Chinese Journal of Biomedical Engineering

Home

Editorial Board

Download

Quick Search

Office Online

Current Issue

Download Articles

Download

Links

2018 Vol. 37, No. 6 Published: 2018-12-20

Reviews Communications Regular Papers

Regular Papers

	641	Automatic Classification of Retinal Optical Coherence Tomography Images via Convolutional Neural Networks with Joint Decision
		Wang Chong, He Xingxin, Fang Leyuan, Guo Siyu, Chen Xiangdong, Nie Fujiao
		DOI: 10.3969/j.issn.0258-8021.2018.06.001
		Optical coherence tomography (OCT) can provide in vivo three dimensional (3D) cross-sectional imaging of human retina at micrometer resolutions, which is a significant tool for the diagnosis and the progression tracking of retinal diseases. In the clinical diagnosis, ophthalmologists need to manually identify various macular lesions at each cross section of the 3D OCT images. Such manual analysis is time-consuming and often yields subjective results. Therefore, it is urgent to develop an automatic classification algorithm to improve the efficiency of OCT-based analysis in daily clinical practice. This paper proposed a novel automatic method, based on the convolutional neural networks (CNN) with joint decision for the classification of OCT images. The proposed joint-decision CNN-based method was able to automatically learn multiple-layer features from original OCT images through a convolutional neural network, these features in each layer were simultaneously utilized to separately make decision of classification. Finally, these decisions were fused together to achieve the final classification. The experimental results on Duke data sets (3231 B-scans) showed that the proposed joint-decision CNN-based method achieved average accuracy of 94.5%, average sensitivity 90.5% and average specificity 95.8%, for the automatic identification of normal macula, age-related macular degeneration and macular edema respectively. The experimental results on HUCM data sets (4322 B-scans) showed that the proposed joint-decision CNN-based method achievesd average accuracy of 89.6%, average sensitivity 88.8% and average specificity 90.8%. The results proved that the richly multiple-layer features of CNN could be used to accurately classify retinal OCT images, hence the algorithm provided effective technical support for the aided diagnosis of retinal diseases in clinical practice.
		2018 Vol. 37 (6): 641-648 [Abstract] ( 457 ) HTML (1 KB) PDF (7313 KB) ( 248 )

	649	Algorithm study of Plant Cell Tracking Based on Dynamic Local Graph Matching
		Qian Weili, Liu Min, Li Jieqin, Liu Xiaoyan
		DOI: 10.3969/j.issn.0258-8021.2018.06.002
		Developing algorithms for plant cell tracking in microscopic image sequences is very critical to the modeling of cell growth pattern and gene expression dynamics. The plant cells are tightly clustered in space and have very similar shapes and intensity distributions, and the images can be translated, rotated in the imaging process, thus tracking plant cell across image sequences is very challenging. This paper proposed a dynamic local graph matching method, which efficiently exploited the feature of the cells area, the angle and distance between adjacent cells to match the plant cells. The most similar cell pair was chosen as the seed cell pair by computing the feature distances between the cells in two adjacent images, and then their neighboring cells were gradually matched starting from the seed pair. During the dynamic local graph matching process, for each iteration, the matched cells were regarded as newly added seed cells, and the cells in the dynamically updated neighborhood with the least feature distance were matched firstly. Experimental results on three unregistered plant cell (Shoot Apical Meristem, SAM) image sequences and their registered image sequences showed that the proposed method improved the tracking accuracy rate by 4% in the registered image sequences and by 30% in the unregistered image sequences when compared with the existing plant cell tracking method. In conclusion, the method is valuable for plant cell population tracking in microscopic image data.
		2018 Vol. 37 (6): 649-656 [Abstract] ( 438 ) HTML (1 KB) PDF (14142 KB) ( 72 )

