Biomedical Signal Processing and Control最新文献

{"title":"Enhancing Mobile Edge Computing customer reviews analysis with Ensemble Sparse Support Vector L1 Regularization Based Crossover Discrete Mycorrhized Algorithm","authors":"Vinodh Kumar S. ,&nbsp;Bharath Babu S. ,&nbsp;J. Vellingiri ,&nbsp;D. Roja Ramani","doi":"10.1016/j.bspc.2025.108019","DOIUrl":"10.1016/j.bspc.2025.108019","url":null,"abstract":"<div><div>Mobile Edge Computing has emerged as a transformative technology, enhancing the efficiency of internet community platforms by enabling real-time data processing and analysis at the network’s edge. These platforms facilitate the exchange of user opinions and ideas, providing valuable insights into user attitudes and preferences. Despite advancements in Mobile Edge Computing, challenges persist in effectively categorizing customer reviews due to latency, data scarcity, and overfitting issues in computational models. This study integrates natural language processing techniques with edge computing infrastructure to analyze reviews closer to their source, thereby minimizing latency and improving overall performance. To enhance the detection accuracy in Mobile Edge customer reviews, this paper proposes an innovative approach called the Ensemble Sparse Support Vector L1 Regularization-based Crossover Discrete Mycorrhized Algorithm. Pre-processing steps, such as tokenization, lemmatization, and stemming, are employed to improve data quality. Feature extraction is performed using a stacked autoencoder, which incorporates multiple layers to address data scarcity issues. To optimize the performance of the Sparse Support Vector L1 Regularization, a novel Crossover Discrete Mycorrhized Optimization Algorithm is introduced, mitigating overfitting and improving classification accuracy. The proposed approach is validated using datasets such as the Mobile Recommendation System Dataset, Mobile Positioning Dataset, Mobile Edge Distance Analysis Dataset, International Phone Checker API, and Financial Fraud Detection Dataset. Experimental results demonstrate superior performance, achieving 98.52% accuracy, 98.39% precision, 98.28% recall, and 98.21% F1-score in aspect category detection. The proposed method addresses critical gaps in latency reduction and data processing accuracy in MEC environments by significantly improving reliability and efficiency in customer review analysis. These findings contribute to more robust, customer-centric Mobile Edge Computing systems, fostering enhanced real-time decision-making and user experience.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"109 ","pages":"Article 108019"},"PeriodicalIF":4.9,"publicationDate":"2025-05-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144071745","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":"Multi-source domain separation adversarial domain adaptation for EEG emotion recognition","authors":"Qingsong Ai ,&nbsp;Chenhuan Wang ,&nbsp;Kun Chen ,&nbsp;Li Ma","doi":"10.1016/j.bspc.2025.108016","DOIUrl":"10.1016/j.bspc.2025.108016","url":null,"abstract":"<div><h3>Background and Objective</h3><div>: Advancements in deep learning have propelled emotion-based brain-computer interfaces (BCI) forward, particularly in medical and rehabilitation applications. Within these BCI systems, electroencephalogram (EEG) serves as a critical modality for monitoring brain activity, playing a pivotal role in the recognition of emotional states. However, individual EEG variances pose challenges to the development of robust emotion recognition models. This research addresses these challenges by leveraging unsupervised domain adaptation, which typically suffers from disrupted shared feature spaces when aggregating multi-subject EEG data.</div></div><div><h3>Methods:</h3><div>In this paper, we propose to retain individual EEG data separately as a source and match it with target for domain adaptation. Meanwhile, multiple independent feature domain matchers are constructed to assist the network in finding domain-invariant features through adversarial learning with discriminative network. Additionally, a self-supervised pseudo-labeling strategy is incorporated to enhance fine-grained emotion classification and to improve confidence levels.</div></div><div><h3>Results:</h3><div>The experimental results demonstrate that, under cross-subject paradigm, the proposed method achieves an average accuracy of 91.68% on SEED dataset and 76.70% on SEED-IV dataset. Additionally, on DREAMER dataset, the binary classification accuracies for Dominance and Arousal are 82.62% and 82.72%. Under cross-session paradigm, the model attains accuracies of 95.42% and 81.89% on SEED and SEED-IV datasets, respectively, outperforming comparative domain adaptation methods and existing models.