CardioDiverse: A Balanced ECG Dataset for Improved Association of Heart Diseases with Relevant Leads.

This study introduces CardioDiverse, a new dataset that has been constructed to address the scarcity of balanced data in existing electrocardiogram (ECG) datasets. CardioDiverse is a combination of multiple datasets integrated to provide rich information to support the association between the different ECG leads and CVD subclasses. The dataset is available to the public through GitHub. Our findings illustrate that particular leads provide clear manifestation to certain CVD subclasses. For instance, leads II, AVF, V3, V4, V5, and V6 were predominantly associated with ST-T wave abnormality, whereas leads II, III, AVF, V1, V5, and V6 were more affiliated with hypertrophy. Furthermore, by extending the analysis to include specific subclasses within the PTB-XL dataset, our study shows promising results in categorizing these subclasses using single-lead ECGs. CardioDiverse is an invaluable resource for researchers, as it provides a more balanced dataset and includes essential data analysis tools within the repository. The data and insights obtained through this study pave the way for future research to investigate the use of these leads for the early detection and diagnosis of cardiovascular diseases.