":4.9,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144922827","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}

{"title":"Enhancing patient treatment classification in healthcare: A comparative analysis of machine learning techniques on electronic health record data","authors":"Cheng yang,&nbsp;Xiaodong Su","doi":"10.1016/j.bspc.2025.108543","DOIUrl":"10.1016/j.bspc.2025.108543","url":null,"abstract":"<div><div>This study is directed at the essential task of patient treatment classification, which is essential for personalized and optimal healthcare. The overall aim is to optimize the accuracy of classification models in order to allocate treatments correctly and avoid the risk of misdiagnosis or ineffective treatments. In pursuit of this aim, the research compares different machine-learning models with the aim of improving the present classification models. The approach involves the assessment of widely used models like Gradient Boosting, K-Nearest Neighbors (KNN), Logistic Regression, Random Forest, and support vector machines (SVM) using an electronic health record (EHR) corpus. By extensive training, cross-validation, and testing, the research compares the performances of each model in patient treatment classification. The results present notable disparities in the performances of the models, with the emphasis being the necessity of the selection of the optimal model for the process. The applied value of the research is significant. By ascertaining the optimal machine learning model for patient treatment classification, the results can be used directly to optimize the accuracy and trustworthiness of healthcare models, ultimately contributing to improved patient results. This research is unique in the extensive assessment of multiple machine-learning models specifically for patient treatment classification. It establishes the optimal model with the foundation for the creation of more trusty and resilient classification models in healthcare.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"112 ","pages":"Article 108543"},"PeriodicalIF":4.9,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144922868","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}

{"title":"Pain intensity estimation via multimodal fusion: Leveraging ternary textures of derivatives in EDA and PPG signals","authors":"Muhammad Umar Khan ,&nbsp;Niraj Hirachan ,&nbsp;Calvin Joseph ,&nbsp;Luke Murtagh ,&nbsp;Girija Chetty ,&nbsp;Roland Goecke ,&nbsp;Raul Fernandez-Rojas","doi":"10.1016/j.bspc.2025.108532","DOIUrl":"10.1016/j.bspc.2025.108532","url":null,"abstract":"<div><div>In the event of pain, the autonomic nervous system reacts by affecting different physiological parameters such as blood pressure, heart rate, skin conductance, and perspiration levels, among others. This research presents an innovative approach to pain intensity recognition through a multimodal system that fuses bio-information from the skin (Electrodermal Activity or EDA) and heart (Photoplethysmograph or PPG) signals. The study involved a self-collected dataset from 61 healthy participants and encompassed two pain intensity levels (low and high) experienced at different anatomical locations (hand and forearm). Employing IIR bandpass filters, the collected EDA and PPG signals were preprocessed. A novel feature extraction method named Ternary Textures of Derivatives (TTD) is proposed, which, when fused with statistical features, exhibited robust potential as a pain intensity biomarker. Feature selection using Joint Mutual Information preceded the utilisation of an Ensemble classifier. The developed multimodal fusion-based pain recognition system outperformed the unimodal (PPG and EDA) approaches by achieving notable accuracies: <span><math><mrow><mn>83</mn><mo>.</mo><mn>1</mn><mtext>%</mtext><mo>±</mo><mn>8</mn><mo>.</mo><mn>8</mn><mtext>%</mtext></mrow></math></span> for No Pain vs. Low Pain, <span><math><mrow><mn>87</mn><mo>.</mo><mn>1</mn><mtext>%</mtext><mo>±</mo><mn>6</mn><mo>.</mo><mn>7</mn><mtext>%</mtext></mrow></math></span> for No Pain vs. High Pain, and <span><math><mrow><mn>74</mn><mo>.</mo><mn>5</mn><mtext>%</mtext><mo>±</mo><mn>6</mn><mo>.</mo><mn>8</mn><mtext>%</mtext></mrow></math></span> for the No Pain vs. Low Pain vs. High Pain scenario. This approach offers an objective means of pain assessment that can furnish valuable insights to clinical teams, aiding in treatment evaluation, surgical decision-making, and overall patient care quality assessment.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"112 ","pages":"Article 108532"},"PeriodicalIF":4.9,"publicationDate":"2025-08-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144920030","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}

