Frontiers in medical technology最新文献

{"title":"Quantum-edge synergy for dynamic acoustic compensation in ultrasound safety: a narrative review of tissue heterogeneity challenges.","authors":"Bin Hu, Wei Shi, Shihong Xiong, Songlin Li, Kelun Wang, Na Gong","doi":"10.3389/fmedt.2026.1775679","DOIUrl":"10.3389/fmedt.2026.1775679","url":null,"abstract":"<p><strong>Background: </strong>Conventional ultrasound safety monitoring relies on the Mechanical Index (MI) and Thermal Index (TI), which lack tissue-specificity. Consequently, real-time safety profiling remains imprecise. Dynamic, tissue-specific corrections such as tissue perfusion rate (TPR) aim to address this limitation. This narrative review examines how integrating these corrections with quantum-cloud computational frameworks may enable enhanced safety and image quality in ultrasonic applications.</p><p><strong>Objective: </strong>We assessed the potential of an integrated quantum-cloud framework to reduce ultrasound-related bioeffects without compromising image quality, compared to standard MI/TI monitoring in patients undergoing clinical ultrasound imaging.</p><p><strong>Methods: </strong>A narrative synthesis of the eligible literature was performed, guided by thematic analysis frameworks. Quantitative outcomes were descriptively summarized with medians and ranges; formal meta-analysis was not conducted because heterogeneity in TPR measurement protocols, device manufacturers, and outcome definitions violated the commutability assumption required for statistical pooling.</p><p><strong>Results: </strong>Qualitative synthesis indicated that tissue-specific, multi-parameter correction strategies were consistently associated with more conservative acoustic exposure and heightened safety awareness than static MI/TI monitoring alone, although the consistency and clinical magnitude of these benefits varied across anatomical applications and study designs. The certainty of the current evidence was judged low to moderate, limited by non-standardized TPR measurement protocols and heterogeneous outcome definitions.</p><p><strong>Conclusions: </strong>Together, these data suggest that multi-parameter acoustic correction coupled with cloud computing holds promise for improving ultrasound safety profiling and imaging precision. However, the current low-to-moderate certainty of evidence, due to methodological heterogeneity, necessitates future validation of edge-cloud synergies to realize robust, personalized monitoring systems.</p>","PeriodicalId":94015,"journal":{"name":"Frontiers in medical technology","volume":"8 ","pages":"1775679"},"PeriodicalIF":4.6,"publicationDate":"2026-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13581935/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148993110","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An evaluation of pathology foundation models with weakly supervised learning for BRAF mutation prediction in melanoma whole slide images.","authors":"Lucy Godson, Jack Breen, Alexander I Wright, Derek Magee, Daljeet Bansal, Darren Treanor, Emily L Clarke","doi":"10.3389/fmedt.2026.1835452","DOIUrl":"10.3389/fmedt.2026.1835452","url":null,"abstract":"<p><strong>Introduction: </strong>Melanoma is the most aggressive form of skin cancer, with around 40-50% of cases harbouring BRAF mutations that can be effectively targeted to improve patient outcomes. Typically, BRAF mutations are identified through DNA sequencing, which can be costly and contribute significantly to turn-around-times. Predicting BRAF status directly from haematoxylin and eosin (H&E) whole-slide images (WSIs) could offer a faster and more accessible alternative within digital pathology workflows.</p><p><strong>Methods: </strong>In this study, we evaluated six pathology foundation models for WSI-based BRAF mutation prediction using an attention-based multiple instance learning (ABMIL) framework. Training and testing models with five-fold cross-validation, using 1,109 WSIs from 745 patients and performing external validation with TCGA and VisioMel cohorts.</p><p><strong>Results: </strong>Models which utilised H-Optimus-1 features achieved the highest AUROC values on the internal test set (0.80 (0.72, 0.88)) and VisioMel (0.79 (0.75, 0.83) external test set, while UNI2-h features demonstrated superior sensitivity and higher TCGA performance (AUROC: 0.76 (0.66, 0.86)). Subgroup analyses indicated that model performance was impacted by histopathological features such as ulceration and tissue area, while t-SNE plots showed that high-attention foundation model features tended to cluster based on the dataset the features were from, rather than biological signal.</p><p><strong>Discussion: </strong>Overall, foundation models combined with weakly supervised learning provide competitive performance for BRAF status prediction in melanoma WSIs, but our findings highlight the need for diverse melanoma datasets and the importance of evaluating model performance beyond a single metric when developing H&E-based computational biomarkers.