Chinese Journal of Biomedical Engineering

Home

Editorial Board

Download

Quick Search

Office Online

Current Issue

Download Articles

Download

Links

2025 Vol. 44, No. 2 Published: 2025-04-20

Reviews Regular Papers

Regular papers

	129	Analysis of Induced Neural Cells Differentiation Stages Based on Transfer Learning with CNN
		Huang Xingye, Wei Guochao, Guo Yunxia, Shen Jiahao, Bi Kun, Zhang Zequn, Huang Yan, Zhao Xiangwei
		DOI: 10.3969/j.issn.0258-8021.2025.02.001
		With the high prevalence of neurological diseases, neural stem cell therapy has emerged as a significant research focus. However, current technologies still face challenges in accurately evaluating the neural stem cell differentiation process. This work aimed to develop a method using convolutional neural networks (CNNs) with transfer learning to efficiently classify images of label-free neuronal cells at different differentiation stages. Additionally, single-cell transcriptome data was leveraged to enhance the accuracy of cell differentiation stage detection and to validate the method′s effectiveness. The dataset including1 026 bright-field images of PC12 cells across different differentiation stages, supplemented by 100 images of HEK-293A or HT1080 cells obtained from publicly available online databases. Four CNNs including ResNet34, were evaluated. The models were conducted pre-training on the ImageNet dataset, followed by fine-tuning using a cell image dataset with annotation. The transcriptomic analysis was conducted on each cell after classifying the differentiation stages of thebright-field neural cell images. The results showed that the ResNet34-TL model performed best in classifying neural cells with an accuracy of 95.8%. The transcriptomic analysis of cells classified as undifferentiated, low-differentiated, and highly-differentiated by ResNet34-TL model indicates significant differences, particularly between the undifferentiated and highly-differentiated groups. Cells in the low-differentiated group exhibited characteristics of both extremes, suggesting a transitional state. The differential expression level of characteristic genes across the three groups revealed that the transcriptome-based classification was consistent with the model's classification results. ResNet34-TL exhibited strong generalization ability in the analysis of neuronal cell differentiation stages, and its capacity to effectively distinguish neural cells at different differentiation stages was further validated through transcriptomic analysis.
		2025 Vol. 44 (2): 129-141 [Abstract] ( 56 ) HTML (1 KB) PDF (0 KB) ( 16 )

	142	EEG Emotion Recognition Method Based on Time Convolutional Neural Network
		Peng Lei, Wei Guohui, Ma Zhiqing, Feng Jinyu, Li Yanjun
		DOI: 10.3969/j.issn.0258-8021.2025.02.002
		Emotion recognition based on EEG signals plays an important role in the field of human-computer interaction. However, due to the time-varying nature of EEG signals (features may vary significantly at different time periods) and multi-scale characteristics (different features are exhibited at different time and spatial scales), existing deep learning methods often struggle to comprehensively capture and extract various emotion related features from EEG signals. To extract the rich emotional information contained in the time-frequency spatial features of EEG signals, an EEG emotion recognition model that integrates convolutional neural networks (CNN), temporal convolutional networks (TCN), and transformer attention mechanism, frequency spatiotemporal attention temporal convolutional networks (FSA-TCN), was proposed in this work. Firstly, the CNN frequency spatiotemporal convolution layer was used to learn frequency domain information, spatial information, and time domain information, and extracted the frequency spatiotemporal features of EEG signals. Next, the TCN was fused with the Transformer attention mechanism to capture the temporal dependence of frequency spatiotemporal fusion features and extract deep EEG fusion features. Finally, the deep EEG fusion features were input into the fully connected layer for classification. This model conducted ablation experiments and subject dependent and cross subject EEG emotion recognition experiments on 76 800 EEG data samples on the DEAP dataset to verify the effectiveness of each module of the model and its effectiveness in EEG emotion recognition. It achieved emotion recognition accuracy of 92.96% and 92.90% in the valence and arousal dimensions, respectively. In addition, the model's generalization performance was validated on the SEED dataset, and its ability to recognize emotions across datasets was evaluated. The results indicated that the model was able to extract frequency spatiotemporal fusion features of EEG signals and mine deep EEG fusion features and achieved high-precision EEG emotion recognition.
		2025 Vol. 44 (2): 142-152 [Abstract] ( 39 ) HTML (1 KB) PDF (0 KB) ( 47 )

