Wanyue Li, Lishen Qiu, J. Zhang, Wenliang Zhu, Lirong Wang
ECG signals are the most commonly used signals in heart disease detection. It contains many waveforms that are directly related to cardiac activity, where the T wave contains much important physiological information. The T wave position detection algorithm is based on the differential threshold method, and the T wave morphological judgment is used before the T wave position detection. The algorithm includes three parts: preprocessing, T wave morphological judgment, T wave position detection. Firstly, the signal is preprocessed to eliminate the effects of noise and other waves. Secondly, a detection window is defined to realize the T wave morphological judgment. Finally, based on the T wave morphology, the T wave position is obtained by a differential threshold method in the detection window. The algorithm was tested on the QT database. By comparing with the manual annotation of the expert in the database, the standard deviation between the algorithm positioning results and the manual labeling results in the database is 30.55 ms at the peak of T wave, and the standard deviation is 47.46 ms at the end of T wave.
{"title":"An Automatic Detection Algorithm for T Wave Position based on T Wave Morphology","authors":"Wanyue Li, Lishen Qiu, J. Zhang, Wenliang Zhu, Lirong Wang","doi":"10.1145/3354031.3354052","DOIUrl":"https://doi.org/10.1145/3354031.3354052","url":null,"abstract":"ECG signals are the most commonly used signals in heart disease detection. It contains many waveforms that are directly related to cardiac activity, where the T wave contains much important physiological information. The T wave position detection algorithm is based on the differential threshold method, and the T wave morphological judgment is used before the T wave position detection. The algorithm includes three parts: preprocessing, T wave morphological judgment, T wave position detection. Firstly, the signal is preprocessed to eliminate the effects of noise and other waves. Secondly, a detection window is defined to realize the T wave morphological judgment. Finally, based on the T wave morphology, the T wave position is obtained by a differential threshold method in the detection window. The algorithm was tested on the QT database. By comparing with the manual annotation of the expert in the database, the standard deviation between the algorithm positioning results and the manual labeling results in the database is 30.55 ms at the peak of T wave, and the standard deviation is 47.46 ms at the end of T wave.","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":"132909471","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}
An automatic multi-organ segmentation method from upper abdominal CT image is proposed in this paper. A group of statistical shape models for multiple organs are generated by learning the statistical distribution of organs' shapes and intensity profiles. Then, a random forest regression model is trained to find the candidate position to initialize the statistical shape model. The proposed method is evaluated at segmentation of four abdomen organs (spleen, right kidney, left kidney and liver) from training set of 26 cases of upper abdominal CT images. The accuracy shows that the initialization improves the accuracy for statistical shape model-based segmentation.
{"title":"Multi-organ Segmentation from Abdominal CT with Random Forest based Statistical Shape Model","authors":"Jiaqi Wu, Guangxu Li, Huimin Lu, Hyoungseop Kim","doi":"10.1145/3354031.3354042","DOIUrl":"https://doi.org/10.1145/3354031.3354042","url":null,"abstract":"An automatic multi-organ segmentation method from upper abdominal CT image is proposed in this paper. A group of statistical shape models for multiple organs are generated by learning the statistical distribution of organs' shapes and intensity profiles. Then, a random forest regression model is trained to find the candidate position to initialize the statistical shape model. The proposed method is evaluated at segmentation of four abdomen organs (spleen, right kidney, left kidney and liver) from training set of 26 cases of upper abdominal CT images. The accuracy shows that the initialization improves the accuracy for statistical shape model-based segmentation.","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":"116666647","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}
Automatic classification of artery and vein vessels in retinal images is still a challenging task. Recent work mainly focuses on the graph analysis of retinal vessels or intensity based feature extraction. In this study, we use one stage multiclass segmentation without any graph-based or vote-based post processing to solve the artery/vein classification problem directly and effectively. We experimentally showed that with limited training data, data augmentation may be at least as crucial as designing complicated deep model architectures in improving the performance of artery/vein classification. In particular, simply with rotation augmentation, the popular deeply supervised U-Net (DS-Unet) is already comparable to or even outperforms the state-of-the-art methods on DRIVE dataset. Our experiments on two datasets show that artery-vein-background segmentation based on deep learning can be used as a promising method for arteriovenous classification and can be combined with conventional methods for better results.