</div></div><div><h3>Conclusion:</h3><div>The proposed approach effectively mitigates the impact of individual EEG variances and significantly enhances the robustness of emotion recognition models. These advances highlight the potential for more personalized and effective interventions in healthcare and rehabilitation settings, providing new scenarios for future research and applications of emotion-based BCI systems.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"109 ","pages":"Article 108016"},"PeriodicalIF":4.9,"publicationDate":"2025-05-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144071741","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":"9IEC: A novel method for exposer region determination in low contrast and nonuniform illumination chest X-ray imaging","authors":"Shivam Gangwar ,&nbsp;Reeta Devi ,&nbsp;Nor Ashidi Mat Isa","doi":"10.1016/j.bspc.2025.107988","DOIUrl":"10.1016/j.bspc.2025.107988","url":null,"abstract":"<div><div>Accurate chest X-ray (CXR) image interpretation is crucial for diagnosing numerous diseases. However, CXRs often suffer from nonuniform illumination and low contrast, leading to misclassification of exposure regions, which affects diagnostic accuracy. Existing methods rely on simplistic intensity-based classification, which results in errors. To address this, we propose the 9IEC algorithm, which introduces a novel integration of intensity, entropy, and contrast to define nine subregions and improve exposure region determination. This approach enables precise image enhancement, leading to superior visual interpretation and improved diagnostic reliability. Extensive qualitative evaluations, including expert surveys, demonstrate that 9IEC outperforms state-of-the-art methods and extends its utility beyond medical imaging.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"109 ","pages":"Article 107988"},"PeriodicalIF":4.9,"publicationDate":"2025-05-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144071743","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 novel biophysical ERG model integrating rod and bipolar cells cascade transduction with JLCSG MOSFET","authors":"Qing-an Ding ,&nbsp;Fangfang Ning ,&nbsp;Yuhua Gao ,&nbsp;Jianyu Li ,&nbsp;Chunyan Liu ,&nbsp;Xiaoyuan Li ,&nbsp;Binghui Hou ,&nbsp;Bing Chen ,&nbsp;Yandong Peng","doi":"10.1016/j.bspc.2025.108010","DOIUrl":"10.1016/j.bspc.2025.108010","url":null,"abstract":"<div><div>Electroretinogram (ERG) Modeling is commonly used in clinical diagnosis and waveform prediction, yet traditional models neglect the impact of protein-regulated gated channels in rod and bipolar cells. A novel ERG biophysical model is proposed to characterize the responses of rod and bipolar cells to flash stimulation. This model is based on the equivalent of junctionless cylindrical surrounding gate (JLCSG) MOSFETs to cyclic nucleotide-gated (CNG) and transient receptor potential cation channel subfamily M member 1 (TRPM1) channels. An improved ERG architecture, the component of this biophysical model, considers the cascade superposition of rod and bipolar cell transduction to enhance the fitting accuracy at higher flash intensity. And by equating the concentration of cascade messengers or transducer proteins to the gate voltage of a JLCSG MOSFET, the generated drain current successfully fits the ERG at flash intensity ranging from 50 to 400 Rh*/rod, with an accuracy of approximately 96.86%. Moreover, the parameter changes caused by retinitis pigmentosa (RP) are elucidated through fitting of model and patient ERGs. The biophysical ERG model breaks the limitations of traditional mathematical models, demonstrating the feasibility of equating biological mechanisms to physical devices.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"109 ","pages":"Article 108010"},"PeriodicalIF":4.9,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144068024","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":"KFCNet: A Key Feature Consistency Network for microscopic urinary sediment image classification","authors":"Shaoqi Wu ,&nbsp;Ge Song ,&nbsp;Mengying Lou ,&nbsp;Shian Wang ,&nbsp;Xuan Chen ,&nbsp;Runge Li ,&nbsp;Minru Zhai ,&nbsp;Hao Chen ,&nbsp;Likangxin Gao ,&nbsp;Feiran Gao ,&nbsp;Linfeng Cong ,&nbsp;Peng Wang","doi":"10.1016/j.bspc.2025.108009","DOIUrl":"10.1016/j.bspc.2025.108009","url":null,"abstract":"<div><div>Diagnosing renal and urinary system disorders heavily relies on the examination