LASSNet: A Four Steps Deep Neural Network for Left Atrial Segmentation and Scar Quantification.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiac Electrophysiology

Background:

• Accurate left atrium (LA) scar quantification is critical for successful atrial fibrillation ablation.
• Manual LA segmentation and scar quantification are time-consuming and prone to inter-observer variability.

Purpose of the Study:

• To develop and validate a deep neural network for automated LA cavity segmentation and LA scar quantification.
• To improve the accuracy and efficiency of scar assessment for guiding ablation strategies.

Main Methods:

• A multi-network sequential deep learning architecture was developed in two stages: LA cavity segmentation and LA scar segmentation.
• Each stage incorporated a region of interest network and a refined segmentation network.
• The model was trained and validated using over 200 late gadolinium-enhanced MRI images from the LAScarQS 2022 Challenge.

Main Results:

• The deep neural network successfully automated LA cavity segmentation and LA scar quantification.
• Performance analysis was conducted based on various parameters, including data triaging.
• The developed method demonstrated improved scar quantification performance compared to existing literature.

Conclusions:

• Automated LA scar quantification using deep learning offers a more accurate and efficient alternative to manual methods.
• This technology has the potential to significantly enhance the guidance of ablation strategies for atrial fibrillation.
• The validated deep neural network provides a robust tool for clinical applications in cardiac imaging.