Detection of respiratory rate using a classifier of waves in the signal from a FBG-based vital signs sensor

• Published in: Computer methods and programs in biomedicine 2019-08, Vol.177 (C), p.31-38
• Main Authors: Krej, Mariusz, Baran, Paulina, Dziuda, Łukasz
• Format: Article
• Language: English
• Subjects: Adult; Algorithms; Discriminant Analysis; Electronic Data Processing; Equipment Design; Female; Fiber Bragg gratings (FBGs); Fiber-optic sensors; Heart Rate; Humans; Machine Learning; Magnetic Resonance Imaging; Magnetic resonance imaging (MRI); Male; Middle Aged; Monitoring, Physiologic - methods; Reproducibility of Results; Respiratory Rate; Signal processing; Young Adult
• ISSN: 0169-2607; 1872-7565
• DOI: 10.1016/j.cmpb.2019.05.014

•Automatic, continuous extraction of respiratory rate from optical signals.•The scheme uses machine learning methods, does not require sophisticated hardware.•Clinically satisfactory results with an RMSE of 1.48 rpm.•The scheme can be easily integrated into a system for monitoring of MRI patients. Monitoring of changes in respiratory rate provides information on a patient's psychophysical state. This paper presents a respiratory rate detection method based on analysis of signals from a fiber Bragg grating (FBG)-based sensor. The detection method is based on a system of software blocks that identify notches in the signal waveforms, determine their parameters, and then transmit them to the classifier, which decides which of them are the characteristic waves of the respiratory cycle. The classifier of respiratory waves was developed by means of machine learning methods and using the training data obtained from 10 volunteers (7 males, 3 females, age: 41.1 ± 8.28 years, weight: 73.6 ± 15.25 kg, height 173.5 ± 6.43 cm), who were lying in the tube of a 3-Tesla magnetic resonance imaging (MRI) scanner. In the verification study, aimed at assessing the performance of the method for detecting respiratory rate, 15 subjects (14 males, 1 female, age: 20.2 ± 3.00 years, weight: 75.47 ± 10.58 kg, height 179.13 ± 6.27 cm) were involved. Clinically satisfactory results of respiratory rate detection were obtained: root mean square error of 1.48 rpm and the limits of agreement at -2.73 rpm and 3.04 rpm. The results indicate a high efficiency of the classifier, i.e., sensitivity: 96.50 ± 3.44%, precision: 95.42 ± 2.84%, and accuracy: 92.99 ± 3.37%. The all-dielectric sensor acquires the respiration curve and the proposed scheme of computation enables for extracting respiratory rate automatically and continuously. This scheme based on machine learning procedures will be integrated into a system to facilitate non-invasive continuous monitoring of MRI patients.