</p>","PeriodicalId":94015,"journal":{"name":"Frontiers in medical technology","volume":"8 ","pages":"1835452"},"PeriodicalIF":4.6,"publicationDate":"2026-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13582659/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148993083","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Systematic review of 3D-printed surgical devices for austere environments: an evaluation guideline.","authors":"Kyung-Hoon Moon, Matthew Cant, Mehdi Saeidi, Arul Ramasamy, Anthony M J Bull","doi":"10.3389/fmedt.2026.1908227","DOIUrl":"https://doi.org/10.3389/fmedt.2026.1908227","url":null,"abstract":"<p><strong>Background: </strong>Three-dimensional (3D) printing is a common manufacturing technique. It is being adopted to produce essential surgical equipment in austere settings around the world. However, the evaluation of these products remains inconsistent. This raises safety concerns and limits translation to clinical use. To address this, this systematic review aims to identify 3D-printed surgical devices intended for austere environments and examine their evaluation methodologies. The findings will inform the development of an evaluation framework that supports the design of patient-safe devices.</p><p><strong>Methods: </strong>Medical and engineering bibliographic databases were searched to identify literature describing 3D-printed surgical devices for use in austere environments. Two independent blinded reviewers screened articles in two stages using abstracts and full texts. The review was conducted in accordance with PRISMA guidelines. In line with the EU Medical Device Regulation, device evaluations were categorised into manufacturing specifications, mechanical performance, sterilisation, biocompatibility, and usability. Reported methods of device evaluation were extracted and compared against the relevant published standards.</p><p><strong>Results: </strong>15 studies met the inclusion criteria. The austere environments included low-middle income countries, military settings, and space exploration. 13 out of 15 studies used Fused Deposition Modeling printers. All materials were thermoplastic. Printed equipment ranged from external fixators to surgical sets. The printing details were well reported. However, there was inconsistency in testing methods, resulting in data that is incomparable across the studies. 2 of 15 papers conducted mechanical testing for external fixators using American Society for Testing and Materials standards but used different parameters. 2 studies used self-developed mechanical testing protocols. 7 of 15 papers reported usability analysis using various methodologies. 3 of 15 papers described testing for sterilisation of the equipment, but only one paper evaluated its effect on the device. There was no biocompatibility testing.</p><p><strong>Conclusion: </strong>Despite the potential implications for patient safety, all 15 studies showed considerable variability in device evaluation methods. To address this, this review recommends adopting established standards from ISO, IEC and ASTM International. The proposed framework covers the key domains: manufacturing specifications, mechanical performance, sterilisation, biocompatibility and usability. Implementation of this structured approach may enhance standardisation and support the development of safer devices.</p>","PeriodicalId":94015,"journal":{"name":"Frontiers in medical technology","volume":"8 ","pages":"1908227"},"PeriodicalIF":4.6,"publicationDate":"2026-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13579019/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148993141","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Overview of research and practice progress in body composition measurement techniques.","authors":"Wenjuan Jing, Wenrong Jing, Shili Luo, Yinhong Chen, Qinghuan Guo, Juan Yang","doi":"10.3389/fmedt.2026.1789046","DOIUrl":"10.3389/fmedt.2026.1789046","url":null,"abstract":"<p><p>The measurement of body composition is one of the most important tools for evaluating an individual's health and nutritional status. Its field has developed rapidly in recent years due to technological advances and growing application needs. Currently, widely used measurement methods include dual-energy x-ray absorptiometry (DXA), bioelectrical impedance analysis (BIA), Magnetic Resonance Imaging (MRI), Computed Tomography (CT), and smartphone software based on Computer Vision (CV) and Artificial Intelligence (AI). Each method has its own advantages in terms of accuracy and convenience. However, these techniques still face problems related to measurement reliability and applicability across different populations. The purpose of this paper is to provide a comprehensive overview of the technologies, target populations, and research fields related to body composition measurement. Body composition measurement is an important part of health management, non-communicable disease management, clinical medicine, and sports and physical training. Along with new scientific advances, this article also presents potential future research directions in body composition measurement.</p>","PeriodicalId":94015,"journal":{"name":"Frontiers in medical technology","volume":"8 ","pages":"1789046"},"PeriodicalIF":4.6,"publicationDate":"2026-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13563127/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148957737","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancing blood transfusion safety through RFID-based verification and human factors engineering.","authors":"Alessandra Angelucci, Fabiana Zummo, Riccardo Giudici, Roberto Fumagalli, Andrea Aliverti, Gabriele Bassi","doi":"10.3389/fmedt.2026.1899583","DOIUrl":"10.3389/fmedt.2026.1899583","url":null,"abstract":"<p><strong>Introduction: </strong>Blood transfusion is a high-risk process, with bedside verification being the last opportunity to intercept potentially fatal errors. This study presents the design, development, and evaluation of an RFID-based system for bedside transfusion verification, grounded in Human Factors Engineering principles.