Sheng Lu, Tong Chen, Fan Yang, Chenglei Peng, S. Du, Yang Li
Single Particle Tracking (SPT) in fluorescence microscopy image is of great importance in the field of computational biology. Automatic or slightly interactive tracking algorithms are essential for the motional analysis of micro particles. Even with prior knowledge, conventional methods may fail when the signal-to-noise ratio (SNR) is too low because they highly depend on the quality of the image and the results of detection. To reliably track particles in the low SNR images, we proposed a novel method based on minimal path theory and attempted to extract complete trajectories between two points. Our method was evaluated on several simulated image sequences and showed its accuracy and robustness in the task of particle tracking.
{"title":"Minimal Path based Particle Tracking in Low SNR Fluorescence Microscopy Images","authors":"Sheng Lu, Tong Chen, Fan Yang, Chenglei Peng, S. Du, Yang Li","doi":"10.1145/3354031.3354035","DOIUrl":"https://doi.org/10.1145/3354031.3354035","url":null,"abstract":"Single Particle Tracking (SPT) in fluorescence microscopy image is of great importance in the field of computational biology. Automatic or slightly interactive tracking algorithms are essential for the motional analysis of micro particles. Even with prior knowledge, conventional methods may fail when the signal-to-noise ratio (SNR) is too low because they highly depend on the quality of the image and the results of detection. To reliably track particles in the low SNR images, we proposed a novel method based on minimal path theory and attempted to extract complete trajectories between two points. Our method was evaluated on several simulated image sequences and showed its accuracy and robustness in the task of particle tracking.","PeriodicalId":286321,"journal":{"name":"Proceedings of the 4th International Conference on Biomedical Signal and Image Processing","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2019-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128481838","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Electroencephalography (EEG)-based BCIs have experienced a significant growth in recent years, especially the passive Brain Computer Interfaces (BCIs) with a wide application in the detection of cognitive and emotional states. But it is still unclear whether more subtle states, e.g., covert selective attention can be decoded with EEG signals. Here we used a behavioral paradigm to introduce the shift of selective attention between the visual and auditory domain. With EEG signals, we extracted features based on Grange Causality (GC) and successfully decoded the attentional shift through a support vector machine (SVM) based classifier. The decoding accuracy was significantly above the chance level for all 8 subjects tested. The features based on GC were further analyzed with tree-based feature importance analysis and recursive feature elimination (RFE) method to search for the optimal features for classification. Our work demonstrate that specific patterns of brain activities reflected by GC can be used to decode subtle state changes of the brain related to cross-modal selective attention, which opens new possibility of using passive BCIs in sophisticated perceptual and cognitive tasks.
{"title":"Application of Granger Causality in Decoding Covert Selective Attention with Human EEG","authors":"Weikun Niu, Yuying Jiang, Yujin Zhang, Xin Zhang, Shan Yu","doi":"10.1145/3354031.3354032","DOIUrl":"https://doi.org/10.1145/3354031.3354032","url":null,"abstract":"Electroencephalography (EEG)-based BCIs have experienced a significant growth in recent years, especially the passive Brain Computer Interfaces (BCIs) with a wide application in the detection of cognitive and emotional states. But it is still unclear whether more subtle states, e.g., covert selective attention can be decoded with EEG signals. Here we used a behavioral paradigm to introduce the shift of selective attention between the visual and auditory domain. With EEG signals, we extracted features based on Grange Causality (GC) and successfully decoded the attentional shift through a support vector machine (SVM) based classifier. The decoding accuracy was significantly above the chance level for all 8 subjects tested. The features based on GC were further analyzed with tree-based feature importance analysis and recursive feature elimination (RFE) method to search for the optimal features for classification. Our work demonstrate that specific patterns of brain activities reflected by GC can be used to decode subtle state changes of the brain related to cross-modal selective attention, which opens new possibility of using passive BCIs in sophisticated perceptual and cognitive tasks.","PeriodicalId":286321,"journal":{"name":"Proceedings of the 4th International Conference on Biomedical Signal and Image Processing","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2019-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116954527","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Ziya Yu, Chi Zhang, Linyuan Wang, Li Tong, Bin Yan
A convolutional neural network with outstanding performance in computer vision can be used to construct an encoding model that simulates the process of human visual information processing. However, training goal of the network may have impacted the performance of encoding model. Most neural networks used to establish encoding models in the past were performed image classification task, the task of which is single. While in the process of human's visual perception, multiple tasks are performed simultaneously. Thus, the existing encoding model does not well satisfy the diversity and complexity of the human visual mechanism. In this paper, we first established a feature extraction model based on Fully Convolutional Network (FCN) and Visual Geometry Group (VGG) with similar network structure but different training goal, and employed Regularize Orthogonal Matching Pursuit (ROMP) to establish the response model, which can predict the stimuli-evoked responses measured by functional magnetic resonance imaging (fMRI). The results revealed that the convolutional neural networks trained by different visual tasks had significant difference in the performance of visual encoding with almost the same network structure. The VGG-based encoding model can achieve a higher performance in most voxels of ROIs. We concluded that classification task in computer vision can better fit the visual mechanism of human compared to visual segmentation task.