	153	EEG Signal Complexity Analysis Based on Multi-Frequency Permutation Entropy
		Niu Yan, Gao Kai, Ding Runan, Wen Xin, Zhou Mengni, Xiang Jie
		DOI: 10.3969/j.issn.0258-8021.2025.02.003
		Complexity analysis is of great importance in electroencephalography (EEG) signal studies. Multivariate entropy methods have proven to be effective techniques for analyzing signal complexity, but existing multivariate entropy studies have set the variables as multichannel time series. The quantification of brain dynamics' complexity from a multi-frequency analysis perspective has not been widely explored. By improving the multivariate permutation entropy (mvPE) algorithm, this study proposed multi-frequency permutation entropy (mFPE) to provide a more detailed measure of brain complexity in the time-frequency dimension. The study analyzed the performance of the algorithm using simulated data and three sets of real EEG data. Firstly, the performance of the mFPE algorithm was analyzed using 1/f noise and Gaussian white noise, as well as simulated data generated by the MIX model. It was found that mFPE had higher sensitivity, shorter data length requirements and good anti-noise performance compared to mvPE. Applying the mFPE algorithm to the analysis of EEG data from 14 Parkinson′s patients and 14 healthy controls, mFPE was able to significantly differentiate between normal and patient brain activity and achieved 78.7% classification accuracy, which was superior to mvPE (72.8%). Secondly, using EEG data from 14 patients with depressive tendencies and 14 healthy controls also revealed a 6.6% improvement in accuracy with mFPE compared to mvPE. Finally, using visual task EEG data from 32 normal subjects, mFPE effectively revealed the changes in EEG activity induced by different task stimuli, and the classification accuracies for different tasks were also higher than those of mvPE. This study showed that the mFPE algorithm provided a new perspective and an effective tool for the dynamic analysis of EEG signal complexity, which is expected to play an important role in the fields of neurological disease diagnosis, brain function research and cognitive science.
		2025 Vol. 44 (2): 153-164 [Abstract] ( 35 ) HTML (1 KB) PDF (0 KB) ( 7 )

	165	Research on Lightweight Transformer Medical Image Segmentation with Multi-Scale FeatureFusion
		Wang Xiaowei, Xing Shuli, Mao Guojun
		DOI: 10.3969/j.issn.0258-8021.2025.02.004
		UNet has been widely used in the field of medical image segmentation, and its U-shaped encoder-decoder structure has become one of the most popular frameworks. However, the classification and localization accuracy of UNet are limited by the local receptive field of convolutions, which restricts its ability to effectively capture long-range dependencies. Transformer has been demonstrated outstanding capabilities in capturing long-range dependencies and serves as the core supporting technology for current large language models, addressing the limitations of convolutional neural networks. In this paper, a novel medical image segmentation model referred to as MoFormer was proposed. Based on the encoding decoding structure of UNet, this model integrated Transformer learning mechanism in the encoder to expand its context aware field of view and enhanced the multi-scale feature extraction ability of local and global information. The proposed MoFormer with random initialization achieved an average Dice coefficient of 0.823 on the BTCV dataset with 50 abdominal CT images. On the ISIC2017 dataset containing 2 750 dermoscopy images, it performed equally well as TransFuse but with 10.91 M fewer parameters. On the polyp dataset which includes 2 590 endoscopic images, it outperformed other popular comparison models, such as PraNet, with the increase in mIoU value by an average of 0.123. Overall, this neural network model balances the number of parameters with segmentation accuracy, demonstrating strong generalization across various medical image datasets.
		2025 Vol. 44 (2): 165-173 [Abstract] ( 69 ) HTML (1 KB) PDF (0 KB) ( 46 )

