Title: Comparison of manually and automatically derived ECG-based left ventricular hypertrophy parameters and their predictive value in survival analysis
Other Titles: Verifikation von Algorithmen zur automatischen Bestimmung von EKG-basierten Parametern zur Detektion von linksventrikulärer Hypertrophie und deren klinische Aussagekraft in Survival-Analysen
Language: English
Authors: Contawe, Corina 
Qualification level: Diploma
Advisor: Kaniusas, Eugenijus  
Assisting Advisor: Mayer, Christopher 
Issue Date: 2017
Number of Pages: 61
Qualification level: Diploma
The presence of left ventricular hypertrophy (LVH) is linked to high risk of cardiovascular mortality, especially in dialysis patients. With the knowledge that LVH is reversible, an accurate method of detection in this cohort would be invaluable. Electrocardiography (ECG) is an universal, inexpensive procedure done at most healthcare settings. It is of great interest to further develop and improve ECG-based LVH parameters for assessing LVH incidence. The standard procedure for determining the parameters is to derive them from manual measurements of the electrocardiogram. Manual measurements are subject to human error and time consuming, thus it would advantageous to develop automatic methods of computing ECG-based LVH parameters. The aim of this thesis is to develop an automatic method of calculating the parameters and compare the results with manual measurements done by an experienced physician. Furthermore, the results, in combination with follow-up data, will be assessed for their predictive value in survival analysis. Four parameters were chosen to be automatically compared, the Sokolow-Lyon Voltage (SLV), the Cornell Voltage (CV), the Cornell Voltage Product (CVP), and the Novacode estimate. The AIT ECGsolver and QRS detectors taken from PhysioNet were used for detection of ECG waveforms and features. An algorithm was developed to compute the criteria by automatically evaluating amplitudes and intervals on a subset of 335 patients of the ISAR study. Recommended thresholds were applied to convert the calculated variables into dichotomized measurements. Both the criteria results and the threshold outcomes were compared with the manually derived matched beat measurements annotated by the physician. The voltage dependent criteria, SLV, CV, and the female Novacode estimate, demonstrated very good agreement between the two methods, both in measurements and LVH outcome. McNemar's test for these criteria showed no statistically significant difference between the measurement methods based on the dichotomized outcomes. Consistent overestimation of the QRS interval by automatic methods in comparison to the manual method proved to be problematic. The CVP and male Novacode are both dependent on QRS duration and demonstrated less agreement between the method measurements. McNemar¿s test showed for the male Novacode that there is a statistically significant difference between the methods, but not for the CVP. Interestingly, the survival analysis showed that CV, CVP, and the automatically derived Novacode were statistically significant risk predictors for all-cause mortality. This is in contrast to published studies which report no significant LVH criteria for all-cause mortality. Automatic methods of parameter calculation proved to be useful, but with limitations. Improvements in QRS interval detection would greatly refine the method. Future opportunities to expand the work done in this thesis would be a comparison of the binary results with accurate left ventricular mass (LVM) measurements, a repeat of the work during known dialysis treatment times (dialysis versus non-dialysis), and a recomparison using a standard method of manual QRS interval measurement.
Keywords: Automatische Auswertung und Verifikation; EKG; Linksventrikuläre Hypertrophie
Automatic detection and verification; ECG; Left Ventricular Hypertrophy
URI: https://resolver.obvsg.at/urn:nbn:at:at-ubtuw:1-99285
Library ID: AC13725812
Organisation: E354 - Institute of Electrodynamics, Microwave and Circuit Engineering 
Publication Type: Thesis
Appears in Collections:Thesis

Files in this item:

Page view(s)

checked on Jul 24, 2021


checked on Jul 24, 2021

Google ScholarTM


Items in reposiTUm are protected by copyright, with all rights reserved, unless otherwise indicated.