	657	Classification and Recognition of P300 Event Related Potential Based on Convolutional Neural Network
		Chou Yuanting, Qiu Tianshuang, Zhong Mingjun
		DOI: 10.3969/j.issn.0258-8021.2018.06.003
		To improve the recognition rate of P300 potentials in the brain computer interface system, a novel method based on the improved convolutional neural network was proposed. By transforming the second serially connected convolutional layer of a traditional convolution neural network to three parallel connected convolutional layers, the method widens the network to enhance the ability of feature extraction in the proposed network. Combining the extracted features with the fully connected layers and sigmoid function, a P300 visual evoked potential classifier was constructed. Targeting to the problem of unbalanced data volume between target and non-target stimulus data in BCIcompetition, this paper adopted an oversampling method. To increase the amount of data, this paper partially averaged the EEG data that contains P300 visual evoked potentials. The training set and test set sample sizes were 25500 and 18000, respectively. Adam optimization method was adopted to train the improved convolutional neural networksupervisely. The analysis results showed that the proposed network achieved an accuracy of higher than 95% when the number of experiments was over 11 times, which is of great significance for the application of brain-computer interface.
		2018 Vol. 37 (6): 657-664 [Abstract] ( 572 ) HTML (1 KB) PDF (1953 KB) ( 573 )

	665	Recognition of EEG Based on Ensemble Empirical Mode Decomposition and Random Forest
		Qin Xiwen, Lv Siqi, Li Qiaoling
		DOI: 10.3969/j.issn.0258-8021.2018.06.004
		It is of great importance to automaticaly monitor and classify epileptic EEG in clinical medicine. In view of the non-stationary characteristics of EEG signals, a new method for feature extraction and recognition of EEG based on ensemble empirical mode decomposition (EEMD) and random forest (RF) was proposed in this paper. In this study 200 single-channel signals of epileptic ictal and interictal EEG were selected from EEG data of Bonn University, and 819400 data were used as samples. Firstly, the EEG signals were decomposed into several intrinsic mode functions (IMF) by EEMD, and then the effective features were extracted from each IMF component. Finally, the features of each IMF component were classified by RF and least squares support vector machine (LSSVM). We compared the classification results of RF and LSSVM. The results showed that the classification effect of RF algorithm on epileptic EEG signals in ictal and interictal periods was effective. The recognition accuracy was 99.60%, which was higher than the accuracy of LSSVM. The proposed method could effectively improve the efficiency of clinical EEG signal analysis.
		2018 Vol. 37 (6): 665-672 [Abstract] ( 377 ) HTML (1 KB) PDF (1168 KB) ( 493 )

	673	Detection of Muscle Fatigue Based on sEMG Signal with AR Model
		Yang Zheng, Wang Liling, Ma Dong
		DOI: 10.3969/j.issn.0258-8021.2018.05.005
		According to the non-stationary characteristics of the surface electromyography signal, the autoregressive model was employed to analyze the surface electromyography signal. This method could rapidly estimate muscle fatigue by analyzing short surface electromyography signal. Surface electromyography signals from 10 subjects with pre-fatigue and post-fatigue were collected and analyzed using an autoregressive model. The autoregressive model parameters was identified by the Legendre?basis function expansion method, transforming the linear non-stationary problem into the linear time-invariant one. The autoregressive model parameters was solved in least square method. Changing rates (in percentage) of pre-fatigue and post-fatigue for the first parameter of the autoregressive model (ACR1), mean power frequency (MPF), median frequency (MF) were calculated and compared using two-tailedsamples t-test. The results showed that the changing rates of ACR1, MPF and MF were 34.33%±2.41%、25.68%±2.03% and 22.80%±2.19%, respectively. And the changing rate for ACR1 was significantly higher than that for both MPF and MF (P<0.05). ACR1 could not only realize the rapid assessment of muscle fatigue on short surface electromyography signal, but also has higher sensitivity than MPF and MF, providing a promising assessment method in the field of the upper muscular strain and the rehabilitation.
		2018 Vol. 37 (6): 673-679 [Abstract] ( 361 ) HTML (1 KB) PDF (2396 KB) ( 777 )

