基于深度卷積神經(jīng)網(wǎng)絡的多元醫(yī)學信號多級上下文自編碼器
doi: 10.11999/JEIT190135
-
1.
北京工業(yè)大學信息學部 北京 100124
-
2.
北京工業(yè)大學計算智能與智能系統(tǒng)北京重點實驗室 北京 100124
Multi-context Autoencoders for Multivariate Medical Signals Based on Deep Convolutional Neural Networks
-
1.
Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China
-
2.
Beijing Key Laboratory of Computational Intelligence and Intelligent System, Beijing University of Technology, Beijing 100124, China
-
摘要:
多元醫(yī)學信號的典型代表有多模態(tài)睡眠圖和多通道腦電圖等，采用無監(jiān)督深度學習表征多元醫(yī)學信號是目前健康信息學領(lǐng)域中的一個研究熱點。為了解決現(xiàn)有模型沒有充分結(jié)合醫(yī)學信號多元時序結(jié)構(gòu)特點的問題，該文提出了一種無監(jiān)督的多級上下文深度卷積自編碼器(mCtx-CAE)。首先改進傳統(tǒng)卷積神經(jīng)網(wǎng)絡結(jié)構(gòu)，提出一種多元卷積自編碼模塊，以提取信號片段內(nèi)的多元上下文特征；其次，提出采用語義學習技術(shù)對信號片段間的時序信息進行自編碼，進一步提取時序上下文特征；最后通過共享特征表示設計目標函數(shù)，訓練端到端的多級上下文自編碼器。實驗結(jié)果表明，該文所提模型在兩種應用于不同醫(yī)療場景下的多模態(tài)和多通道數(shù)據(jù)集(UCD和CHB-MIT)上表現(xiàn)均優(yōu)于其它無監(jiān)督特征學習方法，能有效提高多元醫(yī)學信號的融合特征表達能力，對提高臨床時序數(shù)據(jù)的分析效率有著重要意義。
-
關(guān)鍵詞:
- 多元醫(yī)學信號 /
- 自編碼器 /
- 上下文學習 /
- 卷積神經(jīng)網(wǎng)絡 /
- 深度學習
Abstract:Learning unsupervised representations from multivariate medical signals, such as multi-modality polysomnography and multi-channel electroencephalogram, has gained increasing attention in health informatics. In order to solve the problem that the existing models do not fully incorporate the characteristics of the multivariate-temporal structure of medical signals, an unsupervised multi-Context deep Convolutional AutoEncoder (mCtx-CAE) is proposed in this paper. Firstly, by modifying traditional convolutional neural networks, a multivariate convolutional autoencoder is proposed to extract multivariate context features within signal segments. Secondly, semantic learning is adopted to auto-encode temporal information among signal segments, to further extract temporal context features. Finally, an end-to-end multi-context autoencoder is trained by designing objective function based on shared feature representation. Experimental results conducted on two public benchmark datasets (UCD and CHB-MIT) show that the proposed model outperforms the state-of-the-art unsupervised feature learning methods in different medical tasks, demonstrating the effectiveness of the learned fusional features in clinical settings.
