ePrints.FRI - University of Ljubljana, Faculty of Computer and Information Science

Automatic razlikovanje of pathologic and non-pathologic changes in ECG signals

Jana Faganeli Pucer (2013) Automatic razlikovanje of pathologic and non-pathologic changes in ECG signals. PhD thesis.

Download (4005Kb)


    Heart disease is the leading cause of death in the developed world. ECG signal recording and analysis is the easiest is the prime diagnostic procedure for early diagnosis of heart conditions. It is non-invasive, inexpensive and accessible when compared to other clinical procedures used in the diagnosis of heart disease. In clinical practice, short ECG recordings are usually used, which are recorded in a controlled environment, but long (24-hour) ECG recording (AECG) are also gaining popularity. AECG recordings are used mostly in the diagnosis of different arrhythmias, sometimes even in the diagnosis oh heart ischemia. Our work deals with the automatic detection of pathologic events in long ECG recordings. The first heart pathology we research is heart ischemia. It manifests as ST segment deviation in ECG signals. In the past years a large number of research papers have been published, dealing with the detection of transient ST segment episodes in AECG. The described ST segment episode detectors fail to differentiate between transient ischemic and transient non-ischemic heart rate related episodes. In our work we describe two methods to differentiate transient ST segment episodes. The first method uses a set of features; features that describe heart rate changes, ST segment deviation and morphology changes and QRS complex morphology changes. Using a set of features and different classifiers we show that automatic classification of the two types of episodes can be successful. Following the example of the methods used in exercise ECG (EECG) we define a ST(HR) diagram for AECG. The diagram is used to calculate a subset of features that help us differentiate between ischemic and heart rate related episodes. Similarly as in EECG we observe the diagram in two parts; from the beginning to the extreme of the episode and from the extreme to the end of the episode. We define two overall slopes, two maximal slope and the angle at the extrema of the episode. The slopes are good features (ranked with feature evaluation techniques) while the angle at the extrema is not a good feature. We also observe the heart rate at the beginning, extrema and end of the episodes. The performance of the classification with this set of features is worse than the classification with the set of features described in the last paragraph. The advantage of the classification with the ST(HR) diagram is that it uses a smaller number of features, the features are more comprehensive and easier to calculate. The second anomaly we study in our work is the detection of arrhythmic beats. Here we first develop a QRS detector based on the discrete Morse theory and then an arrhythmia detector. The QRS detector is based on cancelling neighbouring minima and maxima as in discrete Morse theory. With the addition of knowledge from ECG signal theory the performance of our QRS detector is very similar to the best performances published in the literature. Using a similar method as in QRS detection, we develop a procedure that simultaneously finds QRS complexes and up to four of the most significant waves (typical for ECG signals; P, QRS, T and U) between consecutive R waves. With the help of a newly developed algorithm, that evaluates the similarity between two trajectories, we assessed the similarity between consecutive heart beats. Then we classify pairs of heart beats as normal of arrhythmic with the help of two features; the similarity between the pair of beats and the difference between their RR intervals. The performance of the arrhythmia detector is high, especially on records containing a large number of very pathologic heart beats (e.g. PVC).

    Item Type: Thesis (PhD thesis)
    Keywords: electrocardiogram, ischemia, arrhythmia, discrete Morse theory, distance between trajectories, machine learning
    Number of Pages: 117
    Language of Content: Slovenian
    Mentor / Comentors:
    Name and SurnameIDFunction
    doc. dr. Matjaž Kukar267UNSPECIFIED
    Link to COBISS: http://www.cobiss.si/scripts/cobiss?command=search&base=50070&select=(ID=00010431060)
    Institution: University of Ljubljana
    Department: Faculty of Computer and Information Science
    Item ID: 2328
    Date Deposited: 09 Jan 2014 14:45
    Last Modified: 26 Feb 2014 09:21
    URI: http://eprints.fri.uni-lj.si/id/eprint/2328

    Actions (login required)

    View Item