Dorthe Bodholt Saadi

Classification of acute stress using linear and non-linear heart rate variability analysis derived from sternal ECG

Chronic stress detection is an important factor in predicting and reducing the risk of cardiovascular disease. This work is a pilot study with a focus on developing a method for detecting short-term psychophysiological changes through heart rate variability (HRV) features. The purpose of this pilot study is to establish and to gain insight on a set of features that could be used to detect psychophysiological changes that occur during chronic stress. This study elicited four different types of arousal by images, sounds, mental tasks and rest, and classified them using linear and non-linear HRV features from electrocardiograms (ECG) acquired by the wireless wearable ePatch® recorder. The highest recognition rates were acquired for the neutral stage (90%), the acute stress stage (80%) and the baseline stage (80%) by sample entropy, detrended fluctuation analysis and normalized high frequency features. Standardizing non-linear HRV features for each subject was found to be an important factor for the improvement of the classification results.

Automatic Real-Time Embedded QRS Complex Detection for a Novel Patch-Type Electrocardiogram Recorder

Cardiovascular diseases are projected to remain the single leading cause of death globally. Timely diagnosis and treatment of these diseases are crucial to prevent death and dangerous complications. One of the important tools in early diagnosis of arrhythmias is analysis of electrocardiograms (ECGs) obtained from ambulatory long-term recordings. The design of novel patch-type ECG recorders has increased the accessibility of these long-term recordings. In many applications, it is furthermore an advantage for these devices that the recorded ECGs can be analyzed automatically in real time. The purpose of this study was therefore to design a novel algorithm for automatic heart beat detection, and embed the algorithm in the CE marked ePatch heart monitor. The algorithm is based on a novel cascade of computationally efficient filters, optimized adaptive thresholding, and a refined search back mechanism. The design and optimization of the algorithm was performed on two different databases: The MIT-BIH arrhythmia database (Se = 99.90%, P+ = 99.87) and a private ePatch training database (Se = 99.88%, P+ = 99.37%). The offline validation was conducted on the European ST-T database (Se = 99.84%, P+ = 99.71%). Finally, a double-blinded validation of the embedded algorithm was conducted on a private ePatch validation database (Se = 99.91%, P+ = 99.79%). The algorithm was thus validated with high clinical performance on more than 300 ECG records from 189 different subjects with a high number of different abnormal beat morphologies. This demonstrates the strengths of the algorithm, and the potential for this embedded algorithm to improve the possibilities of early diagnosis and treatment of cardiovascular diseases.

A clinical study of short-term sternal photoplethysmography: Recordings from patients with obstructive airways diseases

Traditionally, measurements of the oxygen saturation (SpO2) has been confound to the extremities. In this study, we therefore investigated the possibility for reliable estimation of clinically relevant SpO2 levels from photoplethysmography (PPG) obtained on the sternum of patients with obstructive airway diseases. We initiated the study with a calibration of a prototype sternal PPG sensor. In accordance with the ISO 80601-2-61:2011 guidelines, the calibration was conducted as a controlled desaturation study. We obtained a calibration accuracy of 1.75% which is well within the clinically and commercially accepted range. We then compared the SpO2 levels simultaneously obtained from the sternal PPGs and a commercially available finger pulse oximeter on 28 admitted patients with either asthma or Chronic Obstructive Pulmonary Disease (COPD). The Pearson correlation between the SpO2 levels estimated from the two body locations was found to be 0.89 (p<;0.05) and the mean system bias was only 0.052% with upper and lower limits of agreement of 2.5% and -2.4%, respectively. This finding is very promising for the future design of new sternum based patch technologies that might be able to provide continuous estimates of the SpO2 levels on critically or chronically ill patients.Traditionally, measurements of the oxygen saturation (SpO2) has been confound to the extremities. In this study, we therefore investigated the possibility for reliable estimation of clinically relevant SpO2 levels from photoplethysmography (PPG) obtained on the sternum of patients with obstructive airway diseases. We initiated the study with a calibration of a prototype sternal PPG sensor. In accordance with the ISO 80601-2-61:2011 guidelines, the calibration was conducted as a controlled desaturation study. We obtained a calibration accuracy of 1.75% which is well within the clinically and commercially accepted range. We then compared the SpO2 levels simultaneously obtained from the sternal PPGs and a commercially available finger pulse oximeter on 28 admitted patients with either asthma or Chronic Obstructive Pulmonary Disease (COPD). The Pearson correlation between the SpO2 levels estimated from the two body locations was found to be 0.89 (p<;0.05) and the mean system bias was only 0.052% with upper and lower limits of agreement of 2.5% and -2.4%, respectively. This finding is very promising for the future design of new sternum based patch technologies that might be able to provide continuous estimates of the SpO2 levels on critically or chronically ill patients.

