Machine learning approaches for a 20 AFl classification using BSPM and RQA images
A non-invasive discrimination of different atrial flutter (AFl) mechanisms could lead doctors during ablation procedure, drastically reducing the operation time and the use of invasive cardiac mapping systems.
20 AFl mechanisms have been simulated generating a huge dataset of BSPMs and 12-lead ECGs. On the 12-lead ECGs recurrence plots (RPs) and distance plots (DPs) have been extracted using recurrrence quantification analysis approaches.
Purphose of this thesis is to implement different machine learning approaches to classify the 20 AFl types using the BSPM videos, the RPs, and the DPs. The approches that are going to be implemente are: an atlas approach; k-nearest neighbour approach; convolutional neural network approach.