	680	Coherence Synchronization Analysis Based on Smoothing Minimum Variance Distortionless Response
		Gu Guanghua, Cui Dong, Wang Juan, Qi Shunai, Li Xiaoli
		DOI: 10.3969/j.issn.0258-8021.2018.06.006
		EEG coherence reflects the degree of spectral correlation between two-channels EEG signals, which can assess the connections between neurons in the brain. Combined the coherence method based on minimum variance distortionless response (MVDR) with kernel filter, a new method named smoothing minimum variance distortionless response (SMVDR) was proposed in this study. The simulation analysis indicated that the new method SMVDR provides more accuracy and better anti-noise performance in both narrow-band signals and broad-band signals. SMVDR was used to analyze the EEG coherence of 18 amnesic MCI (aMCI) and 13 normal controls of patients with diabetes between different brain regions in four frequency bands (delta, theta, alpha, beta). We observed a decrease in delta coherence and an increase in beta coherence in left temporal-right temporal region, and an increase in theta coherence in frontal-occipital region, and an increase in alpha coherence in both right temporal-occipital region and frontal-right temporal region in aMCI patients through statistical analysis. Correlation analysis between coherence values and MOCA scores shows that coherence values in alpha band and delta band and MOCA scores had a significant positive correlation in some specific channels, while coherence values in both theta band and beta band and MOCA scores had a significant negative correlation. The new method SMVDR can compute the coherence better between two-channels EEG signals. It is important in exploring the mechanism, early diagnosis and intervention treatment of aMCI.
		2018 Vol. 37 (6): 680-687 [Abstract] ( 411 ) HTML (1 KB) PDF (1160 KB) ( 257 )

	688	A Comparative Study on Dielectric Properties of Human Thyroid Normal and Tumor Tissues from 10 Hz to 100 MHz
		Shi Xuetao, ZhouYimin, Ji Zhenyu, Cai Zhanxiu, You Fusheng, Fu Feng, Dong Xiuzhen
		DOI: 10.3969/j.issn.0258-8021.2018.06.007
		The aim of this study was to investigate the dielectric properties of normal and tumor tissues of human thyroid, make clear the difference between the dielectric parameters of those tissues, and to provide a basis for the rapid identification of pathological thyroid tissues based on dielectric properties. The dielectric properties of 135 thyroid tissues were measured ex vivo at frequency range of 10 Hz to 100 MHz using the standardized measurement platform. Then the samples were classified according to the pathological results, and the differences in dielectric properties of each group were found after statistical analysis. The samples can be divided into three groups: normal (95 cases), malignant tumor (19 cases) and benign tumor (21 cases). All of these samples have shown an increase in conductivity with increasing frequency, and 2 main frequency dispersion regions of α and β were also observed at the frequency range of 10 Hz~100 MHz. Through the modeling analysis, the characteristic parameters of the dielectric properties, ρ_∞(Ω·cm), ρ₀(Ω·cm), f_cα(kHz), δ_α, f_cβ(MHz) and δ_β, were investigated. Normal thyroid tissue was 54.8±27.5, 385.6±3.3, 40.3±10.9, 0.61±0.04, 97.7±1.7 and 0.48±0.05, respectively. Benign tumor tissue was 118.3±8.5, 418.1±84.5, 26.3±13.1, 0.62±0.04, 3.5±0.8 and 0.56±0.06, respectively. Malignant tumor tissue was 67.7±5.1, 592.9±73.3, 10.2±2.6, 0.66±0.06, 6.37±4.1 and 0.36±0.07, respectively. Statistical analysis results indicated that the dielectric properties of normal thyroid tissue significantly different from that of malignant tumor tissue in parameters of ρ₀, f_cα,f_cβand δ_β, which was also significantly different from that of benign tumor tissue in parameters of ρ_∞, f_cα and δ_β. The significant difference was also detected in parameters of ρ_∞, ρ₀, f_cα, f_cβ and δ_β between malignant and benign thyroid tumor tissues. The dielectric properties of the tissue could reflect the changes in the microstructure to a certain extent. These findings provided evidence that there is a certain correlation between the dielectric properties and the nature as well as microstructure of the different types of thyroid tissues, as well as data supporting rapid differentiation and identification of pathological thyroid tissues.
		2018 Vol. 37 (6): 688-693 [Abstract] ( 312 ) HTML (1 KB) PDF (3256 KB) ( 273 )

