Year: 2025
Type: dissertation
Abstract: Heart and vascular diseases are nowadays the leading cause of death globally and their identification and treatment at an early stage are important for preventing deaths. In this thesis the research is about the accuracy of automatically classifying different cardiac arrhythmias from an electrocardiogram (ECG), which records the hearts electrical activity through leads that attach to the skin. Automatic classification of cardiac arrhythmias can save a doctor's time and bring testing to a wider section of people. This can lead to more people getting diagnosed and treatment at an early stage with low cost. The identification is divided into two parts: binary classification and five-class classification. The binary classification is done with two classes: Normal and abnormal. The five-class classification is done with 5 classes: Normal, Myocardial infraction, ST/T change, Conduction disturbance and Hypertrophy. Similar research is usually done with a relatively small number of patients and longer recordings from the patients, where the heartbeats are separated to get a lot of samples from each patient. There have been a few works on an older version of the same dataset used on this thesis, but the feature extraction was different or not done at all. The first step in this thesis was to preprocess the signal to get rid of noise. After this procedure feature extraction was done, where the extracted features were Mel-frequency cepstrum coefficients (MFCC) and power spectrum. Feature extraction can bring meaningful features out of the signal, which help in the classification. In binary classification both feature extraction methods were tested to see which one got better results. MFCC’s got better results with Random Forest classifier. In the five-class classification MFCC’s were used, because they got better results in the binary classification. The five-class classification was done with a convolutional neural network (CNN) with Long short-term memory (LSTM). The problem was to classify different arrhythmias from an ECG-signal. Challenges were the different quality of signals and defining the models to classify the arrhythmias. The signals were preprocessed to get better quality and different models were tuned through trial and error.
Author: Ylinen, Olli
Institution: Tampere University
Language: English
Topic: ECG Monitoring and Analysis
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7132881095