Chinese Journal of Biomedical Engineering

Home

Editorial Board

Download

Quick Search

Office Online

Current Issue

Download Articles

Download

Links

2026 Vol. 45, No. 2 Published: 2026-04-20

Reviews Communications Regular Papers Expert Consensus

Expert Consensus

	129	Expert Consensus on Olfactory Neural Prosthesis Technology for Reconstruction of Olfactory Perception
		Expert Group of Biomedical Measurement Branch, Chinese Society of Biomedical Engineering; Biomedical Sensing Technology Branch, Chinese Society of Biomedical Engineering; Chemical Perception Branch, Chinese Society for Cognitive Science
		DOI: 10.3969/j.issn.0258-8021.2026.02.001
		Drawing upon cutting-edge research and insights from multidisciplinary expert, this consensus seeks to provide feasible technical and clinical pathway recommendations for the perception reconstruction in patients with lifelong anosmia. The document encompasses criteria for patient inclusion and exclusion, bio-inspired olfactory sensing strategies, spatiotemporal dynamic electrical stimulation modulation schemes utilizing brain-computer interfaces, perioperative management, long-term follow-up procedures, and a comprehensive ethical and risk governance framework. It is important to note that research on olfactory neural prostheses is still in the early stages of exploration. Therefore, translational research and clinical trials should be carried out steadily, adhering strictly to ethical standards and quality control measures. The ultimate objective is to achieve the safe, effective, and long-term reconstruction of artificial olfactory perception, thereby enhancing China's impact in the international academic arena within this field.
		2026 Vol. 45 (2): 129-140 [Abstract] ( 51 ) HTML (1 KB) PDF (2072 KB) ( 47 )

Regular Papers

	141	Mamba Dual-Branch Medical Image Fusion Model Based on Channel Attention Mechanism and Multi-Scale Rotational Convolution
		Kong Weiwei, Han Yinbo, Lei Yang, Wang Yuchen, Zhou Haochen
		DOI: 10.3969/j.issn.0258-8021.2026.02.002
		To address the high computational complexity of Transformers, a Mamba dual-branch fusion network based on channel attention mechanism and multi-scale rotation convolution (rotation-enhanced Mamba fusion model, ReMFM) was proposed in this work. First, the channel attention mechanism was ued to capture the inter-channel information correlation of the images to be fused. Next, a rotation convolution module was designed to efficiently extract cross-modal local structural features in both direction and scale dimensions. Finally, an improvement was made to Mamba by designing an attention state space module, which introduced non-causal modeling and global perception capability with single-scan processing. These actions significantly reduced computational complexity and redundancy while ensuring expressiveness. The data used in this experiment were obtained from the Harvard Brain Atlas Database, comprising 357 paired MRI and SPECT images, with 333 pairs allocated for training and 24 pairs for testing. Data augmentation was performed using an overlapping cropping strategy, and all training images were standardized to a uniform size of 120 × 120 pixels. Experimental results showed that ReMFM achieved 0.7438, 0.7457, 0.9767, 0.6884, and 5.0226 in gradient, image feature, Yann measure, visual information fidelity, and mutual information, respectively, with improvements of 2.52%, 15.92%, 2.41%, 14.64%, and 14.29% over seven mainstream Transformer-based models. The proposed model effectively highlights the structural information of the lesion regions while preserving edge textures, producing high-quality fused images.
		2026 Vol. 45 (2): 141-153 [Abstract] ( 64 ) HTML (1 KB) PDF (8908 KB) ( 25 )

