Chinese Journal of Biomedical Engineering

Home

Editorial Board

Download

Quick Search

Office Online

Current Issue

Download Articles

Download

Links

2024 Vol. 43, No. 4 Published: 2024-08-20

Reviews Communications Regular Papers

	0	CONTENTS


		2024 Vol. 43 (4): 0-0 [Abstract] ( 45 ) HTML (1 KB) PDF (251 KB) ( 35 )

Regular Papers

	385	Real-Time Target Detection of Abnormal Regions in Gastrointestinal Endoscopy Based on GE-YOLO
		Fan Shanhui, Lai Jintao, Wei Shangguang, Wei Kaihua, Fan Yihong, Lv Bin, Li Lihua
		DOI: 10.3969/j.issn.0258-8021.2024.04.001
		Gastrointestinal endoscopy is a common clinical examination in early diagnosis and monitoring of gastrointestinal diseases. However, this examination needs to be operated by a professional doctor to identify lesions in real-time, it is extremely dependent on the doctor′s experience which is subjective and easy to cause missed and/or false detection. In this study, GE-YOLO, a real-time detection method for abnormal object under digestive endoscopy based on improved YOLOv7-tiny, was proposed. Using YOLOv7-tiny as the basic framework, the backbone feature extraction network was constructed by using two different feature extraction modules (C3 module and P-ELAN module) to improve the feature extraction capability of the network; and then the coordinate convolution (CoordConv) was used to replace the normal convolution in the up-sampling, which made the model localize the lesion more accurately; furthermore, partial convolution (PConv) was applied to replace the 3×3 convolution in the feature extraction module, which not only guarantee the model detection performance, but also greatly reduced the computation cost and parameter number, and improved the model detection speed; finally, a joint loss function based on IoU and normalized Wasserstein distance was used to make the model more sensitive to small lesions. This model was trained and tested on the labeled images (4 172 in total) in Kvasir-Capsule dataset. The average precision, recall and F1-score of GE-YOLO was 94.2%, 97.2% and 0.957, respectively, and the detection speed was 60 frames per second, which had an improvement of 2.8% in precision, 12.0% in recall and 0.075 in F1-score compared with the results achieved by YOLOv7-tiny. The promising results demonstrated this proposed method can achieve high-precision real-time diagnosis of digestive tract lesions, and is expected to be deployed in clinical endoscopy equipment to provide real-time assistance for doctors during the examination to improve the diagnostic efficiency, which has momentous clinical value and research significance.
		2024 Vol. 43 (4): 385-398 [Abstract] ( 144 ) HTML (1 KB) PDF (16639 KB) ( 62 )

	399	MI-EEG Classification Based on Ensemble Tensor Domain Adaptation
		Gao Yunyuan, Xue Yunfeng, Zhang Congrui, Gao Jan
		DOI: 10.3969/j.issn.0258-8021.2024.04.002
		In clinical applications, EEG signals have been facing problems including high acquisition cost and large differences between users, which restrict the development of motor imaging based on EEG signals. Aiming at the task of cross-subject MI-EEG recognition, a transfer learning method based on ensemble tensor domain adaptation was proposed in this paper. Firstly, the improved Euclidean alignment method was used to co-align the multidimensional EEG data to eliminate the edge distribution shift of the original data. Secondly, an improved joint distribution adaptation method based on tensor subspace was proposed, which obtained different classes of mapping subspaces and performed label prediction of target domain samples. In this paper, experiments were carried out on BCI datasets of 200 samples for 7 people and 144 samples for 9 people, which proved that the proposed method had good performance in cross-domain classification recognition with average accuracy 82.18 % and 76.45 %. The effect of each part of the method was also visually verified, which showed the effectiveness of the ensemble method on cross-domain problems.
		2024 Vol. 43 (4): 399-407 [Abstract] ( 120 ) HTML (1 KB) PDF (3470 KB) ( 71 )