	174	Design of Wearable Flexible Antenna for Transcutaneous Wireless Power Transfer
		Xu Qi, Feng Chenyu, Sun Yuan
		DOI: 10.3969/j.issn.0258-8021.2025.02.005
		The bidirectional peripheral nerve interface (PNI) can help to rebuild the sensory function of amputees. In this work, a flexible wearable antenna was proposed to enable a focused midfield wireless power for the implanted PNI. To generate a focused field, a symmetrical quadruple-feed antenna of 60 mm×60 mm was slotted to optimize the surface current distribution. The simulation results showed that the resonant frequency of the antenna was 1.524 GHz with the absolute bandwidth of 1.385~1.726 GHz, the resonant frequency offsets were less than 0.2 GHz for the bent antennas with the curvature radius varying from 60 mm to 110 mm. The simulated magnetic field intensity at the implanted receiver was 0.012 A/m, 0.058 A/m, and 0.065 A/m for in-phase excitation, time-reversed phase excitation, and orthogonal phase excitation, respectively. The experimental results showed that the resonant frequency of the proposed antenna was 1.531 GHz with the absolute bandwidth of 1.401~1.765 GHz, the variations of the antenna frequency due to curvature were less than 0.3 GHz with the curvature radius of 60 mm, 80 mm, and 100 mm. The measured magnetic fields at the receiver within the simulated tissue gel, generated by the antenna with the time-reversed phase excitation (0.045 A/m) and the orthogonal phase excitation (0.049 A/m) were much greater than that by using the in-phase excitation (0.0035A/m). In conclusion, the symmetrical quadruple-feed antenna with the time reversed phase excitation or the orthogonal phase excitation can generate observable focusing effect.
		2025 Vol. 44 (2): 174-182 [Abstract] ( 30 ) HTML (1 KB) PDF (0 KB) ( 11 )

	183	Cropping Process Study of Micro Probes Based on Finite Element Simulation
		Tan Xinyue, Ni Changmao, Cheng Linyuan, Yang Dengfei, Tong Bei, Huang Yuzhao, Huang Li
		DOI: 10.3969/j.issn.0258-8021.2025.02.006
		In the process of implanting high-density array micro probes into brain tissue, due to the large tip coverage area, the microprobes at corresponding positions in the array need to be cropped to avoid puncture of large blood vessels during implantation, while ensuring that the rest of microprobes are not affected. To determine the best point of the placement and pushing speed of the cropping device, the cropping process was studied by combining experiment and finite element simulation. Firstly, a three-point bending experiment was designed and the mechanical parameters of microprobes needed for simulation calculation were obtained. Secondly, based on the experimental results, a simplified model of array microprobes and cropping device was established in the simulation software Abaqus 2021, and the influence of the placement position, movement speed and alignment deviation of the cropping device on the cropping effect were explored. Finally, the simulation optimized parameters are used to carry out the microprobes cropping experiment. The results showed that the optimal distance between cropping device and the root of the microprobes was 1 mm, and the cropping speed was 7.5 mm/s. The actual success rate of microprobes cropping using this parameter reached 97.9%, and the cropping operation did not affect the other microprobes in the array. The results of this study provided strong support for the subsequent design of the structure and motion parameters of different type high-density microprobes array cropping device.
		2025 Vol. 44 (2): 183-192 [Abstract] ( 26 ) HTML (1 KB) PDF (0 KB) ( 3 )

	193	Study of Hand Joint Kinematics and Biomechanical of Grasping Contact Force
		Liu Xiaojie, Zhang Xushu, Guo Yuan, Wen Yunpeng, Wang Ruixue, Zhang Ming
		DOI: 10.3969/j.issn.0258-8021.2025.02.007
		This work performed biomechanical analysis of finger joint kinematics and grip contact force, aiming to provide reference for design of the motion and force feedback control of prosthetic hands. This study recruited 50 college students and set up three working conditions: manual gripping of different mass weights, cups with different volumes of water, and spheres with different diameters. Hand kinematic data and grip contact force data were collected from the participants. In the hand kinematics experiment, the Vicon motion capture system was used to collect the kinematic data of each marked point affixed to the subject's hand, and the angles of finger joints were calculated according to the space vector angle formula. In the hand contact force acquisition experiment, a thin film pressure sensor was used to acquire contact force information of the hand. USB-210 eight-channel data acquisition card was used as a pressure acquisition device and INSTRON 5544 material testing machine was used to calibrate the sensor. The sensor calibration curve was fitted with an exponential function, and the R2 reached 0.995 9. The contact force of each collection area was calculated. According to the kinematic characteristics of human fingers in different working conditions, the contact force data were analyzed statistically, and the biomechanical characteristics were studied. The results showed that the movement of each finger and the distribution of contact force in hand movement were synergetic, and the thumb, middle finger and index finger had a greater impact on the grasping function of the opponent, contributing 31.59%, 22.2% and 15.88% of the contact force respectively. The three fingertips contact force had significant differences with the increase of weight mass (P<0.05). With the increase of the number of finger joints and the complexity of movement, the contact force between fingertip and palm area also showed significant differences (P<0.05). There was no significant interaction effect between the gender difference of the subjects and the differences between the working conditions (P>0.05). These results provided data and theoretical reference for the design of humanoid prosthetic hand.
		2025 Vol. 44 (2): 193-202 [Abstract] ( 56 ) HTML (1 KB) PDF (0 KB) ( 13 )

	203	Evaluation of the Safety and Efficacy of Breast Cancer Tumor-Bed Degradable Markers in Rats
		Xie Xuejie, Xie Xin, Luo Yingen, Kong Xiangyi, Wang Ningyu, Dai Jianrong
		DOI: 10.3969/j.issn.0258-8021.2025.02.008
		The aim of this investigation is to evaluate the safety and efficacy of breast cancer tumor-bed biodegradable markers in rats. In experiments, a total of 100 rats were divided into surgical group, titanium clip group and degradable labeling group (n=30), and the other 10 rats were set up for the whole degradable observation group. The biodegradable, buckle-shaped marker with round holes and grooves were made of PLA/β-TCP material, and the markers and titanium clips were implanted into the rat mammary tissue. Hematoxylin and eosin staining (HE) and immunohistochemistry (IHC) were used to analyze the inflammatory responses of rat perimammary tissues after marker implantation. Using CT and MR imaging, the degradable markers and titanium clips were observed for the presence of artifacts, and the ability of the degradable markers to achieve effective visualization over the observation period (46 weeks) was evaluated. Safety results showed that the acute immune responses in the tissues around the degradable markers were higher than that of the titanium clips and returned to normal after 12 weeks. The gap between the expression of IL-6 in the degradable labeling group and the titanium clips group was reduced from 12% to 2.8%. The gap between the expression of IL-10 in the degradable labeling group and the titanium clips group was reduced from 14.8% to 8.1%. Statistical analysis showed that the difference in the rate of positive cells for inflammatory factors was significant in the first 4 weeks. The validity results showed that the boundaries of the degradable markers were clear without obvious artifacts on CT and MR images, and the CT values tended to decrease gradually with the increase of time. The results showed that the inflammatory reaction induced by the degradable marker in rats basically disappeared at week 12, and the imaging effect on CT and MR was better than that of the titanium clip marker. In the future, this new biodegradable marker could potentially be an alternative to titanium clips.
		2025 Vol. 44 (2): 203-210 [Abstract] ( 28 ) HTML (1 KB) PDF (0 KB) ( 6 )

Reviews

	211	Research Advance in Lightweight Methods for Sensing and Intelligent Analysis of Lumbar Electromyographic Signals
		Feng Jinghui, Yu Yu, Xi Jianing
		DOI: 10.3969/j.issn.0258-8021.2025.02.009
		The issue of lumbar health has garnered increasing attention in the light of the aging population trend. Lumbar electromyography (EMG) signals contain information about the lumbar and are commonly used to analyze lumbar conditions, and in recent years, related research has focused on detecting lumbar muscle fatigue and assisting in diagnosis. With the development of signal acquisition technology and analysis algorithms, it is possible to collect and transmit real-time data in large quantities from lumbar EMG and analyze them using complex neural network algorithms. However, complex acquisition and analysis systems increase resource consumption and are not conducive to lightweight deployment. In this paper, we reviewed the lightweighting of signal acquisition and analysis systems for lumbar EMG.Regarding the lightweight requirements of the waist electromyography signal acquisition and analysis system, this article reviewed the progress of waist electromyography acquisition technology, and summarized the lightweight research progress in this field from multiple dimensions, including hardware devices, data storage, and algorithm models. The key technologies of related system, such as high-performance communication chips and knowledge distillation neural networks, were introduced. And the future application of waist electromyography technology was also discussed.
		2025 Vol. 44 (2): 211-220 [Abstract] ( 40 ) HTML (1 KB) PDF (0 KB) ( 38 )

