[Paper Review] Simultaneous Left Atrium Anatomy and Scar Segmentations via Deep Learning in Multiview Information with Attention

Guang Yang, Jun Chen+14 more|arXiv (Cornell University)|Feb 2, 2020

Cardiac Valve Diseases and Treatments42 references4 citations

TL;DR

This paper proposes a multiview two-task recursive attention deep learning model that simultaneously segments left atrial anatomy and left atrial scar tissue from a single 3D late gadolinium-enhanced cardiac MRI scan. The method achieves high accuracy (mean Dice score of 93% for left atrium and 87% for scar) in under 0.27 seconds per scan, eliminating the need for separate bright-blood scans and registration steps.

ABSTRACT

Three-dimensional late gadolinium enhanced (LGE) cardiac MR (CMR) of left atrial scar in patients with atrial fibrillation (AF) has recently emerged as a promising technique to stratify patients, to guide ablation therapy and to predict treatment success. This requires a segmentation of the high intensity scar tissue and also a segmentation of the left atrium (LA) anatomy, the latter usually being derived from a separate bright-blood acquisition. Performing both segmentations automatically from a single 3D LGE CMR acquisition would eliminate the need for an additional acquisition and avoid subsequent registration issues. In this paper, we propose a joint segmentation method based on multiview two-task (MVTT) recursive attention model working directly on 3D LGE CMR images to segment the LA (and proximal pulmonary veins) and to delineate the scar on the same dataset. Using our MVTT recursive attention model, both the LA anatomy and scar can be segmented accurately (mean Dice score of 93% for the LA anatomy and 87% for the scar segmentations) and efficiently (~0.27 seconds to simultaneously segment the LA anatomy and scars directly from the 3D LGE CMR dataset with 60-68 2D slices). Compared to conventional unsupervised learning and other state-of-the-art deep learning based methods, the proposed MVTT model achieved excellent results, leading to an automatic generation of a patient-specific anatomical model combined with scar segmentation for patients in AF.

Motivation & Objective

• To eliminate the need for separate bright-blood MRI acquisitions in atrial fibrillation patients by enabling joint segmentation of left atrial anatomy and scar tissue from a single 3D late gadolinium-enhanced cardiac MRI (LGE-CMR) scan.
• To address the challenge of misalignment and registration errors that arise when combining separate anatomical and scar segmentation results from different MRI sequences.
• To develop a deep learning framework capable of jointly learning complex anatomical and pathological tissue patterns in a single inference pass.
• To improve the efficiency and accuracy of patient-specific modeling for atrial fibrillation ablation planning by automating both segmentation tasks simultaneously.

Proposed method

• The method employs a multiview two-task (MVTT) recursive attention network that processes 3D LGE-CMR images from multiple orthogonal views to enhance feature representation.
• The network uses a recursive attention mechanism to progressively refine feature maps, focusing on relevant anatomical and scar regions across the 3D volume.
• The model is trained end-to-end to perform two segmentation tasks simultaneously: left atrium and pulmonary vein segmentation, and left atrial scar segmentation.
• The architecture leverages shared encoder features with task-specific heads to enable joint optimization of both segmentation tasks.
• The method processes 60–68 axial 2D slices per 3D volume, enabling real-time inference with a runtime of approximately 0.27 seconds per scan.

Experimental results

Research questions

• RQ1 Can a single deep learning model accurately segment both left atrial anatomy and left atrial scar tissue from a single 3D LGE-CMR scan?
• RQ2 Does the use of multiview information and recursive attention improve segmentation accuracy compared to single-view or non-recursive approaches?
• RQ3 Can the joint segmentation of anatomy and scar reduce the need for additional bright-blood MRI acquisitions and associated registration errors?
• RQ4 How does the proposed MVTT model compare to state-of-the-art deep learning and unsupervised methods in terms of Dice score and inference speed?

Key findings

• The proposed MVTT model achieved a mean Dice score of 93% for left atrial anatomy segmentation, demonstrating high accuracy in capturing complex cardiac structures.
• The model achieved a mean Dice score of 87% for left atrial scar segmentation, indicating strong performance in identifying fibrotic tissue.
• The method processed each 3D LGE-CMR dataset in approximately 0.27 seconds, enabling real-time clinical application.
• The joint segmentation approach eliminated the need for separate bright-blood scans and avoided registration-related errors common in conventional workflows.