of urine sediment microscopic images. Nonetheless, the heterogeneity of urine constituents and the intricacy of microscopic imagery render conventional manual classification techniques prone to significant subjectivity and diminished accuracy. Deep learning techniques provide an efficient method for classifying urine sediment; yet, current approaches encounter difficulties in addressing the complexity and variety of urinary sediment images, especially in modeling global information. Overdependence on background information can result in erroneous model classification. We propose a Key Feature Consistency Network (KFCNet) to resolve these concerns. KFCNet reduces interference and improves attention on essential variables by completely utilizing characteristics acquired at various network stages, therefore enhancing the accuracy of urine sediment classification. Our network consists of three primary modules: a Local Feature Selection Module (LFS Module), an Interference Feature Learning Module (IFL Module), and a Semantic Enhancement Module (SE Module). Through the process of randomly cropping images and picking critical features, the Local Feature Selection Module enhances the model’s capacity to collect subtle discriminative information. The Interference Feature Learning Module leverages the abundant information in low-level features to acquire and filter interference features via an attention method, hence improving classification robustness. The Semantic Enhancement Module eliminates interference from mid-level features and amplifies critical feature responses, allowing the model to precisely identify urine sediment among intricate backgrounds. Experimental findings indicate that KFCNet excels in urinary sediment classification tasks and markedly surpasses current methodologies across three public datasets. This study presents innovative findings for the automated classification of urine sediment pictures, enhancing the reliability of technical support for clinical diagnosis.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"109 ","pages":"Article 108009"},"PeriodicalIF":4.9,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144071739","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":"Efficient multiscale weighted features based residual ConvLSTM method to detect Parkinson’s disease using electroencephalogram data","authors":"K. Rajesh Kumar,&nbsp;T.R. Ganesh Babu","doi":"10.1016/j.bspc.2025.108057","DOIUrl":"10.1016/j.bspc.2025.108057","url":null,"abstract":"<div><div>Parkinson’s Disease (PD) is a neurological disease that affects the psychological and neural systems. Various factors, including age, medications, and disease state, can affect the Electroencephalogram (EEG) signal. It becomes difficult to establish common features to identify PD. So, this research developed a powerful PD detection model using deep learning to overcome such challenges. Initially, data is taken from different sources. Here, the wave features, temporal features, spatial features, spectral features, and deep features are extracted from the collected data, where the deep features is extracted using the Autoencoder (AE). Then, the extracted features are fed into the Multiscale Weighted Features-based Residual Convolutional Long Short Term Memory (MWF-RconvLSTM). The multiscale weighted features incorporated in the developed model can effectively solve the complexity issue while detecting the disease. Further, convolutional LSTM in the proposed model significantly enhances the model’s ability to understand complex features in the detection of PD. Here, the weights are optimized using the developed Enhanced Peafowl Optimization Algorithm (EPOA). Weight optimization using the developed EPOA can enhance the effectiveness of PD detection. Moreover, the developed model is evaluated with various models to display the effective performance in detecting PD. Finally, the developed EPOA-MWF-RconvLSTM model offers the best result in terms of accuracy is 94.97. Moreover, the conventional model like DMO-MWF-RconvLSTM, BFGO-MWF-RconvLSTM, RHA-MWF-RconvLSTM, and POA-MWF-RconvLSTM achieved the accuracy to be 80.02, 88.49, 82.01, and 90.87. This confirmed that the recommended model is more successful in the detection of PD than other existing models.