Cardiorespiratory monitoring is crucial for the diagnosis and management of multiple conditions such as stress and sleep disorders. Therefore, the development of ambulatory systems providing continuous, comfortable, and inexpensive means for monitoring represents an important research topic. Several techniques have been proposed in the literature to derive respiratory information from the ECG signal. Ten methods to compute single-lead ECG-derived respiration (EDR) were compared under multiple conditions, including different recording systems, baseline wander, normal and abnormal breathing patterns, changes in breathing rate, noise, and artifacts. Respiratory rates, wave morphology, and cardiorespiratory information were derived from the ECG and compared to those extracted from a reference respiratory signal. Three datasets were considered for analysis, involving a total 59 482 one-min, single-lead ECG segments recorded from 156 subjects. The results indicate that the methods based on QRS slopes outperform the other methods. This result is particularly interesting since simplicity is crucial for the development of ECG-based ambulatory systems.

Respiratory sinus arrhythmia (RSA) is one of the forms of cardiorespiratory coupling. It has been suggested as a potential biomarker for diverse illnesses and conditions. In general, methods for non-invasive quantification of the RSA combine information from heart rate variability (HRV) and respiratory signals. Abnormal beats, which commonly occur in different populations, alter the reliability of the HRV and thus hinder the quantification of the RSA. To overcome this problem, several methods for detection and correction of irregular beats have been reported in literature. However, the effect of each of these methods on the quantification of the RSA is not well understood yet. For this reason, an approach that avoids this step might be useful. This paper presents an alternative based on robust regression models. For comparison purposes, an algorithm to detect and correct for irregular beats, in combination with a state-of-the-art RSA estimate, are tested. A similar performance is achieved with both approaches. These results show that the proposed robust methodology is able to capture the strength of the RSA, even when irregular beats are present, avoiding the irregularities correction step.

A real-life validation of a system for simultaneous acquisition of capacitively-coupled ECG (ccECG) and capacitively-coupled bioimpedance (ccBioz) is presented. The heart rate (HR) and respiration rate (RR) estimation performance was evaluated using polysomnography (PSG) signals as ground-truth, in recordings from 28 patients with suspected obstructive sleep apnea (OSA). A ccECG beat detection sensitivity of 98.4% and an R-R interval mean absolute error (MAE) of 17.1 ms were achieved when applying quality-based algorithms. RR MAE values of 3.48 and 6.37 breaths per minute were also achieved when using two different RR extraction methods. High similarity between unobtrusive signals and PSG ground-truth was observed, with a correlation between ccECG and psgECG of 91.5% and a correlation between ccBioz and PSG thoracic belt (TB) of 89.5%. Even in episodes containing OSA events, the characteristic respiration behavior of TB signals was also observed in the ccBioz signals. This shows the potential of ccECG and ccBioz for use in long-term monitoring without adding discomfort to the patient or user. Sleep-related applications as well as more generic cardiorespiratory monitoring in (patient) beds are obvious applications, but also other daily life monitoring can be done using a similar approach (e.g. in seats).

Sleep apnea is a common sleep disorder, whose diagnosis can strongly benefit from home-based screening. As the total sleep time is essential to assess the sleep apnea severity, a sleep-wake classifier was developed based on heart rate and respiration. These two signals were selected as they can be measured using unobtrusive sensors. A 1D convolutional neural network (CNN) was designed to classify 30s epochs of tachograms and respiratory inductance plethysmography (RIP) signals. The input based on beat-to-beat variability allows the use of different sensor types. A dataset of 56 patients with an apnea-hypopnea index (AHI) below 10 was used to train and validate the network. This CNN was applied to an independent test set of ECG and RIP signals of 25 subjects. Of these, 8 subjects were simultaneously monitored using an unobtrusive capacitive-coupled ECG (ccECG) sensor integrated in a mattress. Artefact removal and data correction was performed on this acquired data. The performance on the independent dataset of ECG and RIP is comparable to state-of-the-art, with ? = 0.48. However, application on the ccECG data resulted in a drop in performance, with ? = 0.30. This was caused by a low amount of remaining wake epochs after data cleaning. Importantly, the network classified 30s segments of sleep apnea patients, without relying on past or future information for feature extraction.