CardioLabelNet: An uncertainty estimation using fuzzy for abnormalities detection in ECG.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Electrocardiography (ECG) data, often stored as images, presents challenges for automated abnormality detection.
• Existing methods struggle with identifying multiple cardiac abnormalities simultaneously.

Purpose of the Study:

• To propose CardioLabelNet, a novel two-stage model for automated ECG image abnormality detection.
• To address limitations in current automated ECG analysis techniques.

Main Methods:

• CardioLabelNet employs a two-stage process: fuzzy membership for uncertainty computation and classification for abnormality detection.
• Uncertainty estimation involves global and local entropy calculations using fuzzy membership functions.
• A stacked architecture model is integrated for classifying ECG signal images.

Main Results:

• The proposed CardioLabelNet model demonstrates superior performance compared to existing methodologies.
• Performance was evaluated using segmentation accuracy, Dice similarity coefficient, partition entropy, and classification metrics (accuracy, sensitivity, specificity, AUC).

Conclusions:

• CardioLabelNet effectively detects normal and abnormal patterns in ECG images.
• The model offers an improved approach for automated analysis of cardiac abnormalities in ECG data.