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"109 ","pages":"Article 108057"},"PeriodicalIF":4.9,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144071738","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":"An echo state network for synthesizing the standard 12-lead ECG from a two-lead ECG obtained from a single touch of a wrist-worn device","authors":"Karolina Jančiulevičiūtė ,&nbsp;Daivaras Sokas ,&nbsp;Saulius Daukantas ,&nbsp;Leif Sörnmo ,&nbsp;Andrius Petrėnas","doi":"10.1016/j.bspc.2025.108008","DOIUrl":"10.1016/j.bspc.2025.108008","url":null,"abstract":"<div><h3>Background:</h3><div>With the availability of a wrist-worn device capable of acquiring two ECG leads with a single touch, synthesis of the 12-lead ECG may be accomplished to facilitate clinical interpretation.</div></div><div><h3>Objective:</h3><div>This study proposes an echo state network (ESN) for synthesizing the 12-lead ECG from two leads simultaneously acquired using a wrist-worn device.</div></div><div><h3>Methods:</h3><div>The wrist-worn device, equipped with three electrodes, was used to acquire two ECG leads from 51 healthy participants, 29 patients with acute myocardial infarction, and 12 patients with other cardiovascular diseases. The person-specific synthesis is based on the ESN, a recurrent neural network, trained on a single resting, standard 12-lead ECG through a highly efficient training process. To explore the importance of different electrode touch sites, the participants were instructed to touch sites on the body corresponding to the electrode positions for acquiring the precordial leads V3 and V5, as well as the abdomen.</div></div><div><h3>Results:</h3><div>Using the ESN, the lowest RMS error between the standard and the synthesized ECGs is obtained for leads I and V1, irrespective of participant group and touch site from which the two-lead ECG was acquired. The ESN outperformed a linear regression-based transformation matrix, especially for the precordial leads where the RMS error was up to three times higher than that of the ESN.</div></div><div><h3>Conclusion:</h3><div>ESN-based synthesis of the 12-lead ECG based on a two-lead ECG holds promise as a valuable tool for screening abnormalities in the ECG.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"109 ","pages":"Article 108008"},"PeriodicalIF":4.9,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144071740","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 AI in Healthcare: A Three-Layer Defense to Mitigate Adversarial Noise Impact in Radiology Imaging","authors":"Sheikh Burhan Ul Haque ,&nbsp;Aasim Zafar ,&nbsp;Sheikh Moeen ul haque ,&nbsp;Sheikh Riyaz ul Haq ,&nbsp;Mohassin Ahmad","doi":"10.1016/j.bspc.2025.107969","DOIUrl":"10.1016/j.bspc.2025.107969","url":null,"abstract":"<div><div>Early detection of lung nodules through CT imaging is crucial for timely treatment and improved patient outcomes. Artificial intelligence (AI), particularly deep learning (DL), has shown exceptional promise, often surpassing human expertise in diagnosing lung cancer. However, the vulnerability of DL models to adversarial noise—imperceptible perturbations designed to mislead models—remains underexplored in medical imaging. To the best of our knowledge, this is the first study to comprehensively analyze the effects of targeted and untargeted adversarial noise on DL-based medical diagnosis models. Additionally, we propose a novel three-tier defense strategy to mitigate these adversarial impacts on radiology images. The proposed approach combines modified adversarial training (MAT) during the training phase with Total Variation Minimization (TVM) flowed by bit-plane slicing (BPS) at the testing phase, ensuring robust performance against adversarial attacks in all the phases. MAT strengthens model resilience by exposing it to adversarial examples with varying epsilon values, improving its ability to counter diverse perturbations. At inference, TVM reduces high-frequency adversarial noise while preserving essential image structures, and BPS further enhances robustness by extracting critical features and discarding less significant details prone to adversarial manipulation. A lung nodule classification model was developed using transfer learning with DenseNet-121, trained on the publicly available LIDC-IDRI dataset. The model achieved 95.71% training accuracy and 93.17% testing accuracy on clean images. However, when exposed to adversarial attacks such as Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD), accuracy dropped significantly to 13.74% under FGSM and 1.32% under PGD. The proposed defense strategy successfully restored performance, achieving an average accuracy of approximately 93% against both FGSM and PGD attacks. These results demonstrate that the defense approach effectively mitigates adversarial noise across both training and testing phases, improving the reliability of DL models in medical image analysis. By enhancing robustness in lung cancer detection, this study contributes to the advancement of AI-driven healthcare, ensuring safer and more trustworthy diagnostic systems.