Biomedical Signal Processing and Control最新文献

{"title":"SGT-Net: Spectral generalized transformer for skin lesion segmentation","authors":"Zeye Xu,&nbsp;Jian Ji,&nbsp;Falin Wang,&nbsp;Teng Sun,&nbsp;Junkun Li","doi":"10.1016/j.bspc.2025.108214","DOIUrl":"10.1016/j.bspc.2025.108214","url":null,"abstract":"<div><div>Skin lesion segmentation is vital for melanoma diagnosis. Existing deep-learning models face issues like poor generalization due to image noise and unique-shaped images. This paper presents the Spectral Generalized Transformer, a new network based on the classic encoder–decoder. It includes a redesigned Spectral Adaptive Block (SAB) for better feature capture and connection. SAB uses Fourier Transform to convert features, extracts key features by frequency. A Feature-generalization Decoder (FD) is developed for spectral features. With the reverse attention mechanism and multi-scale fusion, it enhances edge segmentation and model generalization. Evaluated on four public datasets, the model’s mIou on PH2, ISIC2016, ISIC2017, and ISIC2018 reached 92.28%, 87.01%, 83.61%, and 81.52% respectively. The results show it surpasses state-of-the-art methods. Code available at: <span><span>https://github.com/1914010125/Zeye-Xu</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"111 ","pages":"Article 108214"},"PeriodicalIF":4.9,"publicationDate":"2025-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144632595","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":"Semi-supervised medical image segmentation method using multi-scale consistency adversarial learning","authors":"Yujie Feng ,&nbsp;Xue Tang ,&nbsp;Qiuyu Sun ,&nbsp;Weisheng Li ,&nbsp;Shenhai Zheng","doi":"10.1016/j.bspc.2025.108250","DOIUrl":"10.1016/j.bspc.2025.108250","url":null,"abstract":"<div><div>Medical image segmentation is an essential task in medical image analysis, playing a crucial role in improving the accuracy of disease diagnosis and the efficiency of treatment planning. Despite extensive research efforts and notable technological advancements of deep learning, its performance is often limited by the necessity for vast amounts of accurately labeled images, which are inherently costly and labor-intensive to procure. To mitigate this challenge, this study proposes a novel semi-supervised segmentation using Multi-Scale Consistency Adversarial Learning (MSCAL). By leveraging few annotated images, this method constructs a comprehensive data augmentation perturbation space, incorporating both image-level strong–weak perturbations alongside multi-scale feature perturbations. Furthermore, strong–weak consistency regularization and multi-scale adversarial learning strategies across diverse scales of the segmentation network are implemented. Furthermore, the method utilizes adaptive weighted pyramid consistency loss to encourage consistent predictions across scales, and emphasizes consistency in high-confidence regions through the confidence maps outputted by the discriminator. Finally, the advantages of our proposed model are rigorously evaluated on the ACDC and BraTS2019 datasets, where it is systematically compared against ten state-of-the-art semi-supervised methods. Experimental results demonstrate the model’s superiority across DSC, HD95, ASD metrics and <span><math><mi>p</mi></math></span>-value. Notably, with only 3 annotated samples, this method achieves at least 4.2%, 2.2%, and 1.6% gains in segmentation accuracy for the right ventricle, myocardium, and left ventricle, respectively. Ablation studies further corroborate this innovative framework enables the acquisition of richer feature representations and bolstering model robustness for semi-supervised medical image segmentation tasks.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"111 ","pages":"Article 108250"},"PeriodicalIF":4.9,"publicationDate":"2025-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144632594","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":"Assessment of quality of electrocardiograms, seismocardiograms, and gyrocardiograms based on features derived from symmetric projection attractor reconstruction in healthy subjects","authors":"Szymon Sieciński ,&nbsp;Muhammad Tausif Irshad ,&nbsp;Md Abid Hasan ,&nbsp;Rafał Doniec ,&nbsp;Paweł Kostka ,&nbsp;Ewaryst Tkacz ,&nbsp;Marcin Grzegorzek","doi":"10.1016/j.bspc.2025.108170","DOIUrl":"10.1016/j.bspc.2025.108170","url":null,"abstract":"<div><div>Signal quality assessment is essential for biomedical signal processing, analysis, and interpretation. Various methods exist, including averaged numerical values, thresholding, time- or frequency-domain analysis, and nonlinear approaches. The aim of this study was