Chinese Journal of Biomedical Engineering

Home

Editorial Board

Download

Quick Search

Office Online

Current Issue

Download Articles

Download

Links

2022 Vol. 41, No. 6 Published: 2022-12-20

Reviews Communications Regular Papers

Regular Papers

	641	A Novel KPCA-Based RVM Model for Human Activity Classification
		Wu Jianning, Lin Qiuting, Wu Bin
		DOI: 10.3969/j.issn.0258-8021.2022.06.001
		In order to effectively improve the generalization ability in human activity classification task based on small sample data size, a novel human activity classification model was constructed by using the hybrid technique of kernel principal component (KPCA), and relevance vector machine (RVM) was proposed in this paper. The proposed technique was able to take the advantage of kernel function for the combination of KPCA with RVM, that is, the nonlinear gait features containing more human activity discriminative information hidden in the transformed higher dimensional feature space could be exploited by KPCA, which greatly contributed to solve the sparse probability distribution of human activity discrimination by Bayesian learning algorithm in RVM. This significantly improved the generalization performance of human activity classification. A public UCI human activity recognition (HAR) dataset with wearable sensor data from a smartphone was selected to evaluate the feasibility of our proposed model. In the experiment, all 10 299 of sample data were obtained from the collected data including a total of 30 subjects with six different human activity patterns. The cross-validation with 10 times was used to train and test the model. The experimental results showed that our proposed model could reach the best accuracy of 96% when ten relevance vectors were available, which were 5.4% and 3.6% more than KPCA-SVM and CNN-LSTM deep learning model, respectively. This suggested that our model had a superior ability to extract more nonlinear features associated with human activity change and to learn the sparse probability distribution of RVM. In conclusion, the proposed technique could accurately detect a certain human activity pattern based on the small sample size, which would provide a new idea and approach to exactly identify human activity.
		2022 Vol. 41 (6): 641-649 [Abstract] ( 387 ) HTML (1 KB) PDF (3248 KB) ( 326 )

	650	Wearable Sleep Apnea Detection Method Based on Multi-Task and Multi-AttentionResidualShrinkage Convolutional Neural Network
		Shen Qi, Wei Keming, Liu Guanzheng
		DOI: 10.3969/j.issn.0258-8021.2022.06.002
		Sleep apnea syndrome (sleep apnea syndrome, SAS) is a common chronic respiratory disorder, often accompanied by a variety of complications, which seriously plagues human health. The SAS detection method of photoplethysmography (PPG) based on wearable devices has attracted extensive attention because of the advantages of low cost, low load and easy to wear. Aiming at the problem of greater interference of PPG signals based on wearable devices, a sleep apnea detection method based on multi-task multi-attention residual shrinkage convolutional neural network was proposed in this study. First of all, 92 wrist PPG sleep data were collected using smart bracelet devices. Next, a residual multi-attention mechanism convolution block was designed, which efficiently integrated the dual important features of the network in the time domain and the channel domain. Then, the residual shrinkage convolution block was introduced to suppress the signal noise and the redundant features of the network. Through the combination of these two blocks, a backbone network for feature extraction was constructed. The results showed that the accuracy, sensitivity, and specificity of the segment detection achieved 81.82%, 70.27%, and 85.81%, respectively; and the accuracy, sensitivity, and specificity of individual detection achieved 95.65%, 88.89%, and 97.30%, respectively. Compared with peers, the proposed method displayed better performance and was able to integrate to the wearable devices for sleep apnea syndrome detection.
		2022 Vol. 41 (6): 650-662 [Abstract] ( 322 ) HTML (1 KB) PDF (4293 KB) ( 385 )