	408	Active Iterative Optimization of Leukocyte Image Classification Model with Fused Attention
		Jiang Shuying, Li Zhiming, Mo Xian, Sun Ang, Zhang Junran
		DOI: 10.3969/j.issn.0258-8021.2024.04.003
		The number and structural characteristics of leukocytes contain important information about human health, and the counting of different types of leukocytes can provide important basis for the early treatment of many diseases. However, the high cost of collecting and labeling leukocyte data integration and the small number of available leukocyte datasets currently pose challenges for automatic computer-aided leukocyte classification. To address these challenges, an active iterative optimization leukocyte image classification model incorporating attention was proposed in this paper. By attaching a LossNet network for active learning to the ResNet18 backbone network, the most representative samples were selected from a large number of unlabeled samples for labeling, reducing the amount of samples that need to be manually labeled. Meanwhile, to cope with the impact of inter-class imbalance in the leukocyte dataset on active learning, this paper added an active iterative augmentation module to select difficult samples in the training process for data augmentation containing random factors, which formed a two-way information interaction from the bottom up and enhanced the adaptability of the model to imbalanced datasets. Finally, after comparing three attention modules, this paper chose to incorporate the CBAM attention module to enhance the model′s focus on the leukocyte feature regions and improve the model′s performance. In this study, the Raabin-WBC dataset containing 14514 leukocyte microscopy images was used for method validation, and the experimental results showed that the classification accuracy of the model proposed in this paper reached 92.35%, 93.64%, and 94.86% when using 28%, 37%, and 52% samples of the training set, respectively, which was 5.14%, 9.24%, and 2.37% higher than the original ResNet18, respectively, and the model greatly reduced the labeling cost of leukocyte dataset, showing wide application prospectives in medical datasets that was lack of labelling.
		2024 Vol. 43 (4): 408-418 [Abstract] ( 87 ) HTML (1 KB) PDF (5100 KB) ( 110 )

	419	Improved Residual Network Classification of Breast Cancer Pathological Images Based on Multi-Scale Feature Fusion
		Zhuang Jianjun, Wu Xiaohui, Jing Shenghua, Meng Dongdong
		DOI: 10.3969/j.issn.0258-8021.2024.04.004
		In view of the extremely inadequate extraction of pathological features from existing models and the unbalanced number of various types of open breast cancer data sets, the research on multi-classification of breast cancer pathological images is still challenging. In this paper, an improved residual network multi-classification method of breast cancer pathological images based on multi-scale feature fusion was proposed. Firstly, based on ResNet101 residual network, the CBAM attention module was inserted into each residual block. Next, in order to optimize feature extraction, horizontal and vertical multi-scale feature fusion was integrated into the residual network. Finally, the focus loss function was introduced to solve the problem of unbalanced data distribution. Validated by the training of 1582 pathology images with mixed magnifications on BreakHis public dataset, the proposed improved residual network achieved a recognition accuracy of 94.4% on eight classifications of breast cancer pathology images, which was 2.8% better than the original model and outperforms most of the existing publicly available deep learning models. The proposed model provided a more effective method for screening, diagnosis and pathological classification of female breast cancer.
		2024 Vol. 43 (4): 419-428 [Abstract] ( 95 ) HTML (1 KB) PDF (3179 KB) ( 109 )

	429	Research on Chest CT Image Classification Method Combining Attention Mechanism and Lightweight Convolutional Neural Network
		Wang Wei, Xu Yuyan, Wang Xin, Huang Wendi, Yuan Ping
		DOI: 10.3969/j.issn.0258-8021.2024.04.005
		CT images of the same disease type can also show differences due to the different severity of the patient′s disease. At present, main clinical diagnosis methods rely on personal ability and past experience of doctors, and the objectivity needs to be enhanced and the efficiency needs to be improved. In view of these problems, we proposed a CT classification network with attention mechanism-parallel lightweight convolutional neural network for CT classification (PC-CTNet). This network mainly consisted of parallel branch channel shuffle (PCS) module and deep-wise efficient shortcut connection (DES) module. PCS module adopted double branches, fused the features under the multi-scale receptive field. DES module used convolution and efficient channel attention to extract effective deep inter-class differentiation information, and alleviated gradient disappearance by shortcut connection. Experiments were conducted on two chest CT datasets, and the results showed that the classification accuracy of the PC-CTNet model reached 98.46% on the collected dataset with 5 988 CT images in different sizes, and 98.75% on the open-source datasets with 194 922 CT images. The performance indicators of PC-CTNet were close to the existing chest CT classification network, and its parameter and computational complexity was about 0.32 M and 75.58 M, respectively, which was 10.17% and 3.21% of the chest CT classification network in the experimental comparison. The proposed network has higher parameter and computational efficiency, can effectively assist doctors in diagnosis and improve diagnostic efficiency and objectivity.
		2024 Vol. 43 (4): 429-437 [Abstract] ( 95 ) HTML (1 KB) PDF (4184 KB) ( 92 )