	154	GLCFormer: A Breast Cancer TMB Prediction Model Based on Digital Pathological Images
		Zhang Xiaoyan, Liu Yan, Zhao Zheng, Meng Xiangfu, Li Shuai
		DOI: 10.3969/j.issn.0258-8021.2026.02.003
		Tumor mutation burden (TMB) is a key biomarker to predict the efficacy of immune checkpoint inhibitors (ICIs) in breast cancer patients. Existing deep learning methods for WSI-based TMB prediction face two critical challenges: 1) Self-attention easily induces redundant global interaction responses, reducing global modeling efficiency; 2) Lack of adaptability in global-local feature fusion causes redundancy, hindering effective cross-scale integration. To address these issues, this study proposed GLCFormer, a global-local dual-branch parallel collaborative model. ReLU linear attention was integrated into the global branch to suppress weakly correlated features via nonlinear mapping, enhancing computational efficiency and extracting global semantic information from pathological images. To strengthen fine-grained structural representation at the same feature level, the local branch was designed with a multi-scale convolution block composed of parallel depthwise separable convolutions. Dual-branch features were fed into the dynamic cross-scalefusion (DCSF) module, which adaptively generates weights to reduce redundancy and enable efficient global-local integration. Five independent repeated experiments were conducted on TCGA-BRCA (198 patients’ WSI data, 614,253 training samples). Results showed that GLCFormer achieved an AUC of 98.8%±0.1% in TMB classification (4.6% higher than state-of-the-art CoAtNet), with statistical significance (P<0.05) via independent two-sample t-test. In regression, it yielded a mean absolute percentage error (MAPE) of 0.2874, outperforming all comparators. These results validated the proposed method’s favorable accuracy and stability in WSI-level TMB prediction, supporting clinical precise immunotherapy.
		2026 Vol. 45 (2): 154-166 [Abstract] ( 42 ) HTML (1 KB) PDF (6948 KB) ( 10 )

	167	Recognition of Esophageal Cancer and Precancerous Lesions Based on CNN and ViT Complementary Learning Network
		Chen Tuo, Lin Zhigang, Chen Hong, Wu Nengguang, Gao Yang
		DOI: 10.3969/j.issn.0258-8021.2026.02.004
		Accurate identification of esophageal cancer and precancerous lesions can improve the success rate of early intervention and increase patients' survival expectations. However, the lesion area in esophageal endoscopic images has the problem of high intra class diversity and high inter class similarity, which makes it difficult for existing methods to extract effective features and affects recognition performance. Therefore, a classification method for esophageal cancer and precancerous lesions based on a complementary learning network of convolutional neural network (CNN) andvision Transformer (ViT) was proposed to improve recognition accuracy. This network consists of two parallel feature extraction branches, which are used to learn local and global features in the image. First, the local feature extraction branch based on multidimensional attention module can obtain more discriminative local detail features within the lesion area. Second, the global feature extraction branch based on multi-directional attention module is used to learn multi-scale global semantic information. Finally, the cross scale complementary learning module is used to promote complementary learning between branches, improve the feature expression ability of the entire network, and achieve high-precision recognition of the disease. Experimental validation was conducted on a dataset of 3 730 white light endoscopic images of esophageal cancer and precancerous lesions, and the recognition accuracy reached 96.2%, exceeding the baseline model by 5.6 percentage points and outperforming other methods compared in the experiment; The optimal recognition accuracy was also achieved for each category on a public dataset of 6 000 gastrointestinal diseases(Kvasir-dataset), demonstrating good generalization ability. The proposed recognition model based on CNN and ViT complementary learning network can better capture the rich visual features in esophageal endoscopic images, thereby effectively improving recognition accuracy and providing important value for doctor assisted diagnosis.
		2026 Vol. 45 (2): 167-177 [Abstract] ( 40 ) HTML (1 KB) PDF (3531 KB) ( 27 )