{"title":"Different Goal-driven CNNs Affect Performance of Visual Encoding Models based on Deep Learning","authors":"Ziya Yu, Chi Zhang, Linyuan Wang, Li Tong, Bin Yan","doi":"10.1145/3354031.3354045","DOIUrl":"https://doi.org/10.1145/3354031.3354045","url":null,"abstract":"A convolutional neural network with outstanding performance in computer vision can be used to construct an encoding model that simulates the process of human visual information processing. However, training goal of the network may have impacted the performance of encoding model. Most neural networks used to establish encoding models in the past were performed image classification task, the task of which is single. While in the process of human's visual perception, multiple tasks are performed simultaneously. Thus, the existing encoding model does not well satisfy the diversity and complexity of the human visual mechanism. In this paper, we first established a feature extraction model based on Fully Convolutional Network (FCN) and Visual Geometry Group (VGG) with similar network structure but different training goal, and employed Regularize Orthogonal Matching Pursuit (ROMP) to establish the response model, which can predict the stimuli-evoked responses measured by functional magnetic resonance imaging (fMRI). The results revealed that the convolutional neural networks trained by different visual tasks had significant difference in the performance of visual encoding with almost the same network structure. The VGG-based encoding model can achieve a higher performance in most voxels of ROIs. We concluded that classification task in computer vision can better fit the visual mechanism of human compared to visual segmentation task.","PeriodicalId":286321,"journal":{"name":"Proceedings of the 4th International Conference on Biomedical Signal and Image Processing","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2019-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127857969","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
For the binary classification tasks in supervised learning, the labels of data have to be available for classifier development. Cohen's kappa is usually employed as a quality measure for data annotation, which is inconsistent with its true functionality of assessing the inter-annotator consistency. However, the derived relationship functions of Cohen's kappa, sensitivity, and specificity in the literature are complicated, thus are unable to be employed to interpret classification performance from kappa values. In this study, based on an annotation generation model, we develop simple relationships of kappa, sensitivity, and specificity when there is no bias in the annotations. A relationship between kappa and Youden's J statistic, a performance metric for binary classification, is further obtained. The derived relationships are evaluated on a synthetic dataset using linear regression analysis. The results demonstrate the accuracy of the derived relationships. It suggests the potential of estimating classification performance from kappa values when bias is absent in the annotations.