	663	Image Segmentation of Polycystic Ovary Based on Improved U-Net Network
		Wu Liping, Duan Xiaopeng, Ma Yuliang, Zhang Jianhai
		DOI: 10.3969/j.issn.0258-8021.2022.06.003
		Polycystic ovary syndrome (PCOS) is a disease that seriously endangers women′s health. Aiming to solve the problem of low contrast in the targeted area and high background noise in PCOS ultrasound images, a segmentation method of polycystic ovary images based on improved U-Net network was proposed in this paper. Firstly, the PCOS images were preprocessed to reduce the influence of speckle noise and shadows; then, redundant low-frequency feature maps were reduced by octave convolution module and feature fusion is performed; then, the hierarchical residual skip connection module was used to compensate for U-Net semantic gap between encoder and decoder; secondly, experiments were performed using PCOS ultrasound image dataset; finally, validation experiments were performed using ISIC2018, a public dataset containing 2 594 skin lesion images. The proposed method achieved a segmentation accuracy of 88.42% on the PCOS ultrasound image dataset, which was 4.24% higher than that of U-Net; and achieved a segmentation accuracy of 97.5% on the ISIC2018 dataset. The experimental results showed that the proposed method not only improved the segmentation of polycystic ovarian vesicles, but also had better performance in terms of robustness, which could also be referred to other medical image segmentation fields.
		2022 Vol. 41 (6): 663-671 [Abstract] ( 362 ) HTML (1 KB) PDF (7964 KB) ( 244 )

	672	Application of Improved NASNet Algorithm in Breast Ultrasound Diagnosis
		Yi Sanli, She Furong, Yang Xuelian, Chen Dong, Luo Xiaomao
		DOI: 10.3969/j.issn.0258-8021.2022.06.004
		Ultrasound image is of great significance in clinical diagnosis of breast diseases, however, the resolution of the breast ultrasound image is low, and the sample size is small. Although NASNet is suitable for small sample data, it requires many parameters that makes it difficult to train. This paper proposed an improved NASNet classification algorithm to distinguish the benign and malignant breast masses. Firstly, NASNet is pre-trained on Imagenet by transfer learning technology, and the learned features were directly used for benign and malignant tumor recognition on breast ultrasound images, which saved the cost of calculation and improved the accuracy of the calculation. Then, to enhance the ability of the network to extract ultrasonic image features and make the network lightweight, we deeply integrated deep separable convolution into NASNet to construct a large-scale network. Finally, to enhance feature weights that are more relevant to the disease and further enhance the extraction ability of high-order feature information, we added an SE module to screen the channel features that account for more weight in ultrasonic images. To verify the algorithm, we used the training method of 5-fold cross-validations based on the experiments of local hospital data sets and public data sets and compare the algorithm with the widely used classification algorithm. There were 1 350 ultrasound images in the local hospital datasets and 895 ultrasound images in the two public datasets. Based on the data of local hospitals, Acc, Sen, and F1 were 97.52%. The Acc, Sen, and F1 of experiments with public data set as training set and verification set and local hospital data set as test set were 96.31%, 96.31%, and 96.39% respectively. The Acc, Sen, and F1 of the mixed data experiment based on local hospital data and public data were 98.27%. The results showed that the improved algorithm had advantages over other algorithms and was more suitable for the classification of benign and malignant tumors with a small amount of breast ultrasound images.
		2022 Vol. 41 (6): 672-679 [Abstract] ( 188 ) HTML (1 KB) PDF (3720 KB) ( 282 )

	680	The Similarity Evaluation Method of Wireless Capsule Endoscopy Images Based on ImprovedPatchMatch Algorithm
		Tian Hao, Lu Heng, Pan Ning, Hu Huaifei, Liu Haihua
		DOI: 10.3969/j.issn.0258-8021.2022.06.005
		The examination of every single wireless capsule endoscopy (WCE) produces a large quantity of image sequence data (i.e., 50 000~80 000 images), which creates big challenges in clinical practice and can lead to the difficulties in medical diagnosis. It is, therefore, of great significance to advance methods to quantify the abstract-level visual similarity between the WCE images and screen out redundant images in the WCE sequence, so that both the efficiency and the effectiveness of image diagnosis can be improved. In the current study, we proposed a PatchMatch algorithm based on the spatial constraint scheme to capture and evaluate the similarity between the WCE images. This novel method improved the conventional PatchMatch procedure through the achievement of matching the local image patches based on the spatial constraint measures, adding the offset position constraint, restricting the initial matching region, and providing different matching search regions for the image patch to be coordinated. Next, the image-level ensemble descriptor was built by combining the position and texture attribute information of matched image patches. Finally, the similarity of two WCE images was evaluated by a joint probability between ensemble descriptors. A recurring screening experiments of 10 cases of WCE image sequences consisting of more than 5000 WCE imageries were conducted by employing different descriptor operators (such as: SIFT and Hog). The results showed that the precision, recall and F-measure of experiments were 93.73 %, 95.44 % and 94.77 % respectively, indicating the proposed algorithm is effective in analyzing the similarity between the WCE images.
		2022 Vol. 41 (6): 680-690 [Abstract] ( 192 ) HTML (1 KB) PDF (13473 KB) ( 92 )