	694	Research and Development of Virtual Surgery Simulation System with Haptic Feedback for Personalized TKA
		Wang Chunjuan, Liu Fei, Zhang Weijie, Ma Jianbing, Yao Shuxin, Xiao Lin, Cheng Yonghua, Xu Xianhui, Lian Qin, Wang Qi, Zhang Mingjiao
		DOI: 10.3969/j.issn.0258-8021.2018.06.008
		Physicians educated with traditional surgical training modes cannot meet the current surge in demand for knee joint patients. The aim of this work is to develop a simulation system based on virtual reality technology for the personalized knee replacement surgery. The virtual geometric models of knee joint and instruments were established by 3D reconstruction and CAD modeling technology respectively. We studied the voxelization algorithm and applied it to generate the virtual physical models. By using coordinate match method, the force model of the instrument was established between the device interface and the proxy. The AABB bounding box hierarchy tree was used as the collision detection model, we applied the overlapped intersect way to simulate deformation of tissue and the single point constraint force rendering to simulate the operation process. By setting up the hardware platform consisting of Phantom Omni and computer, and configuring the software environment CHAI 3D and OpenGL, the virtual knee replacement surgery simulation system was constructed. We compared the training time of the actual surgical operation of the professionals who experienced system with that of the control group, and used a questionnaire survey to make assessment of the system. Results showed that the system achieved the simulation of tissue deformation and force haptic feedback of drilling and osteotomy along the planned path in TKA, and the real-time performance of the system was excellent, the visual and haptic update frequency was maintained at 60 Hz and 1000 Hz respectively. The statistical results illustrated that there was a statistically significant difference (P=0.04) in the actual operation time between the two groups, the questionnaire results were all above the average excellent level (P<0.05). The constructed simulation system provides a safe, reliable and effective mode for medical training and preoperative drilling of TKA.
		2018 Vol. 37 (6): 694-704 [Abstract] ( 366 ) HTML (1 KB) PDF (8510 KB) ( 181 )

	705	Finite Element Simulation and Analysis of CMUT with Sacrificial Release Techniquies
		Guo Qing, Li Yan, Gao Shang, Zhang Peiyu
		DOI: 10.3969/j.issn.0258-8021.2018.06.009
		Capacitive micromachined ultrasonic transducers (CMUTs) have attracted great attentions in the field of ultrasonic imaging, because the potential of as alternative to piezoelectric ultrasonic sensors. At present, the analysis on CMUTs built by sacrificial layer release technology has less been involved in the analysis of the influence of the processing structures resulted from the releasing process. In this paper, the influence of sacrificial layer release process on CMUTs was analyzed. The finite element analysis software ANSYS was used to analyze the effect of release method on CMUTs’ performance. Moreover, the CMUT with 2.5 MHz was tested using Vibrometer. The results of both simulation and experiment are considerable consistency with an error of 3.9%. It was concluded that the ratio of length and width of membrane on property of CMUT varied much. The resonant frequency of CMUT changed less and the value of percent difference among four modes vary smoothly when the ratio was larger than 4. In addition, the thinner the thickness, the smaller the percentage difference. The effects of supporting walls on vibrating membrane, releasing holes and channels should be considered when building models on CMUT with sacrificial release technique. Finally, the CMUT with 2.5 MHz was tested using a Vibrometer. The results of both simulation and experiment were considerable consistency, which verified the validity of the model presented in this paper.
		2018 Vol. 37 (6): 705-713 [Abstract] ( 309 ) HTML (1 KB) PDF (6941 KB) ( 188 )

	714	Respiratory Signals Prediction based on Particle Swarm Optimization and Back Propagation Neural Networks
		Chang Panchun, Yang Jimin, Yang Juan, You Tao
		DOI: 10.3969/j.issn.0258-8021.2018.06.010
		Objective, Respiratory motion may cause the change of some organs and tissues, such as lung and liver tumors in radiation therapy, which may influence the treatment effect and increase the damage to normal tissues and organs. Hence, it is an essential work to estimate the real-time movement of target in radiation therapy. Method, BP-NN has been widely used in respiratory motion prediction due to its superior non-linear fitting capability. However, BP-NN is easy to fall into local minimum. Results, In this study a novel method using PSO to optimize the BP-NN was proposed to avoid its drawbacks and improve prediction accuracy. Firstly, the PSO method was used to select optimal initial weights and thresholds of neural network. Then, the optimal initial weights and thresholds was utilized to establish artificial neural network (ANN). Finally, the established PSO-NN was used to predict respiratory signals.The preliminary results of 11 patients demonstrate that the mean absolute error reduced from 0.24 to 0.18(25%) and the coefficient correlation increased from 0.82 to 0.86. Conclusion, The proposed method (PSO-NN)could reduce the risk of BP-NN falling into local optimum and has the ability of improving the prediction accuracy of BP-NN method.
		2018 Vol. 37 (6): 714-719 [Abstract] ( 343 ) HTML (1 KB) PDF (1090 KB) ( 301 )

