Development and verification of an ECG processing algorithm for fast post-hoc analysis of large datasets collected with VREACT

Abstract

With todays possibilities of easy and fast medical data collection the amount of gathered data increases at a fast pace. Hence, also the need for quick and reliable analysis algorithms to get an overview and first insights of a dataset is very high. The goal of this diploma thesis was to develop signal processing algorithms with the objective of analyzing the heart rate variability (HRV) of already recorded electrocardiograms (ECGs).The patient data was provided by the Department of Anaesthesia of the Medical University of Vienna and was recorded from patients under general anaesthesia during different surgical procedures. In order to achieve this goal the ECG data first had to be preprocessed and signal distortions and artefacts needed to be eliminated. Then an automated QRS complex detection algorithm was used to detect the exact position of every heart beat and subsequently the beat to beat intervals were calculated. By applying some further signal quality improvement algorithms (artefact detection based on the beat to beat intervals and detection of ectopic heart beats) a mean heart beat detection positive predictive value (PPV) of 98.89% and sensitivity of 97.38% were achieved as compared with the original QRS complex detection algorithm, which achieved a mean PPV of 97.92% and sensitivity of 97.93%. From the beat to beat intervals subsequently the following HRV parameters were calculated in the time domain: the standard deviation of normal to normal heartbeat intervals (SDNN), the root mean square of successive heartbeat interval differences and the percentage of successive normal to normal heartbeat intervals differing by more than 50ms. These parameters were calculated for the whole dataset and for standard 5 minute data-segments and were then statistically analyzed. A correlation between the SDNN and the following metadata was found: American Society of Anesthesiologists physical status classification, diabetes mellitus and inflammation.Furthermore the time under general anaesthesia seems to influence the HRV as most of the patients show a drop in the SDNN by aproximately 50% about 10min after the ECG recording is started. Overall the quality of the beat detection and thus the HRV data produced by the algorithms developed in this work proved to be more than sufficient for the investigation of the present dataset. However the dataset lacks some metadata and details like the mode of the artificial ventilation and drugs given to the patients. Thereby, if there were more sufficient information available, there might be even more statistically/clinically relevant conclusions gained from the HRV investigation. The next step will be a handover of the developed tools to healthcare professionals in order to investigate other ECG datasets and gather feedback from the end users.