	438	Stripe Pooling and Vessel-Constraint Network for Fundus Image Artery/Vein Classification
		Xiao Zhitao, Peng Xinwen, Liu Yanbei, Geng Lei, Zhang Fang, Wang Wen
		DOI: 10.3969/j.issn.0258-8021.2024.04.006
		The ratio of retinal artery to vein diameter is a prerequisite for quantitative analysis of chronic diseases, such as diabetes and hypertension, and is an important risk indicator for many cardiovascular diseases. With the development of deep learning technology, many methods based on convolution neural network have made great progress in the classification of fundus images based on their ability to capture high-level semantics. However, most of the methods are based on superimposed local convolution and pooling operation, which is difficult to be well applied to striped retinal blood vessel segmentation. In this paper, in order to extract the features of retinal blood vessels in the shape of stripes more effectively, we introduced stripe pooling to capture the long-distance dependence of spatial pixels. Taking into account the complex characteristics of arteriovenous interleaving and further combining with spatial pyramid pooling, a new mixed pooling technology was proposed to expand the receptive field and learning context information of the neural network. On the other hand, considering that the proportion of blood vessel and non-blood vessel distribution in the fundus image is extremely unbalanced, this paper introduced a blood vessel enhancement module, which used the information of blood vessel distribution and the information of blood vessel edge constrained by Gaussian kernel function as weights to correct the arteriovenous features and suppress the background features, thus solving the problem of the imbalance between blood vessel and background distribution. Experiments on three internationally available datasets, DRIVE, LES, and HRF, containing 40, 22, and 45 color fundus images respectively, showed that the proposed algorithm achieved results of 0.955, 0.946, and 0.967 in term of BACC scores, which verified that the method combining strip pooling and vascular enhancement effectively solved the problems of complex arteriovenous interlacing and category imbalance in fundus images, achieving accurate classification of retinal arteriovenous malformations, holding a high application value.
		2024 Vol. 43 (4): 438-446 [Abstract] ( 62 ) HTML (1 KB) PDF (4744 KB) ( 35 )

	447	Atrial Fibrillation Detection and ECG Heartbeat Classification Algorithm Based on Inception Module and CNN-BiLSTM
		Zhang Yao, Liu Yanjun, Liu Lei
		DOI: 10.3969/j.issn.0258-8021.2024.04.007
		Automatic ECG classification technology is an important auxiliary diagnostic method for arrhythmia. In order to improve the accuracy of abnormal dynamic ECG heartbeat extraction, an ECG beat classification algorithm based on Inception module and CNN-BiLSTM was proposed in this paper. First, the ECG signal was divided into heartbeat segments with the length of 1000 sampling points. Next, 3 different scales heartbeats were extracted by using the Inception module. The ECG features were further extracted through a 4-layer one-dimensional convolutional neural network (CNN) and a 2-layer bidirectional long short-term memory neural network (BiLSTM). At last, a 1-layer fully connected network and a softmax function were used to reduce the dimension of feature and classify the heartbeat. To improve the classification accuracy, a wavelet denoising technique was used to reduce the noise of the raw data. The data provided by the PhysioNet/Computing in Cardiology Challenge 2017 database were used in experiments. After preprocessed, 60,000 heartbeat samples were selected for classification, and the accuracy (Acc) and F1 score (F1-score) were used as the main evaluation criteria to evaluate the performance of the model. Results showed that the established model had an Acc of 91.38 % for the three types of heartbeats (normal, atrial fibrillation, and others) and F1-score was 91.27%, which was 4.77% and 4.59% higher than that of the combined model using only CNN-BiLSTM (Acc of 86.61%, F1-score of 86.68%), respectively. In conclusion, the proposed CNN-BiLSTM atrial fibrillation detection and ECG beat classification algorithm based on the Inception module has a better classification efficacy than the CNN-BiLSTM combined model.
		2024 Vol. 43 (4): 447-454 [Abstract] ( 64 ) HTML (1 KB) PDF (1043 KB) ( 38 )