{"title":"Relationships of Cohen's Kappa, Sensitivity, and Specificity for Unbiased Annotations","authors":"Juan Wang, Bin Xia","doi":"10.1145/3354031.3354040","DOIUrl":"https://doi.org/10.1145/3354031.3354040","url":null,"abstract":"For the binary classification tasks in supervised learning, the labels of data have to be available for classifier development. Cohen's kappa is usually employed as a quality measure for data annotation, which is inconsistent with its true functionality of assessing the inter-annotator consistency. However, the derived relationship functions of Cohen's kappa, sensitivity, and specificity in the literature are complicated, thus are unable to be employed to interpret classification performance from kappa values. In this study, based on an annotation generation model, we develop simple relationships of kappa, sensitivity, and specificity when there is no bias in the annotations. A relationship between kappa and Youden's J statistic, a performance metric for binary classification, is further obtained. The derived relationships are evaluated on a synthetic dataset using linear regression analysis. The results demonstrate the accuracy of the derived relationships. It suggests the potential of estimating classification performance from kappa values when bias is absent in the annotations.","PeriodicalId":286321,"journal":{"name":"Proceedings of the 4th International Conference on Biomedical Signal and Image Processing","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2019-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126630737","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jenny Ni, Wenliang Zhu, Jinfu Huang, Longfei Niu, Lirong Wang
In order to solve the situation that the elderly can't get timely assistance when they fall down, this paper designs and implements an Android-based fall monitoring application for the elderly (Fall Guard). Combined with existing fall detection devices and cloud server, Fall Guard uses Model-View-Control (MVC) structure and OkHttp network request framework. In this paper, OkHttp is used to send network requests to the cloud server and the fall detection device is bound to the mobile client to obtain user and device information. After successfully obtaining data, the obtained information is displayed on the user interface through JSON parsing, including device positioning, electronic fence, fall alarm information and motion track. One login account can be bound to multiple devices. The login account is set to the emergency contact number of the elderly by default. When the elderly fall, Fall Guard can receive many types of fall alarm prompts, including alarm information list, notification bar reminder, SMS notification, device user status bar information and other prompt functions. Test results show that Fall Guard has good monitoring accuracy and terminal compatibility. On the one hand, it can adapt to different models and brands of Android mobile terminals to achieve accurate positioning and alarm functions. On the other hand, due to its one-to-many management mode, it can be applied to deployment of different application scenarios such as home, community and nursing home.
{"title":"Fall Guard: Fall Monitoring Application for the Elderly based on Android Platform","authors":"Jenny Ni, Wenliang Zhu, Jinfu Huang, Longfei Niu, Lirong Wang","doi":"10.1145/3354031.3354055","DOIUrl":"https://doi.org/10.1145/3354031.3354055","url":null,"abstract":"In order to solve the situation that the elderly can't get timely assistance when they fall down, this paper designs and implements an Android-based fall monitoring application for the elderly (Fall Guard). Combined with existing fall detection devices and cloud server, Fall Guard uses Model-View-Control (MVC) structure and OkHttp network request framework. In this paper, OkHttp is used to send network requests to the cloud server and the fall detection device is bound to the mobile client to obtain user and device information. After successfully obtaining data, the obtained information is displayed on the user interface through JSON parsing, including device positioning, electronic fence, fall alarm information and motion track. One login account can be bound to multiple devices. The login account is set to the emergency contact number of the elderly by default. When the elderly fall, Fall Guard can receive many types of fall alarm prompts, including alarm information list, notification bar reminder, SMS notification, device user status bar information and other prompt functions. Test results show that Fall Guard has good monitoring accuracy and terminal compatibility. On the one hand, it can adapt to different models and brands of Android mobile terminals to achieve accurate positioning and alarm functions. On the other hand, due to its one-to-many management mode, it can be applied to deployment of different application scenarios such as home, community and nursing home.","PeriodicalId":286321,"journal":{"name":"Proceedings of the 4th International Conference on Biomedical Signal and Image Processing","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2019-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130365358","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Hand Gesture Recognition (HGR) is a system to translate the hand gestures express to literature, which is a natural way of communication between deaf-mutes and non-disabled people. However, due to the complexity of relative positions of fingers, hands sizes, and environmental illumination, the hand gesture recognition is difficult. In this paper, a Fully Connected Neural Network (FCNN) algorithm for RGB-D sensor based HGR is proposed. We firstly build datasets of fingers joints and the center coordinates of hands in 3 dimensions. Then we normalize the samples to eliminate the natural difference of hands. Finally, the data are classified using a 3 layers FCNN. Totally 13,000 data of 26 hand gestures are collected. We randomly select 80% of these data for training and 20% of them for testing. According to the experiments, the average recognition accuracy is 94.73%.