Long-term quasi-continuous oxygen saturation levels obtained from sternal photoplethysmography on patients with obstructive lung diseases

Calculation of long-term quasi-continuous oxygen saturation (SpO2) levels is highly relevant for critically ill patients. The purpose of this study is therefore to conduct a preliminary investigation of the clinical reliability of long-term photoplethysmography (PPG) recordings obtained from the sternum of patients admitted to the hospital with obstructive lung diseases. Due to the lack of a gold standard reference that is suitable for long-term monitoring without interfering with the patients activity level, we extracted reliable segments based on knowledge from the basic pulse oximeter theory as well as knowledge about the inherent physiological regulation of the SpO2 levels. We included 15 admitted patients who were monitored with a prototype of a sternal PPG sensor for approximately 20 hours. On average, we found that clinically reliable SpO2 levels could be calculated for 58% of the recording time. Furthermore, the average and standard deviation of the longest period of time with unreliable data was only 23.6 ± 19.38 minutes. This indicates a high potential for quasi-continuous calculation of SpO2 levels from sternal PPGs in many different clinical applications in the future.Calculation of long-term quasi-continuous oxygen saturation (SpO2) levels is highly relevant for critically ill patients. The purpose of this study is therefore to conduct a preliminary investigation of the clinical reliability of long-term photoplethysmography (PPG) recordings obtained from the sternum of patients admitted to the hospital with obstructive lung diseases. Due to the lack of a gold standard reference that is suitable for long-term monitoring without interfering with the patients activity level, we extracted reliable segments based on knowledge from the basic pulse oximeter theory as well as knowledge about the inherent physiological regulation of the SpO2 levels. We included 15 admitted patients who were monitored with a prototype of a sternal PPG sensor for approximately 20 hours. On average, we found that clinically reliable SpO2 levels could be calculated for 58% of the recording time. Furthermore, the average and standard deviation of the longest period of time with unreliable data was only 23.6 ± 19.38 minutes. This indicates a high potential for quasi-continuous calculation of SpO2 levels from sternal PPGs in many different clinical applications in the future.

Investigation of the Minimum Conditions for Reliable Estimation of Clinically Relevant HRV Measures - Introducing a Novel Approach to the Validation of HRV Measurement Systems

The R-peak localization error (jitter) of a heart rate variability (HRV) system has a great impact on the values of the HRV measures. Only a few studies have analyzed this subject and purely done so from the aspect of choice of sampling frequency. In this study we provide an overview of the various factors that comprise the jitter of a system. We propose a method inspired by the field of signal averaged electrocardiography (SAECG) that allows for a quantification of the jitter of any HRV system that records and stores the raw ECG signal. Furthermore, with this method the differences between the HRV measures of the system and HRV measures corresponding to the physiological truth can be quantified. The method is used to obtain the physiologically true R-peak locations of subjects from Physionet’s ‘Normal Sinus Rhythm Database’. The effects of jitter are then analyzed via mathematical modelling for short-term and long-term HRV for various HRV measures. The effects of abnormal beats and missed and false detections are analyzed as well.

Automatic quality classification of entire electrocardiographic recordings obtained with a novel patch type recorder.

Recently, new patch type electrocardiogram (ECG) recorders have reached the market. These new devices possess a number of advantages compared to the traditional Holter recorders. This forms the basis of questions related to benefits and drawbacks of different ambulatory ECG recording techniques. One of the important questions is the ability to obtain high clinical quality of the recordings during the entire monitoring period. It is thus desirable to be able to obtain an automatic estimate of the global quality of entire ECG recordings. The purpose of this pilot study is therefore to design an algorithm for automatic classification of entire ECG recordings into the groups “noisy” and “clean” recordings. This novel algorithm is based on three features and a simple Bayes classifier. The algorithm was tested on 40 ECG recordings in a five-fold cross validation scheme and it obtained an average accuracy of 90% on the test data.