	178	ECSA-EEGNet: A Lightweight EEG Decoding Model with Efficient Channel-SpatialAttention and Multi-Scale Fusion
		Yang Yabing, Dong Zheng, Bao Xueliang, Zhang Peng
		DOI: 10.3969/j.issn.0258-8021.2026.02.005
		To address challenges of low spatial-channel feature fusion efficiency, high computational complexity, and weak cross-subject generalization in existing motor imagery (MI) electroencephalogram (EEG) decoding models, this study aimed to construct a novel decoding architecture that balances high accuracy with lightweight design for resource-constrained embedded applications. We proposed the ECSA-EEGNet, a lightweight model incorporating three key designs. First, a dynamic channel-space attention module was designed, utilizing parallel dual-pooling and adaptive weighting mechanisms to replace traditional fully connected dimensionality reduction, thereby enhancing the perception of key brain region features without information loss. Second, the attention sub-modules were reconstructed by adopting a fully convolutional structure to reduce redundant parameters (approximately 40% reduction compared to traditional mechanisms) and introducing a convolution-batch normalization-mish (CBM) unit to strengthen non-linear modeling capabilities. Finally, a multi-scale channel-space fusion framework was constructed to improve robustness against individual differences through cross-dimensional feature interaction. The model was validated on the BCI Competition IV-2a (4-class) and IV-2b (2-class) datasets involving nine healthy subjects. Results indicated that ECSA-EEGNet achieved an average accuracy of 80.29% on the IV-2a dataset (a 7.89% improvement over the baseline EEGNet) with a Kappa coefficient of 0.72. On the IV-2b dataset, it reached an accuracy of 85.22% (a 9.50% improvement) with a Kappa coefficient of 0.70. The model comprises only 5,022 parameters, approximately one-third of those in TCNet-Fusion. In fine-grained evaluation, Subject 3 in the IV-2a dataset achieved recognition rates exceeding 90% for all four tasks (left hand, right hand, foot, and tongue), with a peak accuracy of 95.83%, confirming the model's stability in complex tasks. These findings demonstrated that ECSA-EEGNet achieved a significant leap in decoding performance with minimal parameter overhead, effectively balancing accuracy and efficiency, thus providing a viable lightweight solution for the clinical deployment of real-time brain-computer interface systems.
		2026 Vol. 45 (2): 178-187 [Abstract] ( 28 ) HTML (1 KB) PDF (4932 KB) ( 17 )

	188	A Four-Channel Heart-Sound Acquisition and Denoising System Based on ECG-Scale andPneumatic Control
		Jiang Houqi, Zhang Chi, She Jin, Li Deyu
		DOI: 10.3969/j.issn.0258-8021.2026.02.006
		Cardiovascular diseases pose a significant threat to the health of Chinese residents. Multi-channel electronic auscultation equipment can provide important information for cardiovascular disease screening and is currently a research hotspot; however, there are issues such asredundant sensors, low stability, and poor portability. This study aims to design a wearable multi-channel heart sound acquisition system to overcome the limitations of traditional equipment in stability, portability, and signal quality, providing technical support for early screening and long-term monitoring of cardiovascular diseases. The proposed multi-channel heart sound acquisition system in this paper adopted a vest-like design. A piezoelectric sensor CM-01B was used as the heart sound acquisition sensor, which is embedded in a gel and integrated into the vest to ensure stable contact between the sensor and the human surface. An air pressure control system was used to automatically inflate the airbag through closed-loop control to ensure that the sensor was uniformly pressed on the acquisition site, improving the stability of signal acquisition.Using the signals collected byan electronic stethoscope(ETZ-1A) as a reference, the frequency domain similarity was used to evaluate four heart sound sites, four lung sound sites, and their combinations, showing that the optimal acquisition sites for heart sounds included aortic valve, mitral valve, upper part of the right sternoclavicular line, and lower part of the left axillary anterior line. This paper proposed a heart sound denoising algorithm -ECG-guided cardiac sound denoising via DTW-PCA template matching algorithm (EDPT), which introduced single-lead electrocardiogram signals to provide an electrocardiogram scale for heart sound denoising; using electrocardiogram signals to segment the cardiac cycle and extract standard heart sound templates for matching filtering to achieve separation and denoising of heart sounds; and evaluating the effect of the EDPT algorithm using signal-to-noise ratio.A total of 16 healthy adult subjects (8 males and 8 females) were recruited for experimental validation. After introducing the electrocardiographic scale, the mean signal-to-noise ratio (SNR) of the separated heart sounds obtained by the EDPT algorithm increased from (18.09±0.50)dB to(29.69±1.49)dB. Pairedt-test analysis demonstrated that the separation performance of the EDPT algorithm was significantly superior to that of the conventional ICA algorithm (P<0.001). These results indicate that the EDPT-based heart sound denoising algorithm developed in this study can more effectively separate heart sound signals, thereby validating both the effectiveness of the algorithm and the feasibility of the proposed system. The multi-channel heart sound acquisition system developed in this study, by introducing the electrocardiogram scale and air pressure control technology, combined with the EDPT heart sound denoising algorithm, significantly improved the stability and signal quality of heart sound acquisition. This system can provide technical support for cardiovascular disease screening based on auscultation and may also be extended to other auscultation fields such as lung sound auscultation in the future.
		2026 Vol. 45 (2): 188-198 [Abstract] ( 30 ) HTML (1 KB) PDF (7481 KB) ( 13 )

