Abstract:

Chronic kidney disease (CKD) affects 13% of the worldwide population and end stage patients often receive haemodialysis treatment to control the electrolyte concentrations. The cardiovascular death rate increases by 10% - 30% in dialysis patients than in general population. To analyse possible links between electrolyte concentration variation and cardiovascular diseases, a continuous non-invasive monitoring tool enabling the estimation of potassium and calcium concentration from features of the ECG is desired. Although the ECG was shown capable of being used for this purpose, the method still needs improvement. In this study, we examine the influence of lead reduction techniques on the estimation results of serum calcium and potassium concentrations.We used simulated 12 lead ECG signals obtained using an adapted Himeno et al. model. Aiming at a precise estimation of the electrolyte concentrations, we compared the estimation based on standard ECG leads with the estimation using linearly transformed fusion signals. The transformed signals were extracted from two lead reduction techniques: principle component analysis (PCA) and maximum amplitude transformation (Max- Amp). Five features describing the electrolyte changes were calculated from the signals. To reconstruct the ionic concentrations, we applied a first and a third order polynomial regression connecting the calculated features and concentration values. Furthermore, we added 30 dB white Gaussian noise to the ECGs to imitate clinically measured signals. For the noisefree case, the smallest estimation error was achieved with a specific single lead from the standard 12 lead ECG. For example, for a first order polynomial regression, the error was 0.0003±0.0767 mmol/l (mean±standard deviation) for potassium and -0.0036±0.1710 mmol/l for calcium (Wilson lead V1). For the noisy case, the PCA signal showed the best estimation performance with an error of -0.003±0.2005 mmol/l for potassium and -0.0002±0.2040 mmol/l for calcium (both first order fit). Our results show that PCA as ECG lead reduction technique is more robust against noise than MaxAmp and standard ECG leads for ionic concentration reconstruction.

Abstract:

In Europe, the prevalence of chronic kidney disease lay at approximately 18.38% in 2016. A common treatment for patients in the end stage of this disease is haemodialysis. However, patients undergoing this therapy suffer from an increased risk of cardiac death. A hypothesis is that the cause is an inbalanced electrolyte concentration. To study the underlying mechanisms of this phenomenon and fight the consequences, a continous non-invasive monitoring technique is desired. In this work, we investigated the possibility to reconstruct the extracellular concentrations of potassium and calcium from ECG signals. Therefore, we extracted 71 ECGs using the simulation results of a modified Himeno et al. ventricular cell model comprising variations of the extracellular ionic concentrations of potassium and calcium. The changes dependent on the different extracellular ionic concentrations were captured with five ECG features. These were used to train an artificial neural network for regression. The study was performed both for noise-free and noisy data. The estimation error for the reconstruction of the potassium concentrations was -0.01±0.14 mmol/l (mean±standard deviation) in the noise- free case, -0.03±0.46mmol/l in the noisy case (30dB SNR). For calcium, the result was 0.01±0.11mmol/l in the noise- free case, 0.02±0.17mmol/l in the noisy case. For both ion types, the result was improved by augmenting the dataset. We therefore conclude that with the calculated features, we are able to reconstruct the extracellular ionic concentrations for both potassium and calcium with an acceptable precision. When analysing noisy signals, the accuracy of the estimation method is still sufficient but can be further improved by an augmentation of the dataset.

Abstract:

Quantifying the atrial conduction velocity (CV) reveals important information for targeting critical arrhythmia sites that initiate and sustain abnormal electrical pathways, e.g. during atrial flutter. The knowledge about the local CV distribution on the atrial surface thus enhances clinical catheter ablation procedures by localizing pathological propagation paths to be eliminated during the intervention. Several algorithms have been proposed for estimating the CV. All of them are solely based on the local activation times calculated from electroanatomical mapping data. They deliver false values for the CV if applied to regions near scars or wave collisions. We propose an extension to all approaches by including a distinct preprocessing step. Thereby, we first identify scars and wave front collisions and provide this information for the CV estimation algorithm. In addition, we provide reliable CV values even in the presence of noise. We compared the performance of the Triangulation, the Polynomial Fit and the Radial Basis Functions approach with and without the inclusion of the aforementioned preprocessing step. The evaluation was based on different activation patterns simulated on a 2D synthetic triangular mesh with different levels of noise added. The results of this study demonstrate that the accuracy of the estimated CV does improve when knowledge about the depolarization pattern is included. Over all investigated test cases, the reduction of the mean velocity error quantified to at least 25 mm/s for the Radial Basis Functions, 14 mm/s for the Polynomial Fit and 14 mm/s for the Triangulation approach compared to their respective implementations without the preprocessing step. Given the present results, this novel approach can contribute to a more accurate and reliable CV estimation in a clinical setting and thus improve the success of radio-frequency ablation to treat cardiac arrhythmias.