AI-driven CNN analysis of ECGs classified four types of atrial cardiomyopathy with 97.9% accuracy, enabling early detection before atrial fibrillation onset.
Can a convolutional neural network accurately classify atrial cardiomyopathy types based on simulated ECG data?
An AI-driven CNN model can accurately classify simulated atrial cardiomyopathy types from ECG data, demonstrating potential for early detection of structural remodeling prior to atrial fibrillation onset.
Absolute Event Rate: 0% vs 0%
Abstract Background Atrial fibrillation (AF) is the most common cardiac arrhythmia, affecting over 2% of the global population and imposing a significant burden on healthcare systems due to its strong association with stroke, heart failure, and mortality. In recent years, artificial intelligence (AI) has been increasingly applied to ECGs for the automatic diagnosis of AF and prediction of AF-related outcomes. However, timely diagnosis and treatment remains a major challenge in AF management, leading to suboptimal patient outcomes. Clinical evidence suggest that early diagnosis and intervention can significantly improve the outcomes by enabling treatment before the onset of advanced cardiomyopathy. Aim This study aims to establish a proof of concept for the application of AI, specifically convolutional neural networks (CNNs), to predict atrial cardiomyopathy type based on ECG data, with the goal of facilitating early AF diagnosis, enabling timely intervention. Methods A dataset of 240 2D tissue models were generated and used with the Fenton-Karma atrial electrophysiology model to simulate atrial electrical conduction. Cardiomyopathy was introduced as a variable distribution of non-conductive fibrotic tissue with different diffusion coefficients, D, representing tissue conductivity. Four distinct tissue types were simulated: healthy (D = 0.1 mm²s-1), gap junctional remodelling (D = 0.01–0.09 mm²s-1), fibrotic remodelling with non-conductive tissue patches (D = 0 mm²s-1), and a combined remodelling condition. ECG P-waves were computed for each subject using the infinite volume conductor method. A CNN was trained to classify the ECGs into the four types. Results Simulations of 2D atrial tissue models with fibrotic patches resulted in disrupted electrical conduction, and hence distinct morphological irregularities in the ECG signals. Tissues with gap junctional remodelling were characterised by slow conduction, displayed as a progressive reduction in ECG amplitude and a smoother, more spread-out waveform. For classifying atrial tissue types from ECG data, the CNN model was evaluated using 10-fold cross-validation, achieving a high mean macro accuracy of 0.979 ± 0.032, a mean macro AUC of 0.9810 ± 0.0267 and a mean macro F1 score of 0.980 ± 0.0058. Conclusion This study demonstrates the feasibility of using AI-driven ECG analysis for the early detection of atrial cardiomyopathy, a key precursor to AF. The CNN model achieved high classification accuracy, highlighting its potential to identify subtle ECG abnormalities associated with structural remodelling before AF onset. This approach could facilitate both early diagnosis and intervention, ultimately improving outcomes for AF patients.AI-ECG prediction of atrial tissue type
Spota et al. (Sat,) reported a other. AI-driven CNN analysis of ECGs classified four types of atrial cardiomyopathy with 97.9% accuracy, enabling early detection before atrial fibrillation onset.