{"title":"RGB-D-based Hand Gesture Recognition for Letters Expression","authors":"Jin Li, J. Yan, Guangxu Li, Liyuan Wang, Fan Yang","doi":"10.1145/3354031.3354044","DOIUrl":"https://doi.org/10.1145/3354031.3354044","url":null,"abstract":"Hand Gesture Recognition (HGR) is a system to translate the hand gestures express to literature, which is a natural way of communication between deaf-mutes and non-disabled people. However, due to the complexity of relative positions of fingers, hands sizes, and environmental illumination, the hand gesture recognition is difficult. In this paper, a Fully Connected Neural Network (FCNN) algorithm for RGB-D sensor based HGR is proposed. We firstly build datasets of fingers joints and the center coordinates of hands in 3 dimensions. Then we normalize the samples to eliminate the natural difference of hands. Finally, the data are classified using a 3 layers FCNN. Totally 13,000 data of 26 hand gestures are collected. We randomly select 80% of these data for training and 20% of them for testing. According to the experiments, the average recognition accuracy is 94.73%.","PeriodicalId":286321,"journal":{"name":"Proceedings of the 4th International Conference on Biomedical Signal and Image Processing","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2019-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132469727","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Feifei Sun, Wenliang Zhu, Gang Ma, Kongpeng Xing, Lirong Wang
In order to promote the automation of medical devices, we designed a test and analysis instrument which we call it as automatic liquid chip system. Based on the fully automatic liquid chip system, we designed various modules, including sample needle module, reagent bin module, reaction plate module, cleaning module, waste bin module and system liquid module. It includes the design of mechanical structure and electronic control system. The control chip of each module is STM32F103RET6. The main control part includes stepping motor, optocoupler sensor and AD converter. The whole communication is carried out through CAN (Controller Area Network) protocol. The serial instructions are sent to the upper computer, and the conversion from serial instructions to CAN instructions is completed in the transfer station. Each module receives CAN instructions, performs FIFO caching, and then performs corresponding operations. Considering the stability of each module, the universality of debugging and the stability of the whole system, this experiment designs the common parts of the module, including the design of stepping motor driver, the generation of software PWM and the configuration of CAN communication protocol. Then we take the cleaning module as an example to design its circuit and workflow.
为了促进医疗器械的自动化,我们设计了一种测试分析仪器,我们称之为自动液体芯片系统。在全自动液片系统的基础上,设计了样品针模块、试剂仓模块、反应板模块、清洗模块、垃圾桶模块和系统液模块。包括机械结构设计和电子控制系统设计。各模块的控制芯片为STM32F103RET6。主要控制部分包括步进电机、光耦传感器和AD转换器。整个通信通过CAN (Controller Area Network,控制器局域网)协议进行。将串行指令发送到上位机,在中转站完成串行指令到CAN指令的转换。每个模块接收CAN指令,进行FIFO缓存,然后进行相应的操作。考虑到各模块的稳定性、调试的通用性和整个系统的稳定性,本实验对模块的通用部分进行了设计,包括步进电机驱动器的设计、软件PWM的生成和CAN通信协议的配置。然后以清洗模块为例,设计了清洗模块的电路和工作流程。
{"title":"Design of Cleaning Module based on CAN","authors":"Feifei Sun, Wenliang Zhu, Gang Ma, Kongpeng Xing, Lirong Wang","doi":"10.1145/3354031.3354053","DOIUrl":"https://doi.org/10.1145/3354031.3354053","url":null,"abstract":"In order to promote the automation of medical devices, we designed a test and analysis instrument which we call it as automatic liquid chip system. Based on the fully automatic liquid chip system, we designed various modules, including sample needle module, reagent bin module, reaction plate module, cleaning module, waste bin module and system liquid module. It includes the design of mechanical structure and electronic control system. The control chip of each module is STM32F103RET6. The main control part includes stepping motor, optocoupler sensor and AD converter. The whole communication is carried out through CAN (Controller Area Network) protocol. The serial instructions are sent to the upper computer, and the conversion from serial instructions to CAN instructions is completed in the transfer station. Each module receives CAN instructions, performs FIFO caching, and then performs corresponding operations. Considering the stability of each module, the universality of debugging and the stability of the whole system, this experiment designs the common parts of the module, including the design of stepping motor driver, the generation of software PWM and the configuration of CAN communication protocol. Then we take the cleaning module as an example to design its circuit and workflow.","PeriodicalId":286321,"journal":{"name":"Proceedings of the 4th International Conference on Biomedical Signal and Image Processing","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2019-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114044729","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yi Ren, Zhipei Huang, Yiming Guo, Jiankang Wu, Yingfei Sun
Kinematic characteristics have been playing a crucial role in assessing the quality of movements and improving training plans. We design five characteristic parameters of table tennis technical movements in this paper, i.e., the normalized path, joint angle, phase duration, root mean square and velocity entropy. Based on the motion data obtained from immersive motion capture system, the validity of these characteristic parameters was verified by analyzing backhand block movement. Twenty subjects with two different skill levels were involved in this test to perform backhand block against the ball. The statistical analysis results revealed that there were significant differences between the parameters of the professional group and those of the novice group, including normalized path, velocity entropy, root mean square and joint angle. Meanwhile, phase duration and joint angle showed practical significance biomechanically. These characteristic parameters could serve as indicators for movement quality assessment and could be extended to other table tennis technical movements as well as further biomechanics research.