{"title":"Retinal Artery/Vein Classification via Rotation Augmentation and Deeply Supervised U-net Segmentation","authors":"Zhaolei Wang, Junbin Lin, Ruixuan Wang, Weishi Zheng","doi":"10.1145/3354031.3354050","DOIUrl":"https://doi.org/10.1145/3354031.3354050","url":null,"abstract":"Automatic classification of artery and vein vessels in retinal images is still a challenging task. Recent work mainly focuses on the graph analysis of retinal vessels or intensity based feature extraction. In this study, we use one stage multiclass segmentation without any graph-based or vote-based post processing to solve the artery/vein classification problem directly and effectively. We experimentally showed that with limited training data, data augmentation may be at least as crucial as designing complicated deep model architectures in improving the performance of artery/vein classification. In particular, simply with rotation augmentation, the popular deeply supervised U-Net (DS-Unet) is already comparable to or even outperforms the state-of-the-art methods on DRIVE dataset. Our experiments on two datasets show that artery-vein-background segmentation based on deep learning can be used as a promising method for arteriovenous classification and can be combined with conventional methods for better results.","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":"124419251","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}
Motor system relies on the recruitment of motor modules to perform various movements. Muscle synergies are the modules used by the central nervous system to simplify the control of complex motor tasks. In this paper, we verified the hypothesis that the control of upper limb motions can be realized by combinations of basic muscle synergies extracted from electromyography signals of simple upper limb motions. Muscle synergies and corresponding activation coefficients were extracted via non-negative matrix factorization from three basic and five complex motions. Similarities across basic synergies and complex synergies were compared. We found that the structure of muscle synergies from complex tasks were similar to the corresponding basic synergies and muscle synergies from basic tasks can be used to reconstruct muscle patterns of the complex tasks. This study demonstrates that different kinds of upper limb motions can be accomplished by the combinations of basic muscle synergies modulated by different activation coefficients.
{"title":"Control of Upper Limb Motions by Combinations of Basic Muscle Synergies","authors":"Bingyu Pan, Yingfei Sun, Licai Sun, Zhipei Huang, Jiankang Wu","doi":"10.1145/3354031.3354038","DOIUrl":"https://doi.org/10.1145/3354031.3354038","url":null,"abstract":"Motor system relies on the recruitment of motor modules to perform various movements. Muscle synergies are the modules used by the central nervous system to simplify the control of complex motor tasks. In this paper, we verified the hypothesis that the control of upper limb motions can be realized by combinations of basic muscle synergies extracted from electromyography signals of simple upper limb motions. Muscle synergies and corresponding activation coefficients were extracted via non-negative matrix factorization from three basic and five complex motions. Similarities across basic synergies and complex synergies were compared. We found that the structure of muscle synergies from complex tasks were similar to the corresponding basic synergies and muscle synergies from basic tasks can be used to reconstruct muscle patterns of the complex tasks. This study demonstrates that different kinds of upper limb motions can be accomplished by the combinations of basic muscle synergies modulated by different activation coefficients.","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":"123464777","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}
As the basic tool for the diagnosis of cardiac diseases, electrocardiogram (ECG) is often contaminated by muscle artifacts, which can cause unreliable interpretation and measurement for ECG. To adequately remove muscle artifacts which contaminate ECG signals, we propose a novel computation framework combining the convolution auto-encoder (CAE) and average beat subtraction in this paper. Firstly, the framework subtracts from the original ECG signal based on an initial average beat, which preserves the characteristics of a heart beat; the average beat is updated according to the original ECG signal to incorporate inter-beat variations. Then, the framework filters the residual ECG signal by a convolution auto-encoder (CAE), which filters out the contaminated parts and keeps the specific information related to the ECG signal. Finally, we combine the filtered residual ECG signal and updated average beat to obtain an enhanced ECG signal. Our framework is evaluated on ECG records from the MIT-BIH Arrhythmia Database, and results show that our framework outperforms existing methods in muscle artifacts removal.
{"title":"Muscle Artifacts Cancellation Framework for ECG Signals Combining Convolution Auto-encoder and Average Beat Subtraction","authors":"Yongfeng Huang, Zijian Ding, Guijin Wang, Jianping Lin, Ping Zhang","doi":"10.1145/3354031.3354041","DOIUrl":"https://doi.org/10.1145/3354031.3354041","url":null,"abstract":"As the basic tool for the diagnosis of cardiac diseases, electrocardiogram (ECG) is often contaminated by muscle artifacts, which can cause unreliable interpretation and measurement for ECG. To adequately remove muscle artifacts which contaminate ECG signals, we propose a novel computation framework combining the convolution auto-encoder (CAE) and average beat subtraction in this paper. Firstly, the framework subtracts from the original ECG signal based on an initial average beat, which preserves the characteristics of a heart beat; the average beat is updated according to the original ECG signal to incorporate inter-beat variations. Then, the framework filters the residual ECG signal by a convolution auto-encoder (CAE), which filters out the contaminated parts and keeps the specific information related to the ECG signal. Finally, we combine the filtered residual ECG signal and updated average beat to obtain an enhanced ECG signal. Our framework is evaluated on ECG records from the MIT-BIH Arrhythmia Database, and results show that our framework outperforms existing methods in muscle artifacts removal.","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":"133117561","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}