</p><p><strong>Methods: </strong>The system employs passive UHF RFID tags on patients, blood units, and healthcare staff to perform continuous silent verification. When all the needed elements for transfusion (patient, matching blood unit, two healthcare operators) are detected by an active RFID reader and no other possible sources of error are present (e.g., another blood unit or patient nearby), the two operators can proceed with a digital checklist. Completion of the digital checklist by the two operators is enforced before transfusion starts. The system was evaluated in simulated scenarios through a series of technical and procedural tests. The development process was informed by iterative feedback from clinical and technical stakeholders, including nurses, critical care physicians specialized in transfusion medicine, and biomedical engineers.</p><p><strong>Results: </strong>Low detection latency (<300 ms), strong Bluetooth Low Energy stability, sufficient power autonomy (>8 h), and consistent alert behavior under error scenarios were confirmed. Environmental interference from liquids and metal reduced tag readability beyond 130 cm, highlighting positioning and presence of obstacles as a deployment consideration.</p><p><strong>Discussion: </strong>A limitation of the system is that it does not close the loop, i.e., there is no flow sensor that confirms if infusion is taking place and lasts as long as expected. Still, the system presents a significant advantage with respect to existing checklists and barcode-based systems, as it performs passive checks and is silent when no source of error is identified. The proposed RFID-based workflow showed technical feasibility and correct functional behavior in predefined simulated transfusion scenarios, supporting its potential as an additional safety layer to be evaluated in future clinical studies. No real-world clinical testing was performed in this study; therefore, the present work should be interpreted as a pre-clinical proof-of-concept. Future studies should evaluate the system in actual transfusion settings and compare it with current standard verification practices.</p>","PeriodicalId":94015,"journal":{"name":"Frontiers in medical technology","volume":"8 ","pages":"1899583"},"PeriodicalIF":4.6,"publicationDate":"2026-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13558210/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148932255","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Translational advances in diabetic nephropathy management: integrating nanoformulation-based drug delivery systems.","authors":"Mohd Danish Ansari, Ashish Vishwakarma, Md Quamuddin, Mansi Aggarwal, Hashim Khan, Muhammad Fauzi Abd Jalil, Syed Mahmood, Adil Farooq Wali, Sirajunisa Talath, Mohamed El-Tanani","doi":"10.3389/fmedt.2026.1951734","DOIUrl":"10.3389/fmedt.2026.1951734","url":null,"abstract":"<p><p>The incidence of diabetes mellitus (DM) is increasing worldwide, especially for its major renal complications, diabetic nephropathy (DN) and chronic kidney disease (CKD). These are caused by chronic hyperglycemia, oxidative stress, hemodynamics, inflammation, and profibrotic signaling, which result in progressive renal structural damage and functional decline. All these mechanisms contribute to podocyte injury, thickening of the glomerular basement membrane, interstitial fibrosis, and progressive loss of renal function. While traditional diagnostic markers like albuminuria or estimated glomerular filtration rate (eGFR) aid in tracking disease with imaging, and emerging markers like neutrophil gelatinase-associated lipocalin (NGAL), kidney injury molecule-1 (KIM-1), and cystatin C can help, none can detect early, subclinical kidney damage. A comprehensive literature review was conducted in PubMed, Google Scholar, Scopus, Web of Science, and ScienceDirect using keywords related to DN pathophysiology, diagnosis, nanomedicine, renal targeting, and nanoformulations, with publications between 2015 and 2026. Intensive glycemic and blood pressure control, inhibition of the renin-angiotensin-aldosterone system, and lipid-lowering therapy are currently used as therapeutic interventions. Poor bioavailability, off-target toxicity, complex dosing regimens, and variable renal drug handling are often responsible for their limited effectiveness. One approach to circumventing these limitations is with nanotechnology-based drug delivery systems. In preclinical studies, the antioxidant, anti-inflammatory, and antifibrotic effects have consistently been reported to be superior, with reduced systemic toxicity, while clinical translation remains in its early stages. This review discusses current knowledge on DN and critically evaluates the pros, cons, and future of nanotechnology-based approaches for delivering targeted nanomedicines to the kidney.</p>","PeriodicalId":94015,"journal":{"name":"Frontiers in medical technology","volume":"8 ","pages":"1951734"},"PeriodicalIF":4.6,"publicationDate":"2026-08-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13551425/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148914822","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"MemAF-KTCBM: a Memory-Aware Fractional optimisation-enabled deep learning model for atherosclerosis classification.","authors":"Nazarkar Pravalika, Jabeena A, Vetriveeran Rajamani","doi":"10.3389/fmedt.2026.1901028","DOIUrl":"10.3389/fmedt.2026.1901028","url":null,"abstract":"<p><p>Atherosclerosis, a chronic vascular disorder caused by the abnormal functioning of blood vessels, remains a major contributor to cardiovascular diseases. Although recent advances in machine learning and deep learning have improved atherosclerosis classification, several critical challenges persist, particularly class imbalance, high computational complexity, limited interpretability, and poor generalisation, which constrain their clinical utility. To address these limitations, this paper proposes a novel Memory-Aware Fractional optimisation-assisted Knowledge-distilled mutual-conditioned dynamic Transformer-enabled Convolutional neural network-bidirectional long short-term Memory (MemAF-KTCBM) framework that employs knowledge distillation and mutual information fusion to reduce computational complexity while preserving classification performance. Initially, coronary artery structures are segmented using a memory-aware fractional optimisation-enabled Modified Pyramid Scene Parsing Network to improve classification reliability. Furthermore, a Memory-Aware Fractional optimisation-enabled Style Generative Adversarial Network is employed to generate high-quality synthetic angiographic images, thereby mitigating class imbalance and reducing prediction bias. In addition, pretrained local descriptors are incorporated to enhance feature extraction, interpretability, and classification reliability. The proposed framework is evaluated using the ARCADE and CADICA datasets using multiple training percentages and 10-fold cross-validation. On the ARCADE dataset, the MemAF-KTCBM model achieved an accuracy of 97.53%, a precision of 97.82%, a recall of 96.94%, an NPV of 0.97, a PPV of 0.98, and an MCC of 0.95. Similarly, on the CADICA dataset, it attained an accuracy of 96.65%, a precision of 96.94%, a recall of 96.07%, an NPV of 0.97, a PPV of 0.97, and an MCC of 0.94. These findings confirm that the proposed framework acts as a robust and effective solution for atherosclerosis classification across heterogeneous atherosclerosis datasets.</p>","PeriodicalId":94015,"journal":{"name":"Frontiers in medical technology","volume":"8 ","pages":"1901028"},"PeriodicalIF":4.6,"publicationDate":"2026-08-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13550192/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148914860","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Severity-aware hierarchical federated learning for privacy-preserving indoor emergency recognition in BLE-based medical IoT environments.","authors":"Chembon Rajeendran Karthik, Shyam Venkatraman, Hemalatha K","doi":"10.3389/fmedt.2026.1900013","DOIUrl":"10.3389/fmedt.2026.1900013","url":null,"abstract":"<p><p>Indoor safety in GPS-denied environments such as hospitals remains critically hindered by the lack of data-local, context-aware emergency localization systems. This paper presents a complete three-layer IoT framework, namely Medical Indoor BLE Emergency System (MIBES), integrating BLE 4.2 and Federated Learning (FL) for real-time emergency detection, localization and severity weighted aggregation of manually triggered emergency alerts. Built upon an ESP32-based hardware architecture, the system employs BLE Safety Tags with long-press emergency triggers and Anchor nodes enhanced with Exponential Moving Average (EMA) smoothing and utilizing the User Datagram Protocol (UDP) for data encapsulation and transport. The key novelty of this work lies in the proposed Transformer-based Severity-Aware Hierarchical Federated Learning (SA-HFL) framework. This system jointly addresses class imbalance and statistical data heterogeneity in emergency localization through the integration of Focal Loss. This design lets important emergency classes get priority without ruining the overall model generalisation. The Transformer model achieves an accuracy of 99.52%, recall of 99.08% and a MCC of 0.9904 when evaluated on a real-world BLE dataset via 5-fold cross-validation and numerically outperforms the standard FedAvg and FedProx model baselines with performance gains that is supported by moderate to large effect sizes though not by statistically significant paired tests. Communication efficiency remains high with only 150,849 parameters exchanged in each aggregation round. Consequently, the results validate the proposed architecture as a scalable, privacy-preserving solution for reliable emergency alert localization and prioritization across healthcare critical infrastructure.