	455	Design of Intelligent Neuronavigation Surgical Robot System Based on Compliance Control
		Wang Jie, Chen Xinrong, Song Zhijian
		DOI: 10.3969/j.issn.0258-8021.2024.04.008
		Robot-assisted minimally invasive surgery has become more and more popular due to its low invasiveness. However, the disorientation and lack of navigational information limit its further applications in natural orifice surgery. Due to the slender and complicated anatomical structure, the surgical instruments of the robotic end effector are prone to injure surrounding tissues during surgical approaches. An intelligent control and navigational surgical robot system was proposed in this paper. The system featured clinical considerations and was designed to provide reliable and safe preoperative and intraoperative positioning. The inverse kinematics strategy with avoiding joint angle limitation ensured that the 7-DOF robot could achieve high flexibility and strong mobility. The system utilized preoperative CT or MRI data of patients for surgical navigation and surgical planning, simultaneously, to avoid damaging the normal tissue of the patient, a compliance control strategy was introduced to control the interaction force between the patient and the surgical instruments to a small range. To improve the surgical accuracy and relieve the doctor′s workload, the intelligent voice control module realized the micro adjustment of surgical instruments. Phantom and cadaver studies were both conducted to evaluate the effectiveness of the proposed system. The experiments results showed that the positioning error of this system was less than 1 mm, and the tracking angle error was less than 2.5 °. Intraoperative navigation can perform real-time surgical target and instrument tracking, and impedance compliance control reduced the contact force between surgical instruments and patients below 1.2 N. Fine tuning based on speech recognition could meet the requirements of intraoperative motion control of surgical instruments. In conclusion, the designed system has broad application prospects in robot-assisted minimally invasive surgery.
		2024 Vol. 43 (4): 455-466 [Abstract] ( 81 ) HTML (1 KB) PDF (12177 KB) ( 34 )

Reviews

	467	Research Progress on Transformer-Based Deep Learning Models for Medical Image Segmentation
		Zhou Lazhen, Chen Hongchi, Li Qiuxia, Li Fangzuo
		DOI: 10.3969/j.issn.0258-8021.2024.04.009
		Accurate segmentation of medical images is a crucial step in clinical diagnosis and treatment. Over the past decade, convolutional neural network (CNN) has been widely applied in the field of medical image segmentation and have achieved excellent segmentation performance. However, the inherent inductive bias in CNN architectures limits their ability to model long-range dependencies in images. In contrast, the Transformer architectures, which focus on global information and the ability to model long-range dependencies, has been demonstrated outstanding performance in biomedical image segmentation. This review introduced the components of Transformer architecture and its applications in medical image segmentation. From perspectives of fully supervised, unsupervised and semi-supervised learning, application values and performances of Transformer architectures in abdominal multi-organ segmentation, cardiac segmentation and brain tumor segmentation were summarized and analyzed. Finally, limitations of Transformer model in segmentation tasks and future optimizations were prospected.
		2024 Vol. 43 (4): 467-476 [Abstract] ( 199 ) HTML (1 KB) PDF (1000 KB) ( 387 )