to evaluate the quality of electrocardiographic (ECG) signals, seismocardiographic signals (SCG), and gyrocardiograms (GCG) based on symmetric projection attractor reconstruction (SPAR) with Takens delay coordinates with fit five classifiers: random forest, gradient boosting, random forest XGB, and support vector machines (SVM) with various number of decision tree-based estimators (100–10,000) and various kernels (linear, radial base function, and polynomial), respectively. The analysis was carried out on a public dataset “Mechanocardiograms with ECG reference” containing 29 concurrent ECG, SCG, and GCG signal recordings. The highest values without SMOTE were observed for ECG signals, SVM with fourth order polynomial kernel (accuracy of 0.6897, PPV of 0.6019, sensitivity of 0.5306, and F1 score of 0.4952), and after applying SMOTE were observed for Gradient Boosting in ECG signal (200 estimators, accuracy 0.7500, PPV of 0.7747, sensitivity of 0.7500, and F2 score of 0.7747 respectively). These findings suggest that the SPAR-based approach is a promising method to accurately assess the quality of cardiovascular signals, including seismocardiograms and gyrocardiograms.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"111 ","pages":"Article 108170"},"PeriodicalIF":4.9,"publicationDate":"2025-07-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144614748","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":"Electrocardiographic alternans as an additional criterion for cardioverter defibrillator implantation in primary prevention of sudden cardiac death","authors":"Ilaria Marcantoni ,&nbsp;Erica Iammarino ,&nbsp;Agnese Sbrollini ,&nbsp;Micaela Morettini ,&nbsp;Cees A. Swenne ,&nbsp;Laura Burattini","doi":"10.1016/j.bspc.2025.108322","DOIUrl":"10.1016/j.bspc.2025.108322","url":null,"abstract":"<div><div>The current Guidelines recommend implantable cardioverter defibrillator (ICD) for primary prevention of sudden cardiac death (SCD) when left ventricular ejection fraction (LVEF) is reduced. Nevertheless, LVEF lacks sensitivity and specificity as a risk index, meaning that additional risk indexes are needed. Electrocardiographic alternans (ECGA) is the every-other-beat morphology oscillation in either ECG wave: P-wave/QRS-complex/T-wave alternans (PWA/QRSA/TWA, respectively). This study aims to investigate ECGA as an additional criterion to decide for ICD implantation for primary prevention of SCD.</div><div>ECGs were acquired during a bicycle-ergometer test in a heart-failure population having ICDs for primary prevention. During follow-up, patients were classified into cases, if device therapy was administered, and controls, if no device therapy occurred. Resting and exercise ECGs were analyzed using the enhanced adaptive matched filter method (EAMFM) to identify ECGA.</div><div>Unlike the exercise condition, the resting condition showed a statistically significant difference in PWA and QRSA between cases and controls. Thus, to classify them, rest-related ECGA features were used to feed a support vector machine (SVM), validated by a leave-one-out cross-validation algorithm. SVM yielded a sensitivity, specificity, and F1 score of 98.49%, 83.33%, and 95.61%, respectively. These results suggest that EAMFM-derived ECGA may act as a further useful feature to stratify the arrhythmia risk, overcoming the insufficient sensitivity and specificity of LVEF only. Thus, the main contribution of this study is the proposal of an additional ECGA-based criterion for identifying patients who may benefit from primary prevention ICD implantation paving the way for a conceivable revision of the current Guidelines.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"110 ","pages":"Article 108322"},"PeriodicalIF":4.9,"publicationDate":"2025-07-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144632012","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":"Tuberculosis detection on chest X-rays using two-dimensional multiscale symbolic dynamic entropy","authors":"Andreia S. Gaudêncio ,&nbsp;Miguel Carvalho ,&nbsp;Pedro G. Vaz ,&nbsp;João M. Cardoso ,&nbsp;Anne Humeau-Heurtier","doi":"10.1016/j.bspc.2025.108346","DOIUrl":"10.1016/j.bspc.2025.108346","url":null,"abstract":"<div><div>Several radiological patterns associated with pulmonary tuberculosis (TB) have been identified on chest X-rays (CXR) used for screening purposes. As a result, several automatic computational tools have emerged for this purpose. We propose a new algorithm, two-dimensional multiscale symbolic dynamic entropy (MSDE<span><math><msub><mrow></mrow><mrow><mn>2</mn><mi>D</mi></mrow></msub></math></span>), to develop a computational tool sensitive to these subtle