{"title":"Kinematic Characteristics of Backhand Block in Table Tennis","authors":"Yi Ren, Zhipei Huang, Yiming Guo, Jiankang Wu, Yingfei Sun","doi":"10.1145/3354031.3354034","DOIUrl":"https://doi.org/10.1145/3354031.3354034","url":null,"abstract":"Kinematic characteristics have been playing a crucial role in assessing the quality of movements and improving training plans. We design five characteristic parameters of table tennis technical movements in this paper, i.e., the normalized path, joint angle, phase duration, root mean square and velocity entropy. Based on the motion data obtained from immersive motion capture system, the validity of these characteristic parameters was verified by analyzing backhand block movement. Twenty subjects with two different skill levels were involved in this test to perform backhand block against the ball. The statistical analysis results revealed that there were significant differences between the parameters of the professional group and those of the novice group, including normalized path, velocity entropy, root mean square and joint angle. Meanwhile, phase duration and joint angle showed practical significance biomechanically. These characteristic parameters could serve as indicators for movement quality assessment and could be extended to other table tennis technical movements as well as further biomechanics research.","PeriodicalId":286321,"journal":{"name":"Proceedings of the 4th International Conference on Biomedical Signal and Image Processing","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2019-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122717303","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Dingyun Liu, Hongxiu Jiang, N. Rao, Cheng-Si Luo, Wenju Du, Zheng-wen Li, Tao Gan
The diagnoses of Early Esophageal Cancer (EEC) based on gastroscopic images is a challenging task in clinic, which relies heavily on subjective artificial detection and annotation. As a result, computer aided diagnosis (CAD) methods that support the clinicians become highly attractive. In this paper, we proposed a CAD method which realized the automatic detection and annotation of EEC lesions in gastroscopic images. The proposed method initially utilized an advanced Deep Learning (DL) network Deeplabv3+ to obtain a preliminary prediction of EEC regions. Then, a post-processing step which referenced the clinical requirements was designed and applied to get the final annotation results. Totally 3190 gastroscopic images of 732 patients were used in this work. The final experimental results show that the EEC detection rate of our method was 97.07%, and the mean Dice Similarity Coefficient (DSC) was 74.01%, which are higher than those of other state-of-the-are DL-based methods. In addition, the false positive output of our method is fewer. Therefore, the proposed method offers a good potential to aid the clinical diagnoses of EEC.