-
表 1 多元卷積自編碼模塊具體配置參數(shù)
編碼單元 卷積層 非線性變換 池化層 元內(nèi)編碼單元 $1 \times 3 \times 16$ ReLU $1 \times 2$ 元間編碼單元 $C \times 3 \times 8$ ReLU $1 \times 2$ 解碼單元 反卷積層 非線性變換 反池化層 元間解碼單元 $C \times 3 \times 8$ ReLU $1 \times 2$ 元內(nèi)解碼單元 $1 \times 3 \times 16$ ReLU $1 \times 2$ 
下載: 導出CSV
表 2 CHB-MIT數(shù)據(jù)庫上的方法比較結(jié)果
方法 AUC-ROC AUC-PR F1分子 準確率 PCA 0.8291 ± 0.0434 0.7021 ± 0.0872 0.6421 ± 0.0223 0.8768 ± 0.0223 SAE 0.5934 ± 0.0377 0.4180 ± 0.1189 0.0668 ± 0.0415 0.7987 ± 0.0309 CAE 0.8657 ± 0.0305 0.7646 ± 0.0881 0.6277 ± 0.1246 0.8690 ± 0.0267 Med2Vec 0.8155 ± 0.1181 0.5870 ± 0.1670 0.6066 ± 0.2363 0.8351 ± 0.0359 Skip-gram+ 0.9090 ± 0.0356 0.7467 ± 0.1540 0.6288 ± 0.2040 0.8898 ± 0.0173 CtxFusionEEG 0.9287 ± 0.0306 0.7833 ± 0.1147 0.7202 ± 0.1485 0.9025 ± 0.0104 Wave2Vec 0.9035 ± 0.0371 0.8839 ± 0.0261 0.8267 ± 0.0184 0.9210 ± 0.0099 m-CAE 0.8946 ± 0.0401 0.8727 ± 0.0189 0.8417 ± 0.0131 0.9324 ± 0.0058 mCtx-CAE 0.9372 ± 0.0495 0.8980 ± 0.0333 0.8493 ± 0.0191 0.9412 ± 0.0110
下載: 導出CSV
表 3 UCD數(shù)據(jù)庫上的方法比較結(jié)果
方法 AUC-ROC AUC-PR F1分數(shù) 準確率 PCA 0.8177 ± 0.0142 0.5764 ± 0.0172 0.5204 ± 0.0275 0.6193 ± 0.0638 SAE 0.7068 ± 0.1372 0.4965 ± 0.0951 0.2760 ± 0.1815 0.4917 ± 0.1364 CAE 0.8386 ± 0.0376 0.5710 ± 0.0429 0.5180 ± 0.0701 0.6208 ± 0.0961 Med2Vec 0.7479 ± 0.0796 0.4836 ± 0.1046 0.3997 ± 0.1361 0.5619 ± 0.0619 Skip-gram+ 0.8010 ± 0.0992 0.5406 ± 0.0995 0.4342 ± 0.1731 0.5884 ± 0.1077 CtxFusionEEG 0.7941 ± 0.1485 0.6358 ± 0.0709 0.5171 ± 0.1994 0.6375 ± 0.1074 Wave2Vec 0.8161 ± 0.0507 0.5984 ± 0.0698 0.5268 ± 0.0661 0.6408 ± 0.0723 m-CAE 0.8446 ± 0.0361 0.5727 ± 0.0215 0.5600 ± 0.0482 0.6562 ± 0.0767 mCtx-CAE 0.8648 ± 0.0258 0.6423 ± 0.0452 0.5655 ± 0.0228 0.6734 ± 0.0562
下載: 導出CSV
-
JOHNSON A E W, GHASSEMI M M, NEMATI S, et al. Machine learning and decision support in critical care[J]. Proceedings of the IEEE, 2016, 104(2): 444–466. doi: 10.1109/JPROC.2015.2501978 RAVI D, WONG C, DELIGIANNI F, et al. Deep learning for health informatics[J]. IEEE Journal of Biomedical and Health Informatics, 2017, 21(1): 4–21. doi: 10.1109/JBHI.2016.2636665 BOOSTANI R, KARIMZADEH F, and NAMI M. A comparative review on sleep stage classification methods in patients and healthy individuals[J]. Computer Methods and Programs in Biomedicine, 2017, 140: 77–91. doi: 10.1016/j.cmpb.2016.12.004 YUAN Ye, XUN Guangxu, JIA Kebin, et al. A multi-view deep learning framework for EEG seizure detection[J]. IEEE Journal of Biomedical and Health Informatics, 2019, 23(1): 83–94. doi: 10.1109/JBHI.2018.2871678 ACAR E, LEVIN-SCHWARTZ Y, CALHOUN V D, et al. Tensor-based fusion of EEG and fMRI to understand neurological changes in schizophrenia[C]. 2017 IEEE International Symposium on Circuits and Systems, Baltimore, USA, 2017: 1–4. JIA Xiaowei, LI Kang, LI Xiaoyi, et al. A novel semi-supervised deep learning framework for affective state recognition on EEG signals[C]. 2014 IEEE International Conference on Bioinformatics and Bioengineering, Boca Raton, USA, 2014: 30–37. L?NGKVIST M, KARLSSON L, and LOUTFI A. A review of unsupervised feature learning and deep learning for time-series modeling[J]. Pattern Recognition Letters, 2014, 42: 11–24. doi: 10.1016/j.patrec.2014.01.008 HOLZINGER A. Machine Learning for Health Informatics[M]. Cham: Springer, 2016: 161–182. SUPRATAK A, LI Ling, and GUO Yike. Feature extraction with stacked autoencoders for epileptic seizure detection[C]. The 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Chicago, USA, 2014: 4184–4187. YAN Bo, WANG Yong, LI Yuheng, et al. An EEG signal classification method based on sparse auto-encoders and support vector machine[C]. 