Exploring the Power of Ppg Matrix for Atrial Fibrillation Detection with Integrated Explainability

Abstract

This paper presents a novel approach for the classification of \ac{af} using \ac{ppg} signals treated as images. The proposed method builds upon previous research that has demonstrated promising results by converting \ac{ppg} signals into image representations and \ac{ecg} signals into heatmaps of matrices.The main contribution of this research lies in two aspects. Firstly, it introduces a novel methodology that treats \ac{ppg} signals as images, allowing for improved classification accuracy. Classification of signals between \ac{af} and \ac{nsr} is performed over the MIMIC PERform Dataset achieving achieving 100\% accuracy. Secondly, this paper addresses the need for interpretability in medical applications by incorporating \ac{xai}. Thus, the classification process is transparent, enabling a comprehensive understanding of how the \ac{cnn} interprets \ac{af} for accurate classification.Overall, this research provides a valuable contribution to the field of \ac{af} classification using \ac{ppg} signals. Combining image-based preprocessing techniques and explainable architectures offers improved accuracy and interpretability, paving the way for enhanced diagnostic capabilities in clinical settings.

Note:
Funding Declaration: This work was supported by the Spanish Ministry of Science, Innovation and Universities grants TED2021-131681B-I00 (CIOMET); and by the Comunidad de Madrid (Spain) under the project PUCFA (PUCFA-CM-UC3M). The authors gratefully acknowledge the computer resources at Artemisa, funded by the European Union ERDF and Comunitat Valenciana, and the technical support provided by the Instituto de Fisica Corpuscular, IFIC (CSIC-UV).

Conflicts of Interest: None

Keywords: Photoplethysmogram, Convolutional Neural Networks, Atrial Fibrillation, Explainable Artificial Intelligence