Arxiv:2002.05534V2 [Cs.LG] 21 Dec 2020

Arxiv:2002.05534V2 [Cs.LG] 21 Dec 2020

ABNORMAL RESPIRATORY PATTERNS CLASSIFIER MAY CONTRIBUTE TO LARGE-SCALE SCREENING OF PEOPLE INFECTED WITH COVID-19 IN AN ACCURATE AND UNOBTRUSIVE MANNER Yunlu Wang1, Menghan Hu1, Qingli Li1, Xiao-Ping Zhang3, Guangtao Zhai2,Nan Yao4 1Shanghai Key Labora. of Multidim. Infor. Proce., East China Normal University, China 2Key Laboratory of Artificial Intelligence, Ministry of Education, China 3Department of Electrical, Computer and Biomedical Engineering, Ryerson University, Canada 4 Shanghai Jianglai Data Technology Co., Ltd, China ABSTRACT Tachypnea, Bradypnea, Biots, Cheyne-Stokes and Central- Apnea). The performance of the obtained BI-AT-GRU is Research significance: The extended version of this paper tested by real-world data measured by depth camera, and the has been accepted by IEEE Internet of Things journal (DOI: results show that the proposed model can classify 6 differ- 10.1109/JIOT.2020.2991456), please cite the journal version. ent respiratory patterns with the accuracy, precision, recall During the epidemic prevention and control period, our study and F1 of 94.5%, 94.4%, 95.1% and 94.8% respectively. In can be helpful in prognosis, diagnosis and screening for the comparative experiments, the obtained BI-AT-GRU specific patients infected with COVID-19 (the novel coronavirus) to respiratory pattern classification outperforms the existing based on breathing characteristics. According to the latest state-of-the-art models. The proposed deep model and the clinical research, the respiratory pattern of COVID-19 is dif- modeling ideas have the great potential to be extended to ferent from the respiratory patterns of flu and the common large scale applications such as public places, sleep scenario, cold. One significant symptom that occurs in the COVID-19 and office environment. is Tachypnea. People infected with COVID-19 have more rapid respiration. Our study can be utilized to distinguish var- Index Terms— Breathing pattern, physiological signal ious respiratory patterns and our device can be preliminarily measurement, recurrent neural network, remote monitoring put to practical use. Demo videos of this method working in situations of one subject and two subjects can be downloaded online (https://doi.org/10.6084/m9.figshare.11493666.v1). 1. INTRODUCTION Research details: Accurate detection of the unexpected abnormal respiratory pattern of people in a remote and un- Respiration is a core physiological process for all living crea- obtrusive manner has great significance. In this work, we tures on earth. A person’s physiological state [1] as well as innovatively capitalize on depth camera and deep learning to emotion [2] and stress [3] may be reflected by representation achieve this goal. The challenges in this task are twofold: the of some respiratory parameters. Therefore, we should pay amount of real-world data is not enough for training to get the attention to respiration. Integrated various respiratory signs, arXiv:2002.05534v2 [cs.LG] 21 Dec 2020 deep model; and the intra-class variation of different types respiratory patterns are able to more comprehensively reflect of respiratory patterns is large and the outer-class variation is the conditions of respiratory activity. Many clinical literatures small. In this paper, considering the characteristics of actual suggested that abnormal respiratory patterns are able to pre- respiratory signals, a novel and efficient Respiratory Simu- dict a few specific diseases [4], providing relatively detailed lation Model (RSM) is first proposed to fill the gap between clues for clinical treatments [5,6]. Unfortunately, these ab- the large amount of training data and scarce real-world data. normal respiratory patterns occur in a way difficult for people Subsequently, we first apply a GRU neural network with to notice themselves. If we could develop a system capable bidirectional and attentional mechanisms (BI-AT-GRU) to of remotely and unobtrusively detecting these unnoticeable classify 6 clinically significant respiratory patterns (Eupnea, abnormal breathing patterns under various scenarios, people who have the diseases may be diagnosed at earliest possible This work is sponsored by the Shanghai Sailing Program time. (No.19YF1414100), the National Natural Science Foundation of China Contact measurement devices are heavy, expensive, and (No.61901172), the STCSM (No.18DZ2270700), and the Science and Technology Commission of Shanghai Municipality (No.19511120100). inconvenient for patients to use [7]. Therefore, non-contact Corresponding author: Menghan Hu ([email protected]) measurement methods are more suitable for detecting ab- normal respiratory patterns. Doppler radar, thermal imaging 2.1. Respiratory patterns technology and camera based on motion detection are usu- ally used for recording respiratory signals. The continual