	691	Detection of Dorsal Hand Vein Based on Improved YOLO Nano and Embedded System
		Zhao Dechun, Tian Yuanyuan, Chen Huan, Zhao Zehan, Chen Yi, Yuan Yang
		DOI: 10.3969/j.issn.0258-8021.2022.06.006
		As the most fundamental medical means, venipuncture remains challenging for medical workers. This paper proposed a vein detection and location method for near infrared image. Firstly, an embedded system based on the near-infrared was designed, by which the vein images of both left and right dorsal hand from 43 subjects were captured to finally build a database composed of 325 dorsal hand vein images after preprocessing. Secondly, YOLO Nano algorithm was improved by trimming the network structure to reduce the model size and the output scale to adapt to the size of the detection target. The spatial pyramid pooling structure was introduced to improve the detection accuracy for its strong detail feature description and efficient feature computation. The database was divided into training set and test set in a proportion of 7∶3 and labeled and expanded. After tested on our embedded system, the results showed that the size of the improved YOLO Nano was reduced by 15%, while the average precision (AP) was increased from 91.68% to 93.23% and the detection time reached 529 ms, reduced by 22% compared to YOLO Nano. The improved YOLO Nano outperformed the original YOLO Nano in terms of both detection speed and accuracy, which realized the detection of puncturable veins.
		2022 Vol. 41 (6): 691-698 [Abstract] ( 236 ) HTML (1 KB) PDF (5988 KB) ( 298 )

	699	Contour Extraction Method of Colony Images Based on Dynamic Synaptic Neural Network
		Cai Zhefei, Fan Yingle, Wu Wei
		DOI: 10.3969/j.issn.0258-8021.2022.06.007
		The accuracy of colony image contour extraction is of great significance for microbial colony morphology and feature analysis. Based on the dynamic synaptic neural network, we constructed an image contour extraction method for colonies. Firstly, simulating the photoelectric conversion process of retinal, proposed an adaptive adjustment model of receptive field scale based on light intensity. Secondly, a LIF model combining the lateral regulation of electrical synapse and chemical synapse was constructed. Using neuron membrane potential, spatial distribution relationship and response time difference to adjust the strength of electrical synapse and chemical synapse, the colony edge sensitive image with or without lateral regulation was obtained. Finally, compared the difference of neurons response time with or without lateral regulation, and using STDP rules to dynamically update the weight of neuronal synaptic, so as to adjust the contour details of colony images. The adjusted contour response with the primary contour response were combined to obtain the final contour information of colony images. Taking 40 colony images collected in the laboratory as research objects, and selecting edge confidence BIdx, average reconstructed similarity MSSIM and comprehensive index EIdx as the evaluation indexes. The results showed that the contours obtained by this study was more accurate, continuous and has less noise. BIdx and MSSIM were 0.651 4±0.056 5 and 0.831 8±0.026 1, respectively. Meanwhile, EIdx was 0.765 7±0.027 4, which was significantly higher (P<0.01) than that of biological vision based methods of OS, BAR and LS. The dynamic synaptic neural network constructed in this paper is suitable for image contour extraction with rich detailed features such as colonies and may provide a new way for the research and application of neural computing models that integrate the biological vision mechanism.
		2022 Vol. 41 (6): 699-707 [Abstract] ( 185 ) HTML (1 KB) PDF (4449 KB) ( 256 )