	720	Biomechanical Analysis for Bioabsorbable Scaffolds
		Qi Yongxiang, Robert Ndondolay, Guan Zhixu, Tang Naijie, Nie Fangfang, Huo Yong
		DOI: 10.3969/j.issn.0258-8021.2018.05.011
		The purpose of the research is to analyze and evaluate the biomechanical capability of bioabsorbable scaffolds during processing and clinical application. Based onfinite element method, hexahedron mesh models were generated. Scaffolds and mock vessels were assumed to be isotropic incompressible materials, with constitutive models of elastoplasticity and hyperelasticity, respectively. Boundary and contact conditions were applied in the basis of practical processing and clinical applications. The Implicit solver of finite element software Abaqus were utilized to simulate the performance of three specific bioabsorbable scaffolds,B-2508, B-3018, and B-3528, under loadings for crimping, dilation, post-dilation and fatigue. The results were demonstrated. During crimping, the peak stress values for scaffold B-2508, B-3018, and B-3528 were 55.47, 55.47, and 50.51 MPa respectively. During dilation, the peak stress values were 64.10, 66.09, and 66.25 MPa respectively. During post-dilation, the peak stress values were 66.10, 65.85, and 67.85 MPa respectively. In high-cycle fatigue graph, all coordinates of mean stress-alternate stress for B-2508 and B-3018 were below the fatigue-critical line 1, while coordinates of a few nodes for B-3528 were above the fatigue-critical line 1. In summary, B-3528 was the best for crimping safety, B-2508 was the best for dilation safety, and B-3018 was the best for fatigue safety. Based on this study, an accurate simulation analytical method of biomechanical capability for bioabsorbable scaffolds were provided, and a theoretical instruction were conducted for product development of bioabsorbable-scaffold-like devices and precise clinical operations.
		2018 Vol. 37 (6): 720-730 [Abstract] ( 358 ) HTML (1 KB) PDF (14679 KB) ( 166 )

	731	Study on 3D Bioprinting of Liver Tissues Based on Coaxial Flow Technique
		Du Xianbin, Xu Mingen, Wang Ling, Zhou Yongyong
		DOI: 10.3969/j.issn.0258-8021.2018.06.012
		The 3D bioprinting provides a new technical possibility for the medical field, which can be widely used in the manufacture of artificial tissues and organs. However, the function and size of the artificial tissue are limited by the vascularization of the tissue. In this study, a coaxial flow extrusion system was used to fabricate hollow filaments in encapsulated liver cells, combined with a biological 3D printing system, so as to fabricate liver tissue with microchannel network by using a layer-by-layer approach. In the experiment, an integrated coaxial flow 3D bioprinting system was built firstly. The influence of material extrusion rate and material concentration was then studied on the size of hollow filaments and the velocity of wire. Subsequently, the biomimetic liver tissue containing multi-layer networks was printed with the liver cell line C3A as the material. Finally, the biomimetic liver tissue containing microchannels was divided into groups and cultured. The cell survival rate of hepatocytes after 24h, 48h and 72h of culture was detected by live-dead cell staining in perfusion group and non-perfusion group. The experimental results showed that in the organs using 3D printing technology based on coaxial flow technique, the hollow filaments could be effectively fused, and the 3D microchannel network inside the support was complete. Besides, the proportion of damaged hepatocyte was less in the printing process, and the survival rate of hepatocytes was over 90% in hollow filaments. Furthermore, there was a significant difference in the cell survival rate between the perfusion group and the non-perfusion group after culture for 72h, suggesting that microchannel perfusion could promote the exchange of material within the tissue and increase the survival rate of liver cells around the microchannel. The printing technology and perfusion system presented in this paper provide a new study approach for the vascularization and culture of artificial tissues.
		2018 Vol. 37 (6): 731-738 [Abstract] ( 317 ) HTML (1 KB) PDF (11830 KB) ( 101 )