2016 IEEE/CIC International Conference on Communications in China, Chengdu, China, 2016: 1–6. LIN Qin, YE Shuqun, HUANG Xiumei, et al. Classification of epileptic EEG signals with stacked sparse autoencoder based on deep learning[C]. The 12th International Conference on Intelligent Computing, Lanzhou, China, 2016: 802–810. YANG Jianli, BAI Yang, LI Guojun, et al. A novel method of diagnosing premature ventricular contraction based on sparse auto-encoder and softmax regression[J]. Bio-medical Materials and Engineering, 2015, 26(S1): S1549–S1558. doi: 10.3233/BME-151454 XUN Guangxu, JIA Xiaowei, and ZHANG Aidong. Detecting epileptic seizures with electroencephalogram via a context-learning model[J]. BMC Medical Informatics and Decision Making, 2016, 16(Suppl 2): 70. doi: 10.1186/s12911-016-0310-7 LI Xiaoyi, JIA Xiaowei, XUN Guangxu, et al. Improving EEG feature learning via synchronized facial video[C]. 2015 IEEE International Conference on Big Data, Santa Clara, USA, 2015: 843–848. YUAN Ye, XUN Guangxu, SUO Qiuling, et al. Wave2Vec: Deep representation learning for clinical temporal data[J]. Neurocomputing, 2019, 324: 31–42. doi: 10.1016/j.neucom.2018.03.074 YUAN Ye, XUN Guangxu, JIA Kebin, et al. A multi-context learning approach for EEG epileptic seizure detection[J]. BMC Systems Biology, 2018, 12(6): 47–57. doi: 10.1186/s12918-018-0626-2 ZHANG Junming, WU Yan, BAI Jing, et al. Automatic sleep stage classification based on sparse deep belief net and combination of multiple classifiers[J]. Transactions of the Institute of Measurement and Control, 2016, 38(4): 435–451. doi: 10.1177/0142331215587568 YULITA I N, FANANY M I, and ARYMURTHY A M. Sequence-based sleep stage classification using conditional neural fields[J]. arXiv preprint arXiv:1610.01935 , 2016. L?NGKVIST M, KARLSSON L, and LOUTFI A. Sleep stage classification using unsupervised feature learning[J]. Advances in Artificial Neural Systems, 2012, 2012: 107046. doi: 10.1155/2012/107046 MASCI J, MEIER U, CIRE?AN D, et al. Stacked convolutional auto-encoders for hierarchical feature extraction[C]. The 21st International Conference on Artificial Neural Networks, Espoo, Finland, 2011: 52–59. HINTON G E and SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786): 504–507. doi: 10.1126/science.1127647 MIKOLOV T, SUTSKEVER I, CHEN Kai, et al. Distributed representations of words and phrases and their compositionality[C]. The 26th International Conference on Neural Information Processing Systems, Lake Tahoe, USA, 2013: 3111–3119. CHOI E, BAHADORI M T, SEARLES E, et al. Multi-layer representation learning for medical concepts[C]. The 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco, USA, 2016: 1495–1504. SHOEB A H. Application of machine learning to epileptic seizure onset detection and treatment[D]. [Ph.D. dissertation], Massachusetts Institute of Technology, 2009. GOLDBERGER A L, AMARAL L A N, GLASS L, et al. PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals[J]. Circulation, 2000, 101(23): E215–E220. doi: 10.1161/01.CIR.101.23.e215 FAWCETT T. An introduction to ROC analysis[J]. Pattern Recognition Letters, 2006, 27(8): 861–874. doi: 10.1016/j.patrec.2005.10.010 DAVIS J and GOADRICH M. The relationship between precision-recall and ROC curves[C]. The 23rd International Conference on Machine Learning, Pittsburgh, USA, 2006: 233–240. HE Haibo and GARCIA E A. Learning from imbalanced data[J]. IEEE Transactions on Knowledge and Data Engineering, 2009, 21(9): 1263–1284. doi: 10.1109/TKDE.2008.239 ZEILER M D. ADADELTA: An adaptive learning rate method[J]. arXiv preprint arXiv:1212.5701, 2012. -
計量
- 文章訪問數(shù): 4763
- HTML全文瀏覽量: 1558
- PDF下載量: 153
- 被引次數(shù): 0
下載:
