Martin Bachler

On Entropy-Based Data Mining

In the real world, we are confronted not only with complex and high-dimensional data sets, but usually with noisy, incomplete and uncertain data, where the application of traditional methods of knowledge discovery and data mining always entail the danger of modeling artifacts. Originally, information entropy was introduced by Shannon (1949), as a measure of uncertainty in the data. But up to the present, there have emerged many different types of entropy methods with a large number of different purposes and possible application areas. In this paper, we briefly discuss the applicability of entropy methods for the use in knowledge discovery and data mining, with particular emphasis on biomedical data. We present a very short overview of the state-of-the-art, with focus on four methods: Approximate Entropy (ApEn), Sample Entropy (SampEn), Fuzzy Entropy (FuzzyEn), and Topological Entropy (FiniteTopEn). Finally, we discuss some open problems and future research challenges.

Non-invasive wave reflection quantification in patients with reduced ejection fraction

The non-invasive quantification of arterial wave reflection is an increasingly important concept in cardiovascular research. It is commonly based on pulse wave analysis (PWA) of aortic pressure. Alternatively, wave separation analysis (WSA) considering both aortic pressure and flow waveforms can be applied. Necessary estimates of aortic flow can be measured by Doppler ultrasound or provided by mathematical models. However, this approach has not been investigated intensively up to now in subjects developing systolic heart failure characterized by highly reduced ejection fraction (EF). We used non-invasively generated aortic pressure waveforms and Doppler flow measurements to derive wave reflection parameters in 61 patients with highly reduced and 122 patients with normal EF. Additionally we compared these readings with estimates from three different flow models known from literature (triangular, averaged, Windkessel). After correction for confounding factors, all parameters of wave reflection (PWA and WSA) were comparable for patients with reduced and normal EF. Wave separations assessed with the Windkessel based model were similar to those derived from Doppler flow in both groups. The averaged waveform performed poorer in reduced than in normal EF, whereas triangular flow represented a better approximation for reduced EF. Overall, the non-invasive assessment of WSA parameters based on mathematical models compared to ultrasound seems feasible in patients with reduced EF.

Online and offline determination of QT and PR interval and QRS duration in electrocardiography

Duration and dynamic changes of QT and PR intervals as well as QRS complexes of ECG measurements are well established parameters in monitoring and diagnosis of cardiac diseases. Since automated annotations show numerous advantages over manual methods, the aim was to develop an algorithm suitable for online (real time) and offline ECG analysis. In this work we present this algorithm, its verification and the development process. The algorithm detects R peaks based on the amplitude, the first derivative and local statistic characteristics of the signal. Classification is performed to distinguish premature ventricular contractions from normal heartbeats. To improve the accuracy of the subsequent detection of QRS complexes, P and T waves, templates are built for each class of heartbeats. Using a continuous integration system, the algorithm was automatically verified against PhysioNet databases and achieved a sensitivity of 98.2% and a positive predictive value of 98.7%, respectively.

Long-term monitoring of cardiorespiratory patterns in drug-resistant epilepsy.

Sudden unexplained death in epilepsy (SUDEP) during inpatient electroencephalography (EEG) monitoring has been a rare but potentially preventable event, with associated cardiopulmonary markers. To date, no systematic evaluation of alarm settings for a continuous pulse oximeter (SpO2) has been performed. In addition, evaluation of the interrelationship between the ictal and interictal states for cardiopulmonary measures has not been reported.

Determinants and covariates of central pressures and wave reflections in systolic heart failure

BACKGROUND In general, higher blood pressure levels and increased central pulsatility are indicators for increased cardiovascular risk. However, in systolic heart failure (SHF), this relationship is reversed. Therefore, the aim of this work is to compare pulsatile hemodynamics between patients with SHF and controls and to clarify the relationships between measures of cardiac and arterial function in the two groups. METHODS We used parameters derived from angiography, echocardiography, as well as from pulse wave analysis (PWA) and wave separation analysis (WSA) based on non-invasively assessed pressure and flow waves to quantify cardiac function, aortic stiffness and arterial wave reflection in 61 patients with highly reduced (rEF) and 122 matched control-patients with normal ejection fraction (nEF). RESULTS Invasively measured pulse wave velocity was comparable between the groups (8.6/8.05 m/s rEF/nEF, P = 0.24), whereas all measures derived by PWA and WSA were significantly decreased (augmentation index: 18.1/24.8 rEF/nEF, P < 0.01; reflection magnitude: 56.3/62.1 rEF/nEF, P < 0.01). However, these differences could be explained by the shortened ejection duration (ED) in rEF (ED: 269/308 ms rEF/nEF, P < 0.01; AIx: 22.2/22.8 rEF/nEF, P = 0.7; RM: 59.3/60.6 rEF/nEF, P = 0.47 after adjustment for ED). Ventricular function was positively associated with central pulse pressures in SHF in contrast to no or even a slightly negative association in controls. CONCLUSIONS The results suggest that the decreased measures of pulsatile function may be caused by impaired systolic function and altered interplay of left ventricle and vascular system rather than by a real reduction of wave reflections or aortic stiffness in SHF.