{"title":"EZhouNet: A framework based on graph neural network and anchor interval for the respiratory sound event detection","authors":"Yun Chu ,&nbsp;Qiuhao Wang ,&nbsp;Enze Zhou ,&nbsp;Qian Liu ,&nbsp;Gang Zheng","doi":"10.1016/j.bspc.2025.108491","DOIUrl":"10.1016/j.bspc.2025.108491","url":null,"abstract":"<div><div>Auscultation is a key method for early diagnosis of respiratory and pulmonary diseases, relying on skilled healthcare professionals. However, the process is often subjective, with variability between experts. As a result, numerous deep learning-based automatic classification methods have emerged, most of which focus on respiratory sound classification. In contrast, research on respiratory sound event detection remains limited. Existing sound event detection methods typically rely on frame-level predictions followed by post-processing to generate event-level outputs, making interval boundaries challenging to learn directly. Furthermore, many approaches can only handle fixed-length audio, limiting their applicability to variable-length respiratory sounds. Additionally, the impact of respiratory sound location information on detection performance has not been extensively explored. To address these issues, we propose a graph neural network-based framework with anchor intervals, capable of handling variable-length audio and providing more precise temporal localization for abnormal respiratory sound events. Our method improves both the flexibility and applicability of respiratory sound detection. Experiments on the SPRSound 2024 and HF Lung V1 datasets demonstrate the effectiveness of the proposed approach, and incorporating respiratory position information enhances the discrimination between abnormal sounds.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"112 ","pages":"Article 108491"},"PeriodicalIF":4.9,"publicationDate":"2025-08-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144922096","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}

{"title":"Self-DenseMobileNet: A robust framework for lung nodule classification using Self-ONN and stacking-based meta-classifier","authors":"Md. Sohanur Rahman ,&nbsp;Muhammad E.H. Chowdhury ,&nbsp;Hasib Ryan Rahman ,&nbsp;Mosabber Uddin Ahmed ,&nbsp;Muhammad Ashad Kabir ,&nbsp;Rusab Sarmun","doi":"10.1016/j.bspc.2025.108588","DOIUrl":"10.1016/j.bspc.2025.108588","url":null,"abstract":"<div><div>Accurate classification of lung nodules can significantly improve treatment outcomes and patient survival rates by enabling early detection and preemptive intervention in cases of suspected lung cancer. Due to the limited availability of sufficiently comprehensive chest X-ray (CXR) datasets, we utilized a composite dataset from the NODE21 challenge to classify nodules and non-nodules in chest radiographs. In this study, we propose a novel computer-aided diagnosis (CAD) framework that includes image standardization and advanced image enhancement techniques for improved quality. We integrated four state-of-the-art deep learning (DL) models such as DenseNet201, MobileViTv2-0.50, MobileViTv2-0.75, and ResNet152, into this framework. We also introduced our own proposed Self-DenseMobileNet, which was trained on four image-enhanced variants of our dataset. The prediction probabilities from this model were transformed into tabular form, and then eight classical machine learning (ML) models were trained. The top three performing classical models were subsequently combined using a stacking algorithm to create a meta-classifier. Finally, We used class activation mapping (CAM) to demonstrate the interpretability of the best-performing model. Our proposed framework achieved an accuracy of 99.28%, a precision of 99.60%, a recall of 99.47%, a specificity of 98.68%, an F1-score of 99.53%, and an area under the curve (AUC) of 0.99 on internal validation images using a Meta-Random Forest Classifier. When tested on a completely different dataset, the external validation results showed an accuracy of 89.40%, a precision of 90.08%, a recall of 92.58%, a specificity of 84.6%, and an F1-score of 91.31%, and an AUC of 0.90. The novel framework resulted in a significant performance enhancement in classifying CXRs with lung nodules. The classification reasoning of the models can also be explained by ScoreCam heatmaps, which further solidifies its robustness. Our proposed model will assist healthcare experts in making informed decisions regarding lung cancer.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"112 ","pages":"Article 108588"},"PeriodicalIF":4.9,"publicationDate":"2025-08-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144922870","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}

