A non-contact camera-based method for respiratory rhythm extraction. (April 2021)

• Record Type: Journal Article
• Authors: Mateu-Mateus, M.; Guede-Fernández, F.; Rodriguez-Ibáñez, N.; García-González, M.A.; Ramos-Castro, J.; Fernández-Chimeno, M.
• Is Part Of: Biomedical signal processing and control. Volume 66(2021)
• Publication Date: 2021-04
• Subjects: Non-contact -- Camera-based -- Respiration -- Computer vision; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102443
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: The aim of this work is to present a non-contact video-based method for respiratory rhythm extraction. The method makes use of a consumer-grade RGB camera, and it is based on computer vision algorithms to detect and track a custom pattern placed on the thorax of the subject. The respiratory signal is extracted by computing the changes in the position of the detected pattern through time. The method has been validated by comparing the extracted respiratory signal versus the one obtained with a reference method in adult population. The reference method was an inductive thorax plethysmography system (Respiband system from BioSignalsPlux™). 21 healthy subjects were measured and four tests were performed for each subject. The respiratory signals and its respiratory cycles were extracted. To characterise the error, the respiratory cycles were assessed with: the Fisher intra-class correlation (ICC), mean absolute error (MAE), the mean absolute percentage error (MAPE) and four Bland-Altman plots were obtained. The results show a > 0.9 correlation for controlled respiration and > 0.85 for unconstrained respiration between the proposed method and the reference method, with low error results (MAPE < 4 % for constrained respiration and < 6 % for unconstrained respiration) and with a high sensitivity when detecting the respiratory cycles ( > 94 % in all cases). From the obtained results we can conclude that the proposed algorithm is adequate to acquire the respiratory signal for rhythm extraction, in real-time with a high performance when compared with the reference method, and that it could be applied to real-life situations.