Challenging Recently Published Parameter Sets for Entropy Measures in Risk Prediction for End-Stage Renal Disease Patients

Heart rate variability (HRV) analysis is a non-invasive tool for assessing cardiac health. Entropy measures quantify the chaotic properties of HRV, but they are sensitive to the choice of their required parameters. Previous studies therefore have performed parameter optimization, targeting solely their particular patient cohort. In contrast, this work aimed to challenge entropy measures with recently published parameter sets, without time-consuming optimization, for risk prediction in end-stage renal disease patients. Approximate entropy, sample entropy, fuzzy entropy, fuzzy measure entropy, and corrected approximate entropy were examined. In total, 265 hemodialysis patients from the ISAR (rISk strAtification in end-stage Renal disease) study were analyzed. Throughout a median follow-up time of 43 months, 70 patients died. Fuzzy entropy and corrected approximate entropy (CApEn) provided significant hazard ratios, which remained significant after adjustment for clinical risk factors from literature if an entropy maximizing threshold parameter was chosen. Revealing results were seen in the subgroup of patients with heart disease (HD) when setting the radius to a multiple of the data’s standard deviation ( r = 0.2 · σ ); all entropies, except CApEn, predicted mortality significantly and remained significant after adjustment. Therefore, these two parameter settings seem to reflect different cardiac properties. This work shows the potential of entropy measures for cardiovascular risk stratification in cohorts the parameters were not optimized for, and it provides additional insights into the parameter choice.

Associations of Novel and Traditional Vascular Biomarkers of Arterial Stiffness: Results of the SAPALDIA 3 Cohort Study

Background and Objectives There is a lack of evidence concerning associations between novel parameters of arterial stiffness as cardiovascular risk markers and traditional structural and functional vascular biomarkers in a population-based Caucasian cohort. We examined these associations in the second follow-up of the Swiss Cohort Study on Air Pollution and Lung and Heart Diseases in Adults (SAPALDIA 3). Methods Arterial stiffness was measured oscillometrically by pulse wave analysis to derive the cardio-ankle vascular index (CAVI), brachial-ankle (baPWV) and aortic pulse wave velocity (aPWV), and amplitude of the forward and backward wave. Carotid ultrasonography was used to measure carotid intima-media thickness (cIMT) and carotid lumen diameter (LD), and to derive a distensibility coefficient (DC). We used multivariable linear regression models adjusted for several potential confounders for 2,733 people aged 50–81 years. Results CAVI, aPWV and the amplitude of the forward and backward wave were significant predictors of cIMT (p < 0.001). All parameters were significantly associated with LD (p < 0.001), with aPWV and the amplitude of the forward wave explaining the highest proportion of variance (2%). Only CAVI and baPWV were significant predictors of DC (p < 0.001), explaining more than 0.3% of the DC variance. Conclusion We demonstrated that novel non-invasive oscillometric arterial stiffness parameters are differentially associated with specific established structural and functional local stiffness parameters. Longitudinal studies are needed to follow-up on these cross-sectional findings and to evaluate their relevance for clinical phenotypes.

Entropy-Based Data Mining on the Example of Cardiac Arrhythmia Suppression

Heart rate variability (HRV) is the variation of the time interval between consecutive heartbeats and depends on the extrinsic regulation of the heart rate. It can be quantified using nonlinear methods such as entropy measures, which determine the irregularity of the time intervals.

Feasibility of oscillometric aortic pressure and stiffness assessment using the VaSera VS-1500: comparison with a common tonometric method.

ObjectivesA number of operator-independent oscillometric devices to measure hemodynamics and arterial stiffness became available recently, but some and in particular VaSera VS-1500 do not provide estimates of aortic pressures and aortic pulse wave velocity (aPWV). The aim of this work was the retrospective application of the ARCSolver algorithm to pulse wave signals acquired with the VaSera VS-1500 device to estimate central systolic blood pressure (cSBP) and aPWV. Materials and methodsARCSolver estimates of cSBP and aPWV, on the basis of brachial cuff measurements, were compared pair-wise with results from the tonometric SphygmoCor device in 68 individuals (mean age 51±18 years). We used variation estimates, correlation coefficients, age group-related t-tests, and the Bland–Altman method to analyze the reproducibility and agreement of the two methods. ResultscSBP reproducibility expressed as variability was 14.9% for ARCSolver and 11.6% for SphygmoCor. PWV reproducibility was better for ARCSolver, with a variation estimate of 6.5%, compared with 20.9% using SphygmoCor. The mean cSBP difference was 0.5 mmHg (SD 6.9 mmHg) and 0.32 m/s (SD 1.20 m/s) for PWV, respectively. The age-related differences between ARCSolver and SphygmoCor are in line with previous studies. Bland–Altman plots showed considerable agreement between the two methods without signs of systematic bias. ConclusionThese results show that the combined application of the ARCSolver method with the VaSera VS-1500 device is feasible and the results are comparable with tonometric determination of cSBP and aPWV. This successful application of the ARCSolver may potentially help to improve cardiovascular risk stratification and prevention at an early stage in a community setting.

Implementation and verification of an enhanced algorithm for the automatic computation of RR-interval series derived from 24 h 12-lead ECGs

An important tool in early diagnosis of cardiac dysfunctions is the analysis of electrocardiograms (ECGs) obtained from ambulatory long-term recordings. Heart rate variability (HRV) analysis became a significant tool for assessing the cardiac health. The usefulness of HRV assessment for the prediction of cardiovascular events in end-stage renal disease patients was previously reported. The aim of this work is to verify an enhanced algorithm to obtain an RR-interval time series in a fully automated manner. The multi-lead corrected R-peaks of each ECG lead are used for RR-series computation and the algorithm is verified by a comparison with manually reviewed reference RR-time series. Twenty-four hour 12-lead ECG recordings of 339 end-stage renal disease patients from the ISAR (rISk strAtification in end-stage Renal disease) study were used. Seven universal indicators were calculated to allow for a generalization of the comparison results. The median score of the indicator of synchronization, i.e. intraclass correlation coefficient, was 96.4% and the median of the root mean square error of the difference time series was 7.5 ms. The negligible error and high synchronization rate indicate high similarity and verified the agreement between the fully automated RR-interval series calculated with the AIT Multi-Lead ECGsolver and the reference time series. As a future perspective, HRV parameters calculated on this RR-time series can be evaluated in longitudinal studies to ensure clinical benefit.