	708	Diagnosis of Alzheimer′s Disease with Sparse Association of Whole Brain Anatomical Markers
		Zheng Fei, Tang Qiling, Liu Ruxuan, Zhang Meiling, Ge Wei
		DOI: 10.3969/j.issn.0258-8021.2022.06.008
		Alzheimer′s disease (AD) is a complex neurodegenerative disease with progressive impairment of memory and other mental functions, which is the main cause of death in the elderly. How to make an accurate diagnosis of AD is crucial. Existing research methods ignore the relationship between potential features. However, according to biological verification, it is known that many brain regions in the human brain are interconnected anatomically and functionally. Therefore, focusing on the characteristic correlation of different brain regions is beneficial to improve the detection performance of brain cognitive diseases. In this paper, we propose a data-driven method for automatic recognition of anatomical markers in whole brain structural magnetic resonance imaging (sMRI), extract block features based on anatomical points, deeply fuse the features of each block using global correlation, and realize the correlation of various brain regions by calculating the interaction between blocks. Secondly, according to the difference of correlation degree between blocks, thresholding processing is carried out, and redundant information is removed by sparse correlation module to further improve the distinguishing ability of features. Finally, classification model is constructed by using the deep features after sparse to predict Alzheimer′s disease individuals. The experimental results showed that the accuracy and sensitivity of the method reached0.936 8 and 0.921 1, respectively, when the ADNI-1 dataset containing 198 AD patients and 224 healthy subjects were trained, and the ADNI-2 dataset containing 152 AD and 196 healthy subjects were tested. The proposed method takes into account the relationship between blocks and the difference of correlation degree, and can diagnose Alzheimer′s disease more effectively.
		2022 Vol. 41 (6): 708-716 [Abstract] ( 174 ) HTML (1 KB) PDF (4815 KB) ( 216 )

	717	Effects of a Carbon Fiber Reinforced Polyvinyl Alcohol Conductive Hydrogel on Macrophages and Vascular Endothelial Cells
		Xiao Bo, Xu Shilin, Wu Fengxin, Liu Xuanxin, Huang Yuguang, Xu Haiyan
		DOI: 10.3969/j.issn.0258-8021.2022.06.009
		The effects of hydrogel dressings on the immune cells and vascular endothelial cells are closely related to wound healing. This work is aimed to investigate the effects of a carbon fiber reinforced polyvinyl alcohol conductive hydrogel (PVA/CFs) on inflammatory response in mouse macrophages and the formation of adherent junctions between HUVEC cells. The above two kinds of cells were cultured on PVA and PVA/CFs with CF content of 1.5%, 3% and 5%, respectively. The expression of TNF-α, IL-1β and IL-6 in inflammatory response genes of macrophages (RAW264.7) was detected by real-time PCR. The cytokins secretion of TNF-α, IL-1β and IL-6 were tested by ELISA kits. The growth and proliferation of human umbilical vein endothelial cells (HUVEC) and RAW264.7 on PVA/CFs were detected by CCK-8 assay. The protein expression of VE-Cadherin in HUVEC was determined by laser scanning confocal microscopy and western blot assay. All quantitative detection experiments were set at least three replicates (n≥3). PVA/CFs did not significantly change the gene expression levels of TNF-α, IL-1β and IL-6 and the cytokins secretion levels in RAW264.7 cells compared with PVA. All groups were no significant difference (P>0.05). However, the secretion of TNF-α, IL-1β and IL-6 in RAW264.7 cells treated with LPS was increased by 2.35, 4.27 and 27.1 folds of that in the control group, respectively. The expression of VE-cadherin in HUVEC cells on PVA/CFs was increased, suggesting that PVA/CFs could strengthen adherent junctions of endothelial cells. In addition, PVA/CFs could support the growth and proliferation of RAW264.7 and HUVEC cells, and there was no significant difference in cell viability between PVA/CFs and PVA groups (P>0.05). PVA/CFs not only did not cause any severe inflammatory responses of macrophages, but also supported the growth and proliferation of macrophages and endothelial cells, and promoted the formation of integrated adherent junctions between endothelial cells.
		2022 Vol. 41 (6): 717-723 [Abstract] ( 172 ) HTML (1 KB) PDF (2019 KB) ( 283 )