use Respiratory patterns can be divided into normal breath- of radar has the potential risk of the explosion to released ing (Eupnea) and abnormal breathing such as Bradypnea, radiation [8]. Thermal imaging cameras are susceptible to Tachypnea, Biots, Cheyne-Stokes and Central-Apnea. Stan- ambient heat [9, 10]. The method based on motion detection dard waveforms are available at the website (https://doi.org/ is able to solve the above problems, because its measure- 10.6084/m9.figshare.9981236.v2). ment is based on the variation in displacement instead of the changes in pixel intensity. Depth camera is a good method 2.2. Respiratory Simulation Model (RSM) based on motion detection which has been widely utilized in the acquisition and analysis of respiratory signals [11, 12]. Respiration is a cyclic process of inhalation and exhalation, Nonetheless, most relevant work focused on improving the which are reflected in the rise and fall of the waveform. Thus, accuracy of extracting respiratory signals and restraining the respiratory signals measured by non-contact method can be noise in movement [13]. Some paid attention to the detec- approximated by sine wave. Actual measured respiratory sig- tion of abnormal respiratory patterns, but did not classify nals, especially those measured by non-contact method, are them specifically [14]. Thus, the available clinical clues were prone to deviation due to environmental changes, leading to limited. fluctuations in respiratory depth and frequency within a cer- tain range. Affected by body movements in measurements, Deep learning has been extensively used in many fields. signals also tend to have longitudinal and oblique deviation. In terms of respiratory pattern detection, Cho et al. capital- Considering the possible deviation mentioned above, the ac- ized on the convolutional neural networks (CNN) to achieve tual measured respiratory signals can be defined by the fol- the identification of deep breathing [15]. Kim et al. used 1D lowing equation: CNN to classify the respiratory signals estimated by radars into four categories [16]. Therefore, the classification of res- yi2! = ai sin(bix) + ci + dix (1) piratory signals extracted by the non-contact measurement system with the aid of deep learning is a study worth trying where ! is a constant time window, indicating the cycle of and of much significance. In the above study, all the data sets examining a respiratory pattern; i is a variable time, which required for model construction were obtained, though, by is used to represent changes of signals belonging to the same assessing the respiratory activities of the test subjects. This respiratory pattern; the point dividing each period of i can be approach for capturing different types of respiratory patterns called a ‘breakpoint’; ai is respiratory depth; bi is respiratory yields a limited set of data. Were there no large amount of rate; ci and di are longitudinal deviation degree and oblique training data to support, the neural network might not be able deviation degree of respiratory signal, respectively. Gaussian to give full play to its advantages. Furthermore, researchers white noise is added to the simulated respiratory signal to often present studies that the classification models generally make it closer to the real-world measured respiratory signal. adopt the general network architecture in the field of deep By changing values of ai, bi, ci and di in equation (1), learning without specific designs for respiratory pattern clas- different features in the process of respiration can be simu- sification. At the same time, the network is not optimized lated, such as speed, depth and time of apnea. A specific according to the characteristics of the collected data. partial waveform can be generated by randomizing the pa- In this paper, we pioneer an AT-BI-GRU deep neural net- rameters of each waveform in a pre-set range. We combine work for classifying abnormal respiratory patterns. To fill the every partial waveform through breakpoints then specific res- gap between the large amount of training data and scarce real- piratory patterns can be obtained. The random range of pa- world data, we first propose Respiratory Simulation Model rameters is set for the 6 respiratory patterns by reference to (RSM) to generate abundant reliable data. The obtained clas- standard waveforms and real-world waveforms, and it can be sifier validated by 605 real-world data measured by depth acquired in our released RSM code(https://doi.org/10.6084/ camera can be used to achieve the remote and unobtrusive m9.figshare.9978833.v1). To illustrate the process of generat- measurement of respiratory patterns. ing data, we select an abnormal respiratory pattern to demon- strate (https://doi.org/10.6084/m9.figshare.9981311.v1).

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    6 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us