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"109 ","pages":"Article 107969"},"PeriodicalIF":4.9,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144071502","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":"An innovative integrated processing framework for enhanced decoding of Mental-Imagery MEG signals","authors":"Fan Wang ,&nbsp;Peng Wang ,&nbsp;Peng Ding ,&nbsp;Anmin Gong ,&nbsp;Yanxiao Chen ,&nbsp;Yuhang Xue ,&nbsp;Yunfa Fu","doi":"10.1016/j.bspc.2025.108068","DOIUrl":"10.1016/j.bspc.2025.108068","url":null,"abstract":"<div><div>Magnetoencephalography (MEG) has emerged as a pivotal tool for developing brain-computer interfaces (BCI) based on mental imagery tasks, owing to its superior spatiotemporal resolution. However, existing decoding strategies for MEG data in mental imagery tasks face two major challenges: (1) inadequate exploitation of the high-dimensional characteristics of MEG signals, and (2) a lack of tailored algorithms for efficient feature extraction and classification. To address these issues, we propose a novel integrated processing framework for enhancing mental imagery-based BCI performance. First, an innovative channel selection method combining correlation coefficient and variance entropy product (CC-VEP) is introduced to identify task-relevant channels while suppressing noise and redundancy. Next, discriminative features are extracted using the DivCSP algorithm with intra-class regularization terms. Subsequently, probabilistic outputs from base classifiers (K-Nearest Neighbors, Random Forests, Support Vector Machines, and Extreme Learning Machines) are fused through a modified multi-criteria decision-based fusion (MCDM-MCF) strategy to generate final class labels. Experimental results on a public MEG dataset demonstrate that the proposed framework achieves a 12.25% improvement in classification accuracy compared to the average accuracy of all base classifiers in traditional methods, and a 7.97% improvement over other existing methods. Moreover, it effectively balances specificity and sensitivity, enhances robustness, and offers a new approach for efficient decoding in mental imagery-based BCI systems.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"109 ","pages":"Article 108068"},"PeriodicalIF":4.9,"publicationDate":"2025-05-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144068156","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":"Harnessing deep learning for biomarker identification in osteoarthritis pathological sections","authors":"Weiao Ying ,&nbsp;Yi Du ,&nbsp;Qing Wang ,&nbsp;Juan Zheng ,&nbsp;Zhouliang Yang","doi":"10.1016/j.bspc.2025.108001","DOIUrl":"10.1016/j.bspc.2025.108001","url":null,"abstract":"<div><div>Osteoarthritis (OA) is a degenerative joint disease in which accurate identification of treatment-related biomarkers and fibrosis biomarkers in pathological sections is essential for evaluating therapeutic efficacy. This study assessed several deep learning algorithms—U-Net (Universal Network), convolutional neural networks (CNNs), residual networks (ResNets), and densely connected convolutional networks (DenseNets)—for their ability to automatically analyze immunohistochemically stained pathological images from OA patients. Among these models, U-Net achieved the highest segmentation accuracy for both biomarkers and fibrosis markers, significantly improving recognition efficiency. Using U-Net-based image analysis, a comparative evaluation of treatment modalities indicated that arthroscopic surgery most effectively reduced inflammation-related factors, including matrix metalloproteinases (MMPs), tumor necrosis factor-alpha (TNF-α), and interleukin-1 beta (IL-1β). A progressive decline in these markers was observed at 1, 3, and 6 months post-treatment. These results offer critical evidence of therapeutic efficacy and support the integration of artificial intelligence (AI)-assisted pathological analysis into OA treatment pathways to inform precise clinical decision-making. Future studies should investigate he integration of U-Net with other deep learning models to further improve biomarker identification and optimize OA management.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"109 ","pages":"Article 108001"},"PeriodicalIF":4.9,"publicationDate":"2025-05-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144068157","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}