Automated non-contact detection of central apneas using video. (January 2020)

• Record Type: Journal Article
• Authors: Geertsema, Evelien E.; Visser, Gerhard H.; Sander, Josemir W.; Kalitzin, Stiliyan N.
• Is Part Of: Biomedical signal processing and control. Volume 55(2020)
• Publication Date: 2020-01
• Subjects: Epilepsy -- Event detection -- Pattern recognition -- SUDEP -- Video analysis; 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.2019.101658
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: An automated non-contact algorithm for detection of central apneas is proposed. It detects cessation of oscillatory movements of breathing in absence of body motion. All apnea episodes were detected in the signals from at least one camera angle. Integrating camera inputs capturing different angles improved detection sensitivity. >99% specificity was achieved with individual cameras and integrated camera inputs. Abstract: Central apneas occurring in the aftermath of epileptic seizures may lead to sudden death. Contact-sensors currently used to detect apneas are not always suitable or tolerated. We developed a robust automated non-contact algorithm for real-time detection of central apneas using video cameras. One video recording with simulated apneas and nine with real-life apneas associated with epileptic seizures, each recorded from 3 to 4 angles, were used to develop the algorithm. Videos were preprocessed using optical flow, from which translation, dilatation and shear rates were extracted. Presence of breathing motions was quantified in the time-frequency spectrum by calculating the relative power in the respiratory range (0.1–1 Hz). Sigmoid modulation was calculated over different scales to quantify sigmoid-like drops in respiratory range power. Each sigmoid modulation maximum constitutes a possible apnea event. Two event features were calculated to enable distinction between apnea events and movements: modulation maximum amplitude and total spectral power modulation at the time of the event. An ensemble support vector machine was trained to classify events using a bagging procedure and validated in a leave-one-subject-out cross validation procedure. All apnea episodes were detected in the signals from at least one camera angle. Integrating camera inputs capturing different angles increased overall detection sensitivity (>90%). Overall detection specificity of >99% was achieved with both individual cameras and integrated camera inputs. These results suggest that it is feasible to detect central apneas automatically in video, using this algorithm. When validated, the algorithm might be used as an online remote apnea detector for safety monitoring.