[Paper Review] Abnormal respiratory patterns classifier may contribute to large-scale screening of people infected with COVID-19 in an accurate and unobtrusive manner

Yunlu Wang, Menghan Hu+4 more|arXiv (Cornell University)|Feb 12, 2020

Anomaly Detection Techniques and ApplicationsComputer Science22 references197 citations

TL;DR

The paper proposes BI-AT-GRU with a Respiratory Simulation Model to classify six respiratory patterns from non-contact depth-camera data, achieving 94.5% accuracy on real-world data.

ABSTRACT

Research significance: The extended version of this paper has been accepted by IEEE Internet of Things journal (DOI: 10.1109/JIOT.2020.2991456), please cite the journal version. During the epidemic prevention and control period, our study can be helpful in prognosis, diagnosis and screening for the patients infected with COVID-19 (the novel coronavirus) based on breathing characteristics. According to the latest clinical research, the respiratory pattern of COVID-19 is different from the respiratory patterns of flu and the common cold. One significant symptom that occurs in the COVID-19 is Tachypnea. People infected with COVID-19 have more rapid respiration. Our study can be utilized to distinguish various respiratory patterns and our device can be preliminarily put to practical use. Demo videos of this method working in situations of one subject and two subjects can be downloaded online. Research details: Accurate detection of the unexpected abnormal respiratory pattern of people in a remote and unobtrusive manner has great significance. In this work, we innovatively capitalize on depth camera and deep learning to achieve this goal. The challenges in this task are twofold: the amount of real-world data is not enough for training to get the deep model; and the intra-class variation of different types of respiratory patterns is large and the outer-class variation is small. In this paper, considering the characteristics of actual respiratory signals, a novel and efficient Respiratory Simulation Model (RSM) is first proposed to fill the gap between the large amount of training data and scarce real-world data. The proposed deep model and the modeling ideas have the great potential to be extended to large scale applications such as public places, sleep scenario, and office environment.

Motivation & Objective

• Motivate remote, unobtrusive detection of abnormal respiratory patterns relevant to COVID-19 screening.
• Address data scarcity by generating large synthetic training data via a Respiratory Simulation Model (RSM).
• Develop and validate a deep learning classifier tailored to respiratory waveform characteristics.

Proposed method

• Introduce Respiratory Simulation Model (RSM) to generate diverse synthetic respiratory waveforms with parameters a_i, b_i, c_i, d_i and noise.
• Acquire real-world depth-camera data from 20 subjects imitating six patterns using Kinect v2 and ROI-based depth signals.
• Propose BI-AT-GRU, a bidirectional GRU with an attention mechanism, to classify six respiratory patterns from time-series data.
• In training, use RSM-generated data; in testing, evaluate on depth-camera measured data.
• Train BI-AT-GRU and compare with BI-AT-LSTM, GRU, and LSTM to assess performance.

Experimental results

Research questions

• RQ1 Can a deep neural network trained on simulated respiratory data accurately classify real-world respiratory patterns from depth-camera signals?
• RQ2 Do bidirectional and attention mechanisms improve respiratory pattern classification performance over standard GRU/LSTM models?
• RQ3 What is the classification accuracy of the proposed BI-AT-GRU on real-world depth-camera data for six patterns?

Key findings

• BI-AT-GRU achieves 94.5% accuracy, 94.4% precision, 95.1% recall, and 94.8% F1 on real-world data.
• BI-AT-GRU outperforms BI-AT-LSTM, GRU, and LSTM on the same test set.
• Bidirectional and attention mechanisms improve performance over their non-bidirectional/non-attentional counterparts.
• Classification errors mainly occur when distinguishing Cheyne-Stokes from Central-Apnea due to amplitude-related variability and movement.
• The approach demonstrates feasibility of remote, unobtrusive respiratory pattern classification with potential large-scale applications.