patterns variations and noise robustness for evaluating CXR images from healthy and TB-diagnosed individuals. The one-dimensional SDE algorithm was previously shown to be more efficient in detecting amplitude variations and in computational calculations (compared to other entropy algorithms). Additionally, we also extracted first-order statistical parameters like standard deviation (SD), and mean of positive pixels (MPP), among others. These MSDE<span><math><msub><mrow></mrow><mrow><mn>2</mn><mi>D</mi></mrow></msub></math></span> and first-order texture features were used to detect TB in each lung individually. The MSDE<span><math><msub><mrow></mrow><mrow><mn>2</mn><mi>D</mi></mrow></msub></math></span> was validated using a synthetic dataset and optimized for the best set of parameters. We verified that, for both lungs, the MSDE<span><math><msub><mrow></mrow><mrow><mn>2</mn><mi>D</mi></mrow></msub></math></span> values were significantly different between healthy and TB CXR images (<span><math><mrow><mi>P</mi><mo>&lt;</mo><mn>0</mn><mo>.</mo><mn>05</mn></mrow></math></span>), and the effect size was <span><math><mo>|</mo></math></span>d<span><math><mo>|</mo></math></span> <span><math><mo>&gt;</mo></math></span>0.23. From the first-order parameters, only the mean, SD, entropy, and MPP were statistically different between both groups for the left lung (<span><math><mrow><mi>P</mi><mo>&lt;</mo><mn>0</mn><mo>.</mo><mn>05</mn></mrow></math></span>; <span><math><mo>|</mo></math></span>d<span><math><mo>|</mo></math></span> <span><math><mo>&gt;</mo></math></span>0.22). For the right lung, all first-order features significantly differentiated TB patients (<span><math><mrow><mi>P</mi><mo>&lt;</mo><mn>0</mn><mo>.</mo><mn>05</mn></mrow></math></span>; <span><math><mo>|</mo></math></span>d<span><math><mo>|</mo></math></span> <span><math><mo>&gt;</mo></math></span>0.28). Finally, we show that a multi-layer perceptron obtained 86.4 and 85.2% accuracy in detecting TB in the left and right lungs, respectively. The highest sensitivity values achieved in this study were 71.4% and 81.8% for the left and right lungs, respectively.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"111 ","pages":"Article 108346"},"PeriodicalIF":4.9,"publicationDate":"2025-07-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144614750","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":"Effective muscle synergies extraction pipeline to characterise the Box and Block Test movement","authors":"Emma Colamarino ,&nbsp;Jlenia Toppi ,&nbsp;Floriana Pichiorri ,&nbsp;Valeria de Seta ,&nbsp;Giulia Savina ,&nbsp;Ilaria Mileti ,&nbsp;Eduardo Palermo ,&nbsp;Donatella Mattia ,&nbsp;Febo Cincotti","doi":"10.1016/j.bspc.2025.108252","DOIUrl":"10.1016/j.bspc.2025.108252","url":null,"abstract":"<div><div>The Box and Block test (BBT) is a clinical test for the evaluation of the unilateral manual dexterity. To the best of our knowledge, no study has yet explored it through the muscle synergy approach. Since each analysis step impacts on the muscle synergies influencing the results interpretation, this study aims at i) optimising the procedure of muscular synergies extraction in BBT and ii) applying it to characterise BBT in healthy individuals. Electromyographic (EMG) data (8 muscles per arm) were recorded from 16 participants during the BBT performed with each upper limb, separately. Muscle synergies were extracted by means of the Non-Negative Matrix Factorization method and the impact of following two parameters was estimated: (i) cut-off frequency of low-pass filter designed to compute the envelope of the EMG signals (ii) type of normalisation to be applied to the EMG envelopes. Results show that the procedure of muscular synergies extraction requires to be optimised with particular attention to the cut-off frequency of low-pass filter designed to compute the envelope of the EMG signals and the type of normalisation to be applied to the EMG envelopes. From the physiological perspective, three synergies seem to be enough to capture main BBT movement mechanisms. The evaluation of the impact of processing steps on the muscle synergies and the characterization of the BBT will pave the way for the definition of a standard procedure to be applied also in a pathological context.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"110 ","pages":"Article 108252"},"PeriodicalIF":4.9,"publicationDate":"2025-07-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144623669","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":"Cross-subject seizure detection with vision transformer and unsupervised domain adaptation","authors":"Hailing Feng,&nbsp;Shuai Wang,&nbsp;Hongbin Lv,&nbsp;Chenxi Nie,&nbsp;Wenqian Feng,&nbsp;Hao Peng,&nbsp;Yanna Zhao","doi":"10.1016/j.bspc.2025.108341","DOIUrl":"10.1016/j.bspc.2025.108341","url":null,"abstract":"<div><div>Automatic seizure detection is of critical importance for clinical epilepsy treatment. Due to the variability of Electroencephalography (EEG) patterns across different individuals, most existing seizure detection methods fails to generalize across patients. To tackle this issue, this paper proposes a cross-subject seizure detection combines Vision Transformer (ViT) and unsupervised domain adaptation (UDA). Specifically, to enhance the generalization ability of the ViT backbone across different subjects, an adversarial network is introduced on the class token to disentangle global transferable features. Meanwhile, the multi-head attention mechanism is replaced by the transfer adaptation module (TAM) to disentangle transferable features at the patch level. Additionally, to retain the discriminative features related to epileptic seizures, a discriminative clustering module (DCM) is introduced to constrain the model. Our experiments on the CHB-MIT dataset demonstrate that the proposed method achieves strong performance in both evaluation paradigms: in epoch-based analysis it attains 89.20% accuracy, 91.05% sensitivity, and 94.54% specificity, while in event-based evaluation it maintains 89.23% sensitivity with a low false detection rate of 0.42/h. The results verify the feasibility of this method in cross-subject seizure detection.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"111 ","pages":"Article 108341"},"PeriodicalIF":4.9,"publicationDate":"2025-07-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144632592","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":"GSC-ABTA: A group-level brain sources connectivity framework based on adaptive block tensor analysis","authors":"Bashir Najafabadian ,&nbsp;Ali Motie Nasrabadi ,&nbsp;Saeid Rashidi","doi":"10.1016/j.bspc.2025.108336","DOIUrl":"10.1016/j.bspc.2025.108336","url":null,"abstract":"<div><h3>Background</h3><div>This study presents a group analysis method for identifying shared brain connectivity patterns using tensor analysis. The method’s efficacy is evaluated through a validation framework, considering various scenarios of group brain data generation and diverse control parameters.</div></div><div><h3>Methodology</h3><div>The proposed group estimation method for source-level brain connectivity begins by modeling the activity of brain and noise sources using a quasi-real six-layer head model to solve the direct problem. Pseudo-EEG data are then generated at the group level for three scenarios: Volume Conduction Effect (VC), Inter-Trial Variability (ITV), and Time Varying Connectivity (TV). The Group-Level Source Connectivity based on Adaptive Block Tensor Analysis (GSC-ABTA) is used to solve the inverse problem and estimate group-level source activity. This method allows for trial-dependent streaming updates in the group estimation of brain sources. Finally, a tensorial multivariate autoregressive model is developed in an adaptive format, taking into account a forgetting parameter for determining the contribution of observations in estimating effective brain connectivity coefficients at the group level. Statistical analysis was performed for six control parameters (including data length, signal-to-noise ratio, density, percentage of real connections added to the model, model order, and the number of trials) and compared with tensorial and non-tensorial methods in the three proposed scenarios. Additionally, the framework was validated with real data.</div></div><div><h3>Results</h3><div>The proposed method outperforms other methods in the VC scenario for all control parameters and in the ITV and TV scenarios for most control parameters. These findings underscore the importance of adaptive updating in extracting the activity of the sources for group investigation, facilitating the group extraction of brain connectivity coefficients on a more generalizable scale.