ECG-derived respiration in atrial fibrillation
Abstract: In past years, the demand for an indirect extraction of respiration and the interest in a joint study of respiratory and cardiac systems represented the driving forces for the development of Electrocardiogram Derived Respiration (EDR) algorithms. The main advantage of them consists in deriving a sufficiently reliable surrogate respiratory signal by only exploiting the normal electrocardiogram (ECG) equipment, without requiring the common devices used to record respiration, which are cumbersome and expensive, besides to possibly interfere with natural breathing. However, the validity of EDR methods has been mainly demonstrated on healthy subjects and in certain clinical applications. At the present time, the possibility to extract non-invasively the respiratory signal during arrhythmia, by applying an EDR method, has never been explored on a sufficiently large dataset and with a systematic study. Atrial Fibrillation (AF) represents the most common arrhythmia, characterized by a fast and irregular beating with an increasing incidence that is especially prominent in the developed world. Therefore, this master thesis aims to verify the feasibility of extracting the respiratory rate from ECG during Atrial Fibrillation (AF). Four EDR methods are implemented and evaluated, by selecting only among the techniques based on respiration-induced variations in beat-to-beat morphology, since the abnormal heart rhythm in AF does not lend itself to be used for deriving respiration. The different techniques are tested on a dataset of 49 patients, containing a two non-orthogonal leads ECG recording and a simultaneous belt respiratory signal for each of them. The recordings are of about ten minutes long and were acquired in rest phase. A code workflow has been designed to handle the characteristics of the signals and reliably estimate the respiratory rate from the surrogate respiratory signal and from the reference signal. Since the fibrillation activity may mask the respiratory information contained in the beat morphology, an important stage of the workflow is represented by the subtraction of the fibrillation signal from the QRS interval. The actual benefit of this step is tested by comparing the methods performances by selectively allowing or not its performing. The estimation accuracy of each method is assessed by comparing the respiratory rate estimates from EDR signal and reference signal in terms of mean absolute and relative intrasubject error, percentage of the record duration where an estimate is given from both signals and RMS error computed on the entire dataset. The results do not point out any improvement in the performances of the methods after removing the f-waves from the QRS complex. This do not ensure that different results may be observed by applying another technique, since the method applied in this thesis suffers of sensitivity to noise that should be further investigated. From the comparison of the results of the methods, it turned out that the methods that extract respiratory information independently from the two leads outperformed the other ones, which combine this information from both leads and derive only one rotation angle series. The method based on QRS slopes and R waves angles estimated the respiratory frequency of the subjects in the dataset more accurately than did the other tested methods, achieving a mean intrasubject error μ=0.0227±0.0217 (8.45% ± 8.83 %). In general, all the tested methods achieved estimation errors higher with respect to previous studies on healthy subjects, but still comparable as order of magnitude. This indicates that further studies on the extraction of electrocardiography derived respiration in patients with atrial fibrillation are justified.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)