Endafd: an End-to-End System for Ecg Based Atrial Fibrillation Detection

Atrial fibrillation is one of the most common arrhythmias with significant clinical implications. When it comes to detecting and categorising different types of heart diseases, machine learning and deep learning techniques have shown great results. There is a tendency to use convolution-based neural networks, as they have been specifically designed to categorise and recognise images. Image classification entails extracting image attributes in order to identify certain patterns in the dataset. The use of a basic ANN for this classifications would be computationally costly due to the large number of trainable parameters. Traditional convolution-based detection approaches, on the other hand, occasionally provided skewed findings since they had to integrate all 12 ecg leads to acquire the result. They will occasionally overlook the findings of a single lead abnormality if no other leads are shown to be defective. However, in the instance of a transformer, they take a single lead value at a time and check to see if there is any irregularity within that lead. Although the training may not always be successful owing to the small number of input samples, the results are typically rigged in favor of the negative since the number of negative samples is considerably larger than the positive ones We need a way to generate a larger number of balanced samples via the model, and then train the model to evaluate the original input. This is where the model comes in. A unique end-to-end model is suggested, consisting of two vision transformers connected by a generative adversarial network.