Reviews

	724	A Review on Generative Adversarial Networks in Medical Image
		Yu Miao, Xu Zhenghua
		DOI: 10.3969/j.issn.0258-8021.2022.06.010
		As an important means to assist the detection of diseases and make better diagnostic decisions, the definition of medical image is of great significance for clinical treatment. The unique confrontation training idea of generative adversarial network (GAN) can generate high-quality samples. The success in the field of computer vision makes GAN a bright prospect. In this article, the application of GAN in medical image denoising was reviewed. Firstly, the basic theory, advantages and disadvantages of GAN were introduced. Then, the derivation model of GAN for medical image denoising was introduced in detail, and various loss functions that can help improve the denoising performance of GAN for medical images were summarized. And other deep learning frameworks, which can be nested into the GAN model and play an auxiliary role in medical image denoising, were presented as well. The methods to improve the performance of GAN network for medical image denoising were summarized. Finally, the application prospects, challenges and possible future research directions of GAN in medical image denoising were discussed.
		2022 Vol. 41 (6): 724-731 [Abstract] ( 394 ) HTML (1 KB) PDF (1050 KB) ( 1185 )

	732	Advances in the Application of Wearable Noninvasive Glucose Sensors in Diabetes Management
		Shao Shengqi, Fan Kai, Li Dujuan, Liu Chaoran, Wang Gaofeng
		DOI: 10.3969/j.issn.0258-8021.2022.06.011
		Currently great efforts have been made in the development of new wearable electrochemical sensors worldwide. Different from traditional invasive monitoring strategies, such sensors can continuously monitor biomarkers in the body fluids in non-invasive ways, and have significant application potentials in the disease prevention, diagnosis and management. In the global trend of increasing diabetes, non-invasive continuous monitoring of blood glucose is very important in the management of diabetes. This paper reviews the application of wearable non-invasive glucose sensors in diabetes management in recent years, briefly introduces the development and principle of several electrochemical glucose sensors based on different body fluids (sweat, tears, saliva). The advantages of the wearable sensor in diabetes management and the challenges and problems in the research and development of the sensor are described. At the end of the paper, the development prospective, commercialization trend and potential of wearable noninvasive glucose sensor in the future market are discussed.
		2022 Vol. 41 (6): 732-743 [Abstract] ( 404 ) HTML (1 KB) PDF (1580 KB) ( 1184 )

	744	Machine Learning Methods for Prediction of Dose Distribution and Response to Treatmentin Tumor Precision Radiotherapy: A Review
		Liu Guocai, Gu Dongdong, Liu Jinguang
		DOI: 10.3969/j.issn.0258-8021.2022.06.012
		Intensity modulated radiation therapy (IMRT) is a main technology for tumor treatment in clinics. In order to design a clinically acceptable and executable IMRT plan, key factors including radiotherapy dose calculation, dose distribution prediction and optimization evaluation are required to carefully consider. Meanwhile, it also needs to predict and evaluate the outcome, toxicity and side effects of radiotherapy and chemotherapy. This article reviewed machine learning methods based on the images for dose distribution prediction and responses to the tumor radiotherapy and chemotherapy, including deep learning methods for dose prediction as well as deep learning, radiomics, logistic regression methods for outcome prediction of IMRT, stereotactic body radiotherapy (SBRT), volumetric arc modulated radiation therapy (VMAT). Finally, the future research directions and research contents were proposed.
		2022 Vol. 41 (6): 744-758 [Abstract] ( 261 ) HTML (1 KB) PDF (954 KB) ( 775 )

Communications

	759	Post-Stroke Lower Limb Dyskinesia Assessment Based on sEMG Features
		Lu Xiao, Zhang Wentong, Su Panpan, Lu Qian, Zhao Kunkun, Yang Junjie, He Chuan
		DOI: 10.3969/j.issn.0258-8021.2022.06.013

		2022 Vol. 41 (6): 759-763 [Abstract] ( 181 ) HTML (1 KB) PDF (748 KB) ( 218 )

	764	Algorithms to Automatically Count Splitting and Merging Number of Spatial Extending PeripheralNerve Internal Fascicular Groups
		Zhong Yingchun, Yi Xiaohong, Huang Jianhao, Qi Jian, Zhu Shuang, Luo Peng
		DOI: 10.3969/j.issn.0258-8021.2022.06.014

		2022 Vol. 41 (6): 764-768 [Abstract] ( 161 ) HTML (1 KB) PDF (3364 KB) ( 132 )

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