DECNet: Left Atrial Pulmonary Vein Class Imbalance Classification Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Pulmonary vein anatomical classification is critical for atrial fibrillation radiofrequency ablation surgery.
• Deep learning models struggle with imbalanced data and subtle anatomical variations, impacting classification accuracy.

Purpose of the Study:

• To address the challenge of unbalanced classification of left atrial pulmonary veins.
• To propose a novel deep learning network, DECNet, for accurate pulmonary vein morphology classification.

Main Methods:

• Developed DECNet, integrating multi-scale feature-enhanced attention and dual-feature extraction classifiers.
• Multi-scale attention enhances deep features using channel and spatial weights.
• Dual-feature classifier mitigates learning bias and overfitting from data imbalance.

Main Results:

• DECNet achieved average accuracies of 78.81% on clinical data and 83.44% on the DermaMNIST dataset.
• The proposed method demonstrated enhanced deep feature expression and accurate classification of left atrial pulmonary vein morphology.

Conclusions:

• DECNet effectively improves the classification accuracy of left atrial pulmonary vein anatomy.
• The model provides valuable support for preoperative assessment and clinical treatment planning in atrial fibrillation surgery.