{"title":"Exploring multi-scale time group for common spatial pattern feature based motor imagery EEG classification","authors":"Jian-Xun Mi ,&nbsp;Rong-Feng Li ,&nbsp;Ke Liu ,&nbsp;Weisheng Li","doi":"10.1016/j.bspc.2025.108591","DOIUrl":"10.1016/j.bspc.2025.108591","url":null,"abstract":"<div><div>Feature extraction is a pivotal challenge in brain–computer interface (BCI) systems that utilize motor imagery (MI). Discriminative features are crucial for enhancing the classification accuracy of MI tasks within MI-BCI systems. The common spatial pattern (CSP) method has been extensively employed for extracting band-power features in this context. Prior research indicates that the efficacy of the CSP algorithm can be significantly improved by identifying optimal filter bands or selecting appropriate time windows. Most existing methods adopt a fixed-size time window, which may overlook EEG patterns occurring in windows of varying sizes. Furthermore, current approaches have not explored the potential of optimizing the filtering band for different time windows, thereby potentially compromising the effectiveness of the CSP algorithm. In this paper, we introduce a novel method called multi-scale time group common spatial pattern (MTGCSP), which aims to optimize both the time window for MI tasks based on a multi-scale sliding time window preprocessing strategy and the filtering band for each time window at a specific scale. Specifically, the EEG signals are initially filtered across multiple sub-bands. Subsequently, we use the multi-scale sliding time window preprocessing strategy to segment the filtered spectral signals into multiple subsequences of varying lengths using the multi-scale sliding window. To extract robust CSP features from the multi-scale sliding time window subsequences, we propose a sparse joint optimization objective function that incorporates sparse group constraints. The resulting feature subset is then fed into a support vector machine (SVM) classifier with a linear kernel to perform MI-EEG classification tasks. Experimental results from three public datasets demonstrate that the MTGCSP method outperforms other state-of-the-art techniques.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"112 ","pages":"Article 108591"},"PeriodicalIF":4.9,"publicationDate":"2025-08-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144920186","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}

{"title":"HHGCN: Hybrid Hypergraph Convolutional and Graph Convolutional Network for prognostic prediction of intracerebral hemorrhage","authors":"Haowei Duan ,&nbsp;Wensong Yang ,&nbsp;Yuqi Ma ,&nbsp;Jingliu He ,&nbsp;Peng Xie ,&nbsp;Shanxiong Chen","doi":"10.1016/j.bspc.2025.108568","DOIUrl":"10.1016/j.bspc.2025.108568","url":null,"abstract":"<div><h3>Problem:</h3><div>Intracerebral hemorrhage (ICH) is a severe form of stroke characterized by high mortality and disability rates. Accurate early prognosis is crucial for guiding clinical treatment strategies, yet predicting ICH outcomes remains challenging due to the disease’s complexity and the diverse information modalities involved.</div></div><div><h3>Aim:</h3><div>This study aims to enhance the accuracy of early prognostic prediction for ICH by proposing a Hybrid Hypergraph Convolutional and Graph Convolutional Network (HHGCN). The network integrates imaging features, radiomic features, and clinical features, modeling them as graph and hypergraph structures to capture the intricate relationships within and between modalities.</div></div><div><h3>Methods:</h3><div>Utilizes a pre-trained 3D ResNet34 model to extract deep learning image features, radiomic techniques to process medical images, along with structured clinical scales. These features are structured into graph and hypergraph frameworks, allowing for intra-modal and inter-modal feature extraction through graph convolution and hypergraph convolution. A Hybrid Modal feature Fusion (HMFF) module is designed to synthesize these features, enhancing the model’s predictive capabilities.</div></div><div><h3>Results:</h3><div>Through cross-validation on a multimodal ICH prognosis dataset, achieved an accuracy of 81.47%, an F1 score of 0.8158, and an AUC value of 0.8433, outperforming other advanced methods.</div></div><div><h3>Conclusion:</h3><div>Proposes a graph and hypergraph-based model for ICH prognosis, which integrates multimodal data to enhance prediction accuracy, offering a robust framework for early prognostic prediction of ICH. Its integration of multimodal data through advanced graph and hypergraph convolutional techniques provides a comprehensive and accurate predictive tool.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"112 ","pages":"Article 108568"},"PeriodicalIF":4.9,"publicationDate":"2025-08-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144920187","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}