基于胃镜图像的早期食管癌诊断在临床上是一项具有挑战性的任务,它严重依赖于主观的人工检测和注释。因此,支持临床医生的计算机辅助诊断(CAD)方法变得非常有吸引力。本文提出了一种CAD方法,实现了胃镜图像中EEC病变的自动检测与标注。该方法首先利用先进的深度学习(DL)网络Deeplabv3+对EEC区域进行初步预测。然后,参照临床需求设计并应用后处理步骤,得到最终标注结果。本研究共使用了732例患者的3190张胃镜图像。最终实验结果表明,该方法的EEC检测率为97.07%,平均Dice相似系数(DSC)为74.01%,高于其他基于state- are dl的方法。此外,我们的方法的假阳性输出更少。因此,该方法在辅助脑电图临床诊断方面具有很好的潜力。
{"title":"Computer Aided Annotation of Early Esophageal Cancer in Gastroscopic Images based on Deeplabv3+ Network","authors":"Dingyun Liu, Hongxiu Jiang, N. Rao, Cheng-Si Luo, Wenju Du, Zheng-wen Li, Tao Gan","doi":"10.1145/3354031.3354046","DOIUrl":"https://doi.org/10.1145/3354031.3354046","url":null,"abstract":"The diagnoses of Early Esophageal Cancer (EEC) based on gastroscopic images is a challenging task in clinic, which relies heavily on subjective artificial detection and annotation. As a result, computer aided diagnosis (CAD) methods that support the clinicians become highly attractive. In this paper, we proposed a CAD method which realized the automatic detection and annotation of EEC lesions in gastroscopic images. The proposed method initially utilized an advanced Deep Learning (DL) network Deeplabv3+ to obtain a preliminary prediction of EEC regions. Then, a post-processing step which referenced the clinical requirements was designed and applied to get the final annotation results. Totally 3190 gastroscopic images of 732 patients were used in this work. The final experimental results show that the EEC detection rate of our method was 97.07%, and the mean Dice Similarity Coefficient (DSC) was 74.01%, which are higher than those of other state-of-the-are DL-based methods. In addition, the false positive output of our method is fewer. Therefore, the proposed method offers a good potential to aid the clinical diagnoses of EEC.","PeriodicalId":286321,"journal":{"name":"Proceedings of the 4th International Conference on Biomedical Signal and Image Processing","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2019-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132226066","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Mitsuaki Nagao, Huimin Lu, Hyoungseop Kim, T. Aoki, S. Kido
Recently, visual screening based on CT images become the useful tool in the medical diagnosis. However, due to the increasing data volumes and the computational complexity of the algorithms, image processing technique for the high quality visual screening is still required. To this end, some computer aided diagnosis (CAD) algorithms are proposed. Meanwhile, cancer is a leading cause of death in the world. Detection of cancer region in CT images is the most important task to early detection and early treatment. We design and develop a framework combining convolutional neural networks (CNN) with temporal subtraction techniques-based non-rigid image registration algorithm. However, conventional CNN has the issue that as the layers deeper, global information close to input images is lost. Therefore, we add a skip connection to conventional CNN. By adding a new skip connection, the proposed CNN network maintains the global information without loss of important features of input image. All in all, our proposed method can be built into three main steps; i) pre-processing for image segmentation, ii) image matching for registration, and iii) classification of abnormal regions based on machine learning algorithms. We perform our proposed technique to 25 thoracic MDCT sets and obtain the AUC score of 0.951.
{"title":"Detection of Abnormal Regions on Temporal Subtraction Images based on CNN","authors":"Mitsuaki Nagao, Huimin Lu, Hyoungseop Kim, T. Aoki, S. Kido","doi":"10.1145/3354031.3354049","DOIUrl":"https://doi.org/10.1145/3354031.3354049","url":null,"abstract":"Recently, visual screening based on CT images become the useful tool in the medical diagnosis. However, due to the increasing data volumes and the computational complexity of the algorithms, image processing technique for the high quality visual screening is still required. To this end, some computer aided diagnosis (CAD) algorithms are proposed. Meanwhile, cancer is a leading cause of death in the world. Detection of cancer region in CT images is the most important task to early detection and early treatment. We design and develop a framework combining convolutional neural networks (CNN) with temporal subtraction techniques-based non-rigid image registration algorithm. However, conventional CNN has the issue that as the layers deeper, global information close to input images is lost. Therefore, we add a skip connection to conventional CNN. By adding a new skip connection, the proposed CNN network maintains the global information without loss of important features of input image. All in all, our proposed method can be built into three main steps; i) pre-processing for image segmentation, ii) image matching for registration, and iii) classification of abnormal regions based on machine learning algorithms. We perform our proposed technique to 25 thoracic MDCT sets and obtain the AUC score of 0.951.","PeriodicalId":286321,"journal":{"name":"Proceedings of the 4th International Conference on Biomedical Signal and Image Processing","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2019-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117009519","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}