Simultaneous left atrium anatomy and scar segmentations via deep learning in multiview information with attention

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Three-dimensional late gadolinium-enhanced (LGE) cardiac magnetic resonance (CMR) imaging is crucial for assessing left atrial (LA) scar in atrial fibrillation (AF) patients.
• Accurate segmentation of LA anatomy and scar tissue is essential for patient stratification and guiding ablation therapy.
• Current methods often require separate acquisitions and registration, introducing potential errors and inefficiencies.

Purpose of the Study:

• To develop and validate a joint segmentation method for simultaneously segmenting LA anatomy and scar tissue from a single 3D LGE CMR acquisition.
• To eliminate the need for additional imaging sequences and registration steps.
• To improve the accuracy and efficiency of scar assessment in AF patients.

Main Methods:

• A novel multiview two-task (MVTT) recursive attention model was developed for joint segmentation.
• The model operates directly on 3D LGE CMR images, segmenting both LA anatomy and scar tissue.
• Performance was evaluated using Dice scores and computational efficiency was measured.

Main Results:

• The MVTT model achieved high accuracy, with mean Dice scores of 93% for LA anatomy and 87% for scar segmentation.
• Simultaneous segmentation of LA anatomy and scar was highly efficient, taking approximately 0.27 seconds per dataset.
• The proposed method outperformed conventional unsupervised learning and other state-of-the-art deep learning techniques.

Conclusions:

• The MVTT recursive attention model offers an accurate and efficient solution for joint LA anatomy and scar segmentation from 3D LGE CMR.
• This approach facilitates the automatic generation of patient-specific anatomical models with scar segmentation for AF patients.
• The method has the potential to enhance clinical decision-making in AF management and ablation therapy.