</p>","PeriodicalId":94015,"journal":{"name":"Frontiers in medical technology","volume":"8 ","pages":"1900013"},"PeriodicalIF":4.6,"publicationDate":"2026-08-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13550126/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148914875","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Bionic vision technologies: progress and perspectives on retinal prostheses and optogenetics for the treatment of advanced retinal degeneration.","authors":"Na Long, Meng Si, Yin Jiang, Fan Chen, Bei Chen, Fei Jiang, Shangang Pan, Xueqin Chu, Huarong Wu","doi":"10.3389/fmedt.2026.1898982","DOIUrl":"10.3389/fmedt.2026.1898982","url":null,"abstract":"<p><p>Advanced degenerative retinal diseases [e.g., retinitis pigmentosa (RP) and age-related macular degeneration (AMD)] are characterized by irreversible photoreceptor loss, and conventional interventions rarely prevent progression to profound vision loss at advanced stages. Retinal prostheses and optogenetic therapies are two leading bionic strategies to bypass absent photoreceptors, yet both are limited by a shared \"encoding-biointerface\" bottleneck: converting external scenes into interpretable neural activity in surviving retinal circuits with adequate selectivity, safety margins, and long-term stability. This review critically compares the two approaches in spatial resolution, neural adaptation, and biosafety, while explicitly distinguishing clinical evidence from preclinical and proof-of-concept studies. Retinal prostheses provide near-term light perception and limited form vision but are constrained by current spread, non-selective pathway/axon recruitment, interface reactions, and cortical decoding burden. Optogenetic restoration offers cell-type-biased reactivation and potentially more physiological processing, but currently remains coupled to external light-delivery hardware and is limited by scattering, heterogeneous transduction, safety-limited irradiance, restricted dynamic range, and adeno-associated virus (AAV)-related inflammatory risk. We highlight emerging integrative directions, including opto-electronic hybrid interfaces, adaptive encoding, broadband subretinal nanomaterial-based photovoltaic nanoprostheses, and experimental acoustic modalities (e.g., focused ultrasound and sonogenetics), alongside ethical and accessibility considerations for translation.</p>","PeriodicalId":94015,"journal":{"name":"Frontiers in medical technology","volume":"8 ","pages":"1898982"},"PeriodicalIF":4.6,"publicationDate":"2026-08-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13551047/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148914815","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"HSD-Net: a dual-branch CNN-BiLSTM network with hybrid cepstral fusion for heart sound classification.","authors":"Caijian Hua, Ye Tian, Liuying Li, Xia Zhou","doi":"10.3389/fmedt.2026.1883548","DOIUrl":"10.3389/fmedt.2026.1883548","url":null,"abstract":"<p><strong>Introduction: </strong>Cardiovascular diseases (CVDs) represent a major global health threat, making early detection crucial. While cardiac auscultation is cost-effective, its reliance on clinical expertise leads to significant variability in diagnostic accuracy. Existing automated heart sound classification methods suffer from inadequate feature representation and limited model performance due to constraints imposed by the randomness, nonstationarity, and complex acoustic environment of phonocardiogram (PCG) signals.</p><p><strong>Methods: </strong>In this paper, an improved dual-branch CNN-BiLSTM network with hybrid cepstral fusion, named HSD-Net, is proposed for automated heart sound classification. To overcome feature representation limitations, we introduce a hybrid cepstral fusion feature set that combines Mel-Frequency Cepstral Coefficients (MFCC) and Inverse Mel-Frequency Cepstral Coefficients (IMFCC) with their first-order delta coefficients, capturing complementary low-frequency and high-frequency acoustic information. This feature set is processed by a dedicated dual-branch architecture: a convolutional neural network (CNN) branch extracts localized spectrotemporal patterns, while a bidirectional Long Short-Term Memory (BiLSTM) branch models long-range temporal dependencies. A Squeeze-and-Excitation (SE) attention mechanism is integrated to adaptively recalibrate feature importance.</p><p><strong>Results: </strong>Extensive evaluations on two public datasets demonstrate that the proposed HSD-Net achieves superior performance compared with existing methods. On the PhysioNet/Computing in Cardiology Challenge 2016 dataset, the model achieved an accuracy of 96.25%, a sensitivity of 95.85%, and a specificity of 96.50%. On the additional five-class Yaseen dataset, the model achieved an average accuracy of 99.32% in valvular heart disease classification. Ablation studies quantitatively confirmed that both the hybrid features and the dual-branch architecture contribute significantly to the overall performance.</p><p><strong>Discussion: </strong>The proposed HSD-Net provides an effective framework for automated heart sound analysis. This work offers a reliable technical foundation for developing clinical decision-support systems, with significant potential for application in early screening and telemedicine.</p>","PeriodicalId":94015,"journal":{"name":"Frontiers in medical technology","volume":"8 ","pages":"1883548"},"PeriodicalIF":4.6,"publicationDate":"2026-08-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13529997/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148876238","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}