A Domain Data-Driven Metric to Improve Traditional Visualization and Interaction with ECG Large Databases

The exponential growth of electrocardiogram (ECG) data presents significant challenges, particularly in efficiently handling large volumes of raw ECG data, leading to manipulation, interpretation, and analysis difficulties. Consequently, there is an urgent need to develop novel strategies that simplify and enhance the analysis of these extensive datasets. Recent advances in computer-assisted and visualization technologies offer promising solutions to these issues by improving ECG analysis. These advances also constitute a unique opportunity to design new tele-education, teleconsultation, and recommendation systems scenarios. This study introduces a novel data-driven approach for interacting with ECG collections by constructing a latent space with a Convolutional Neural Network (CNN) and grouping the ECG collection using the labels of the database. This approach introduces three key contributions: a data-driven metric for case retrieval, a customized grouping of data using the associated diagnosis of each case, and an adapted interface to dynamically interact with relevant data. Results demonstrate the approach's effectiveness at representing the latent space as a simple Euclidean distance to the nearest neighbors. The user-friendly interface enables uploading new signals, diagnostic predictions, and comparisons with existing ECGs, facilitating in-depth interaction and detailed analysis. This strategy not only enhances the accuracy and efficiency of the ECG analysis but also has the potential to transform the way healthcare professionals manage large volumes of ECG data in dynamic clinical settings.