	477	Mechanisms and Research Progress of Low-Intensity Ultrasound in Drug Delivery
		Zhu Wenwu, Yang Hui, Hu Kai, Yan Guofei, Wang Fan
		DOI: 10.3969/j.issn.0258-8021.2024.04.010
		Noninvasive low-intensity ultrasound (LIUS) has been applied in multiple fields of clinical therapy and widely investigated and discussed due to its excellent therapeutic outcome. Recently, researches on LIUS in promoting drug delivery in many applications have been booming, covering multiple drug delivery scenarios and application directions, including transdermal drug delivery (sonophoresis), sonodynamic therapy, and brain blood barrier (BBB) opening. This article delves into mechanisms of LIUS from above three perspectives, covering many biophysical effects including thermal effect, cavitation effect, mechanical effect, and physiological regulatory effect, with particular emphasis on the importance of local cavitation and mechanical effects. In addition, this article highlights the effectiveness of LIUS in drug delivery, while also points out issues that need to be further addressed, such as controllability of drug release and safety of clinical treatment. Domestic and foreign studies have shown that the combined application of ultrasound and nanomaterials has enormous potential. In the future, combining ultrasound with nanomaterials and other technologies is expected to lead to innovative development in the field of drug delivery and provide more effective treatment plans for clinical treatment.
		2024 Vol. 43 (4): 477-488 [Abstract] ( 93 ) HTML (1 KB) PDF (883 KB) ( 104 )

	489	Research Progress in Brain Magnetic Source Localization Reconstruction Algorithm
		Yang Yanling, Yao Xufeng, Luo Shichang, Shi Cheng, Gao Xiumin, Wu Tao
		DOI: 10.3969/j.issn.0258-8021.2024.04.011
		As a non-invasive functional neuroimaging method of human brain, magnetoencephalography has been widely concerned in clinical applications due to its high temporal resolution and non-invasive characteristics. The inverse problem of inferring the distribution of current sources in the brain from the data of scalp magnetic field is the central problem in the research of brain magnetic source localization. The difficulty lies in the uniqueness and ill-posed feature of the inverse problem. The reconstruction methods are divided into two categories: distributed source model and dipole localization. Therefore, this article systematically discussed the development of magnetoencephalography and magnetic source imaging. The distributed source model includes minimum norm estimation, low resolution brain electromagnetic tomography, focal underdetermined system solution, Bayesian estimation, beamformer and sparse source imaging. Dipole localization includes maximum entropy method, least-squares minimum norm, multiple-signal classification algorithm, neural network and genetic algorithm. Existing problems and development trends were analyzed, highlighting that multimodal reconstruction methods that integrate multiple brain function technologies are expected to become the most important detection technology for neural function diagnosis.
		2024 Vol. 43 (4): 489-498 [Abstract] ( 60 ) HTML (1 KB) PDF (3372 KB) ( 73 )

	499	Application of Near-Infrared Fluorescence Photothermal Nanoparticles in Photothermal Therapy of Cervical Cancer
		Zhu Lijun, Xiong Jiabao, Du Zhong, Ma Rong, Zhang Xueliang, Nuernisha Alifu
		DOI: 10.3969/j.issn.0258-8021.2024.04.012
		Diagnosis and treatment of cervical cancer are of great significance to improve the survival rate and prognosis of patients. With the rapid development of nanomedicine in disease theranostics systems, near-infrared (NIR) fluorescence and photothermal therapy (PTT) based on fluorescent nanomaterials have provided promising solutions to improve the poor sensitivity, significant side effects and high incidence of postoperative metastasis in the diagnosis and treatment of cervical cancer. In this article, we reviewed the characteristics of bio-modified nanoparticles from NIR imaging and PTT, elaborated on the application of biologically modification new photothermal nanomaterials in the PTT of cervical cancer with the non-invasive, rapid, precise diagnosis and treatment, and low toxic side effects as the characteristic. Potentials of NIR imaging assisted PTT in preclinical research and clinical translation were discussed as well. Meanwhile, new NIR-photothermal nanoparticles and their preclinical experimental basis for translation to clinical treatment of cervical cancer were briefly introduced.
		2024 Vol. 43 (4): 499-507 [Abstract] ( 69 ) HTML (1 KB) PDF (1255 KB) ( 59 )

Communications

	508	Research on Fatigue Process Estimation of Muscle Tissue Based on Ultrasonic Radiofrequency Nakagami Model
		Ran Jianqing, Lv Qian, Zhang Xueqing, Gao Jie, Guo Jianzhong
		DOI: 10.3969/j.issn.0258-8021.2024.04.013

		2024 Vol. 43 (4): 508-512 [Abstract] ( 68 ) HTML (1 KB) PDF (1251 KB) ( 25 )

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