Quantitative analysis of complex atrial intramural microstructure is a crucial step towards understanding the mechanism behind atrial fibrillation (AF) maintenance. Siamese network was adopted to extract features from computationally simulated multi-layer fibrosis structure. Through analysis of the features produced by the feature extractor, the difference between Non-sustained and Sustained simulations was comprehended intuitively and electrophysiologically. Complex conduction pathway marked by the feature extractor might be an indicator for AF radio-frequency ablation clinically.