Outlier Detection in ECG

Automatic abnormal ECG detection algorithms are crucial for treating heart health problems and saving lives. This study's objective is to verify whether using unsupervised learning methods and more specifically probabilistic and deep learning models such as autoencoders, variational autoencoders (VAE), diffusion Models, partial shift variational autoencoder (our implementation of half VAE half prediction model), normalizing flows and Gaussian mixture models to detect outliers in ECG data is possible. An outlier for our case is an abnormal ECG signal, a one belonging mostly to a sick person while the normal case is that of a normal healthy person. The results have shown that the models distinguish between normal and abnormal data to a specific degree, with the VAE achieving an area under the curve (AUC) of the receiver operating characteristic curve score of 0.85 on the publicly available PTB-XL dataset and 0.83 on the publicly available CPSC dataset. Moreover, the VAE achieved anAUC of 0.89, 0.80 and 0.81 when distinguishing between normal and conduction disturbance, myocardial infarction, and STIT Change respectively. This indicates that a VAE when optimized itself and fed with more proper data may be able to be used in medical applications.