	199	Motor Imagery Decoding Based on Spectral Attention
		Chen Xuejun, Chen Hongda, Zhang Chenhua, Chen Weijie
		DOI: 10.3969/j.issn.0258-8021.2026.02.007
		Aiming to solve the problem of insufficient ability of feature extraction by the traditional motor imagery decoding algorithm for long time series and low recognition rate of EEG signals, a novel CNN wavelet decomposition spectral encoder EEG decoding algorithm was proposed in this work, based on the fusion idea of convolutional neural network and transformer encoder and combining wavelet decomposition and spectral attention. First, the convolution module was constructed for local feature extraction, and the multi-level sub-features were obtained by one-level wavelet decomposition. Based on the transformer encoder, spectral attention was introduced to replace the traditional multi-head attention, and a new FFT-Former encoder was constructed followed by the input of multi-level sub-features into the spectrum attention for frequency domain modeling. Next, the time-domain features obtained by wavelet reconstruction technology were input into the feedforward network to enable the model to learn more complex feature dependencies. Finally, the feature information of convolution module and FFT-Former was fused to design a classifier to achieve low computational complexity and high accuracy EEG signal recognition. Verified by public datasets, the average accuracy of motor imagery decoding of CWFT network model on BCI Competition IV-2a and BCI Competition IV-2b reached 85.65% and 89.54% respectively, and the Kappa coefficient reached 80.86% and 79.07% respectively. By comparing with other comparison algorithms, we demonstrated that the proposed algorithm had excellent classification performance and provided a new idea for the construction of motor imagery brain-computer interface.
		2026 Vol. 45 (2): 199-210 [Abstract] ( 65 ) HTML (1 KB) PDF (5153 KB) ( 15 )

	211	Friction-Based Self-Powered Sensing Technology for Intelligent Monitoring of Sleep Breathing
		Cao Yixin, Liu Bing, Yu Yutong, Zhu Heyao, Yi Huansheng, Lu Guohua, Xu canhua, Qi fugui
		DOI: 10.3969/j.issn.0258-8021.2026.02.008
		Real-time monitoring of sleep apnea syndrome is crucial for early diagnosis and health management. Current clinical monitoring techniques face challenges such as complex wiring, the need for a continuous power supply, and poor portability, which hinder long-term home-based monitoring. To address these issues, this study proposed a self-powered intelligent sleep respiration monitoring system based on a triboelectric nanogenerator (TENG) for real-time monitoring and warning of sleep respiratory status. The system converted respiratory movements into electrical signals through a contact-separation mode triboelectric sensing mechanism. The sensor employed a PDMS/thermally expandable microsphere composite film and FEP as triboelectric layers, with optimized fabrication parameters (copper film size: 5 cm × 5 cm, thickness: 100 nm; PDMS-to-microsphere mass ratio: 100:1; spin-coating speed: 4000 rpm; heating conditions: 110℃ for 10 min). A dual-stream decision fusion method combining energy feature extraction and support vector machine (SVM) classification was adopted to achieve accurate recognition of respiratory status. Test results on 50 sets of sleep respiratory data from 10 volunteers showed that the system accurately extracted key parameters including respiratory rate, apnea events, and cumulative duration, with an apnea event recognition accuracy exceeding 90%, and triggered real-time alarms based on thresholds. The developed system features self-powering, high sensitivity, wearability and low cost, making it suitable for long-term respiratory monitoring in home environments and providing a feasible solution for home-based screening and health management of sleep apnea syndrome.
		2026 Vol. 45 (2): 211-219 [Abstract] ( 41 ) HTML (1 KB) PDF (5115 KB) ( 14 )