Reviews

	739	A Review of Segmentation of Pulmonary Airway in Lung CT Scans
		Duan Huihong, Gong Jing, Wang Lijia, Li Xinyu, Nie Shengdong
		DOI: 10.3969/j.issn.0258-8021.2018.06.013
		Pulmonary airway is the only access between the human body and the external environment, therefore the anatomy information of pulmonary airway is helpful for diagnosis of respiratory system disease. Computed tomography (CT) is one of the main methods for respiratory disease diagnosis. However, due to the large amounts of patients and images, manual reading of CT images is tedious and time-consuming. The automatic segmentation and extraction of pulmonary airway tree is the precondition of automatic analysis and computer-assisted diagnosis. Hence, according to the research progress of segmentation of pulmonary airway in recent years, we first introduced the background and the meaning of the airway segmentation. Then we analyzed the traditional methods, the method based on tube structure detection and machine learning, and the problem they met. Finally, we proposed that integrating the step of segmentation and leakage limitation can improve the accuracy and the number of branch, which means segmenting as many airways as possible at first and then eliminating the leakage.
		2018 Vol. 37 (6): 739-748 [Abstract] ( 516 ) HTML (1 KB) PDF (4376 KB) ( 584 )

	749	Research Advancements on the Variation and Prediction of Brain Control Performance for Brain-Computer Interfaces (BCIs)
		Zheng Yufu, Xu Minpeng, Ming Dong
		DOI: 10.3969/j.issn.0258-8021.2018.06.014
		Brain-computer interface (BCI) technology has attracted significant attention over recent decades, and has made remarkable progress. However, Brain control performance variations across and even within subjects severely degrade the reliability of BCI systems, which has become one of the most important challenges in the real-life application of brain-computer interfaces (BCIs). Understanding the underlying causes is essential to improve the stability of BCI systems, which could be approached by predicting BCI performance. Reliable prediction on individual’s ability to control BCIs can distinguish BCI illiteracy and can help eliminate the performance differences among users; while reliable prediction across trials for the same subject can help improve the whole BCI performance. This paper reviewed recent studies on the prediction of BCI performance, which laid emphasis on the BCI illiteracy and the causes of the BCI performance variation. It also provided some possible directions for future studies.
		2018 Vol. 37 (6): 749-755 [Abstract] ( 341 ) HTML (1 KB) PDF (789 KB) ( 510 )

	756	Research Progress of Protein Adsorption on Hydroxyapatite Surface
		Fu Yakang, Zhang Kehong, Weng Jie, Liu Yaowen
		DOI: 10.3969/j.issn.0258-8021.2018.06.015
		This paper summarized the research progress of protein adsorption on hydroxyapatite (HAP) in the past 30 years, including adsorption mechanism, crucial factors and related application exploration. The adsorption mechanisms include electrostatic attraction, hydrogen bond and van der Waals force. The crucial factors include exposing crystal faces of HAP, specific surface area, concentration of PO₄^3- and Ca²⁺, pH and temperature in the adsorption environment. In addition, this article introduced the application research of protein adsorption on HAP in the field of drug sustained-release carrier and bone tissue engineering scaffold. This review aimed to provide comprehensive introduction of advances for exploring the biological activity and special adsorption function design of HAP.
		2018 Vol. 37 (6): 756-764 [Abstract] ( 619 ) HTML (1 KB) PDF (3596 KB) ( 658 )

Communications

	765	The Enrichment of Tumor Cells by Microporous Membrane with Different Pore Size
		Zhao Xiao, Liu Wenjuan, Li Lei, Yang Shuiqing, Yu Chunhong, Cui Ningxuan, Wang Ying, Yi Zongchun
		DOI: 10.3969/j.issn.0258-8021.2018.06.016

		2018 Vol. 37 (6): 765-768 [Abstract] ( 352 ) HTML (1 KB) PDF (828 KB) ( 359 )

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