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"110 ","pages":"Article 108336"},"PeriodicalIF":4.9,"publicationDate":"2025-07-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144613821","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":"Evaluating the impact of blood rheology in hemodynamic parameters by 4D Flow MRI in large vessels considering the hematocrit effect","authors":"Hernán Mella ,&nbsp;Felipe Galarce ,&nbsp;Tetsuro Sekine ,&nbsp;Julio Sotelo ,&nbsp;Ernesto Castillo","doi":"10.1016/j.bspc.2025.108145","DOIUrl":"10.1016/j.bspc.2025.108145","url":null,"abstract":"<div><div>Aortic hemodynamic parameters estimated from 4D Flow Magnetic Resonance (MR) velocity measurements are often estimated using a constant Newtonian viscosity, neglecting blood’s shear-thinning behavior. The aim of this work is to estimate and assess whether Newtonian viscosity is sufficient to quantify these parameters, given the non-Newtonian nature of blood. Additionally, we demonstrate that shear-thinning effects remain observable in large vessels despite artifacts commonly present in 4D Flow MR images.. To address this, we quantified the impact of blood rheology and hematocrit (Hct) on Wall Shear Stress (WSS), the rate of viscous Energy Loss (<span><math><msub><mrow><mover><mrow><mi>E</mi></mrow><mrow><mo>̇</mo></mrow></mover></mrow><mrow><mi>L</mi></mrow></msub></math></span>), and the Oscillatory Shear Index (OSI) based on velocity data obtained from 4D Flow MR images. Using a Hct-dependent power-law non-Newtonian model with experimentally derived rheological parameters, we analyzed these metrics across a broad range of Hct values at physiological temperatures in both in-silico and in-vivo MR datasets.</div><div>The results reveal significant differences between Newtonian and non-Newtonian models. In in-silico experiments, WSS and <span><math><msub><mrow><mover><mrow><mi>E</mi></mrow><mrow><mo>̇</mo></mrow></mover></mrow><mrow><mi>L</mi></mrow></msub></math></span> differed by up to +189% and +112% at systole, with reductions of −74% and −80% at diastole, respectively, while OSI differences ranged from −23% to −30%. For in-vivo data, WSS and <span><math><msub><mrow><mover><mrow><mi>E</mi></mrow><mrow><mo>̇</mo></mrow></mover></mrow><mrow><mi>L</mi></mrow></msub></math></span> deviations reached −44% and −60% at systole, ranging from −69% to +73% at diastole, with OSI differences averaging −21%. These findings highlights the importance of accounting for non-Newtonian blood rheology when estimating hemodynamic parameters from 4D Flow MR images in large vessels, enhancing the accuracy of cardiovascular disease assessments using in-vivo aortic data.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"111 ","pages":"Article 108145"},"PeriodicalIF":4.9,"publicationDate":"2025-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144604624","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 Knowledge-Driven Evidence Fusion Network for pancreatic tumor segmentation in CT images","authors":"Kaiqi Dong ,&nbsp;Yan Zhu ,&nbsp;Yu Tian ,&nbsp;Peijun Hu ,&nbsp;Chengkai Wu ,&nbsp;Xiang Li ,&nbsp;Tianshu Zhou ,&nbsp;Xueli Bai ,&nbsp;Tingbo Liang ,&nbsp;Jingsong Li","doi":"10.1016/j.bspc.2025.108281","DOIUrl":"10.1016/j.bspc.2025.108281","url":null,"abstract":"<div><div>Accurate pancreatic tumor segmentation remains challenging due to complex anatomical structures and diverse tumor appearances. This study presents a Knowledge-Driven Evidence Fusion Segmentation Network (KEFS-Net), a framework that systematically integrates radiological and anatomical knowledge from medical reports with imaging features to enhance segmentation accuracy. KEFS-Net consists of three key components: (1) a knowledge-driven attention network that leverages large language models, discrete information bottleneck, and cross-attention to enhance CT image segmentation performance by capturing informative features from medical reports, (2) an evidence fusion strategy based on Dempster–Shafer theory that optimizes segmentation results by evaluating the consistency between textual knowledge and image predictions, and (3) a masked learning approach that ensures robust performance in clinical scenarios with incomplete tumor descriptions. The framework was evaluated on both the Medical Segmentation Decathlon (MSD) dataset and an external clinical dataset from the First Affiliated Hospital (FAH) of Zhejiang University School of Medicine. Experimental results demonstrate superior performance compared to state-of-the-art methods, achieving Dice of 59.10% and 59.42% respectively for tumor segmentation on the MSD and external dataset. The approach shows particular strength in handling diverse tumor characteristics including size variations, boundary ambiguity, and complex anatomical locations. This knowledge-driven framework represents a significant advancement in leveraging domain knowledge through multi-modal integration for improved pancreatic tumor segmentation. Our code is available at <span><span>https://github.com/Singlesnail/KEFS-Net</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"111 ","pages":"Article 108281"},"PeriodicalIF":4.9,"publicationDate":"2025-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144614239","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}