	221	Key Challenges in Task-Induced Fatigue Monitoring Based on EEG Signals
		Wang Tao, He Feng
		DOI: 10.3969/j.issn.0258-8021.2025.02.010
		With the advancement of electroencephalogram (EEG) technology, EEG-based fatigue monitoring models have become crucial tools for ensuring safety and improving efficiency in high-risk fields such as transportation and medical surgeries. These models provide objective fatigue assessments and early warnings, reducing human error and accident risks. However, their application in real-world scenarios faces challenges, including issues with the comfort and convenience of EEG acquisition, motion artifacts and electromyographic noise interfering with feature extraction, and variability in EEG signals both between and within individuals, which results in the complication of model generalization. Additionally, the lack of standardization in fatigue data labeling and collection leads to inconsistencies and biases. This review summarized these challenges and explored potential solutions, aiming to advance the practical application of EEG-based fatigue monitoring systems. Along with the progress of EEG acquisition technologies, signal processing algorithms and machine learning models, the accuracy, convenience and generalizability would be continuously improved in the future systems, enhancing work safety and efficiency.
		2025 Vol. 44 (2): 221-231 [Abstract] ( 53 ) HTML (1 KB) PDF (0 KB) ( 34 )

	232	Current Research on Hydrogel Adhesives for Cardiac and Vascular Repair
		Chen Shihan, Fan Jiefu, Lu Qingsheng
		DOI: 10.3969/j.issn.0258-8021.2025.02.011
		Hydrogels are highly biocompatible biomaterials composed of crosslinked three-dimensional networks of hydrophilic polymers, which can be structurally and functionally designed to modulate their physicochemical properties and thus have a wide range of biomedical application potential. Among them, hydrogels with adhesion ability are widely used as medical tissue adhesives in wound repair of skin, muscle, nerve, bone, cornea, organs and blood vessels due to their ability to mimic physiological environment, excellent biocompatibility, outstanding mechanical properties and most importantly - strong adhesion to tissues. In this paper, we introduced the current state of research on hydrogel adhesives, providing a comprehensive overview of their adhesion mechanisms to tissues, fundamental design principles, associated characteristics, and application contexts, with particular emphasis on their use in cardiac and vascular repair. Additionally, the limitations of existing hydrogel adhesive applications and anticipates future advancements were analyzed, aiming to offer a valuable reference for their clinical application in surgical settings.
		2025 Vol. 44 (2): 232-240 [Abstract] ( 38 ) HTML (1 KB) PDF (0 KB) ( 34 )

	241	Iron Oxide Nanoparticle: Preparation and Application in Tumor Diagnosis and Treatment
		Du Kai, Zhang Zhuoling, Li Tinghua, Rao Wei
		DOI: 10.3969/j.issn.0258-8021.2025.02.012
		Despite the significant challenges presented by the complex pathological processes of tumors, the rapid development of nanotechnology provides new solutions. Iron oxide nanoparticle-based nanocomposites have shown enormous potential in the realm of tumor diagnostics and therapy, effectively employed in applications such as magnetic resonance imaging, targeted drug delivery, magnetically activated mechanical therapy, and magnetic hyperthermia therapy. This paper systematically reviewed and outlined the advancements in the preparation methods of iron oxide nanoparticles, encapsulation, and applications in tumor diagnostics and therapy. Researchers have employed various synthesis methods and modification strategies to fabricate iron oxide nanomaterials with diverse sizes, shapes, and surface properties. By integrating these characteristics, researchers can analyze their impact on targeting, drug delivery efficiency, circulation time in vivo, imaging quality, and biocompatibility. Rational design of iron oxide nanoparticle-based nanocomposites facilitates the combined use of multimodal imaging and diversified therapeutic modalities, enhancing the effectiveness of tumor diagnostics and treatment while also accelerating the clinical translation of these composite materials.
		2025 Vol. 44 (2): 241-256 [Abstract] ( 35 ) HTML (1 KB) PDF (0 KB) ( 37 )

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