A Decision-Making System with Reject Option for Atrial Fibrillation Prediction Without ECG Signals

• Published in: Ingénierie et recherche biomédicale Vol. 43; no. 6; pp. 573 - 584
• Main Authors: Mroueh, M., Alshamaa, D., Mourad-Chehade, F., Abdallah, F.
• Format: Journal Article
• Language: English
• Published: Elsevier Masson 01.12.2022
• Subjects: Computer Science; Fusion of evidence; Belief functions theory; Decision making; Atrial fibrillation
• ISSN: 1959-0318
• DOI: 10.1016/j.irbm.2022.04.008

Objectives: This paper presents a new method for Atrial Fibrillation detection based on the belief functions theory. Materials and methods: The theoretical framework allows to handle missing and uncertain data, to aggregate evidence in an independent order of sources of information and to reject a decision in case of insufficient supporting evidence. The proposed method is evaluated on real signals acquired from Intensive Care Units available in the MIMIC-III database and compared to state-of-the-art technologies and methods. Results: The precision of the suggested method is 90.03%, which is 2% more than existing methods in the literature. Conclusion: While almost all existing methods rely on high frequency sampled ECG signals, mainly at 125 Hz, to achieve a good accuracy, our proposed approach achieves a comparable performance using low frequency sampled physiological signals at 0.016 Hz without the need for an ECG which allows for a significant reduction in energy consumption, in data size and in processing complexity.