Reviews

	220	Research Progress of Machine Learning in Sports Injuries
		Yu Jun, Wu Chengliang, Tang Jia, Zhang Yugui
		DOI: 10.3969/j.issn.0258-8021.2026.00.000
		With advantages in high-dimensional data processing and complex nonlinear relationship mining, the machine learning technology provides a new approach for risk prediction and mechanism analysis in sports injuries. This paper focused on research progress in this field, discussing key roles of machine learning in injury risk identification, prediction, and early warning, and providing theoretical support for the development of a scientific and accurate sports injury prevention and control system. First, from the technical process of model construction, we summarized the common methods and technical characteristics of the three core links: feature engineering, model establishment, and optimization validation. Second, we compared the technical principles, application scenarios, and optimization paths of seven mainstream algorithms: decision tree, logistic regression, support vector machine, random forest, XGBoost, neural network, and hybrid learning, and describe their application effects in injury risk prediction. Finally, we analyzed the challenges faced by current research in terms of methodological heterogeneity, sample limitations, and model interpretability, and outline future research directions, including the construction of standardized systems, the improvement of data platforms, and the development of interpretable algorithms.
		2026 Vol. 45 (2): 220-232 [Abstract] ( 51 ) HTML (1 KB) PDF (932 KB) ( 15 )

	233	Research Progress in Virus Detection Based on Nanopore Sensors
		Wu Lingzhi, Qi Ke, Zhang Xuanhao, Tang Lihua
		DOI: 10.3969/j.issn.0258-8021.2026.02.010
		In recent years, infectious diseases caused by various pathogenic microorganisms have frequently emerged, posing a serious threat to human health and medical safety. There has been a growing demand for large-scale clinical diagnosis of viruses. With its unique single-molecule detection capability, simple operation process, and potential portability, nanopore technology is emerging as an important tool for virus detection. This article reviewed advances of nanopores on viral detection from several aspects, including key technologies and diagnostic applications. The typical cases of the nanopore platforms, such as nanopore sequencing, single particle counting, and single molecular identification, were further summarized on different scenarios of the pathogen surveillance, along with a brief overview of the nanopores integrated with other technologies, like machine learning and microfluidics, further enhancing detection capabilities and expanding their application. These technologies will be help for developing new approaches for precise epidemic prevention and control.
		2026 Vol. 45 (2): 233-242 [Abstract] ( 30 ) HTML (1 KB) PDF (4771 KB) ( 34 )

	243	The Research Progress of Multi-source Photosensitizers for Tumor Therapy
		Huang Yuzi, Wang Wei, Li Yan, Zhang Yuting, Geng Peng, Lan Haichuang, Huang Wenquan, Xiao Shuzhang
		DOI: 10.3969/j.issn.0258-8021.2026.02.011
		Malignant tumors pose a significant threat to human health, and breakthroughs in treatment technologies are of crucial clinical importance. Photodynamic therapy (PDT), as a non-invasive and highly selective cancer treatment strategy, has gained increasing attention in recent years due to its unique advantages,including precise targeting and strong controllability. This technique utilizes photosensitizers (PSs) that, when exposed to light of a specific wavelength, generate cytotoxic reactive oxygen species (ROS), which selectively induce apoptosis in tumor cells. As the core component of PDT, the physicochemical properties of photosensitizers—such as light absorption characteristics, ROS quantum yield, and biocompatibility—directly determine the therapeutic efficacy and clinical translation potential. This review systematically summarized the latest researchprogress on inorganic, organic, and metal-organic framework-based photosensitizers, exploring their potential applications in tumor treatment from multiple perspectives. It also highlighted the key scientific challenges and technical bottlenecks in the field. Finally, the review discusses the molecular design strategies and development directions for next-generation high-performance photosensitizers, providing theoretical insights for the design and optimization of new PSs.
		2026 Vol. 45 (2): 243-251 [Abstract] ( 36 ) HTML (1 KB) PDF (1279 KB) ( 25 )

Communications

	252	Experimental Study on Optimal Metabolic Energy Consumption of Human Body and EfficientPower-Assist Walking Indicators of Exoskeletons
		Liu Qiming, Wang Shan, Li Tao, Guo Shijie
		DOI: 10.3969/j.issn.0258-8021.2026.02.012

		2026 Vol. 45 (2): 252-256 [Abstract] ( 30 ) HTML (1 KB) PDF (5991 KB) ( 11 )

Copyright © Editorial Board of Chinese Journal of Biomedical Engineering
Supported by:Beijing Magtech