Key result
The BAt-RSFR deep learning model achieved high segmentation accuracy for bi-atrium cavities and walls in LGE-MRI, with Dice Similarity Coefficients of 0.9975, 0.8328, and 0.8727.
Why the study?
AF increases atrial pressure leading to cavity enlargement and wall stretching, and accurate segmentation of atrial cavities and walls on LGE-MRI is needed to improve clinical assessment.
The proposed BAt-RSFR deep learning model demonstrates high accuracy for bi-atrium segmentation in LGE-MRI, potentially improving clinical assessment and anatomical measurement in patients with atrial fibrillation.
DL model may enable atrial segmentation on LGE-MRI in AFib; leaves open prospective validation before clinical use.
Atrial Fibrillation (AFib) is the most prevalent form of Cardiac Arrhythmia, characterized by unsynchronized electrical activity in the heart chambers. In AFib, the ineffective atrial contraction increases atrial pressure, leading the atrial cavity to enlarge and the atrial walls to stretch. The purpose of this study is to design a deep learning based model to segment the atrial cavities and walls using Late Gadolinium Enhanced Magnetic Resonance Imaging (LGE‐MRI). This non‐invasive imaging technique improves the clinical assessment of AFib patients. This proposed study introduces BAt‐RSFR (Bi Atrium‐ Residual Spatial Feature Refinement), a 3D deep learning model for bi‐atrium segmentation. This model integrates Residual Convolutional and Spatial Feature refinement mechanisms to enhance relevant spatial features. The attention mechanism enhances feature refinement by adjusting spatial weights, helping the model to highlight relevant regions. The proposed model is trained on the Multiclass bi‐atrium LGE‐MRI dataset provided by MICCAI 2024, Multiclass bi‐atrium segmentation challenge. The model's performance is evaluated using three key metrics: Dice Similarity Coefficient (DSC) and Hausdorff distance (HD) and Intersection over Union (IoU) for the left and right atrium cavity, atrium walls and achieved DSC of 0.9975, 0.8328, 0.8727 and corresponding HD of 1.9048 mm, 3.3109 mm, 2.8562 mm and IoU of 0.9949, 0.7935, 0.7469 respectively. To further assess the clinical relevance, left and right atrial chamber volume are calculated using standard voxel based volumetric method from both ground truth and model predictions, confirming the model's ability to preserve clinically meaningful anatomical measurement. These metrics indicate that the proposed model demonstrates improved segmentation accuracy compared with existing models.
No takes yet. Share an insight, caveat, or question.
Abinaya et al. (2026) studied Atrial Fibrillation. BAt-RSFR (Bi Atrium- Residual Spatial Feature Refinement) 3D deep learning model vs. existing models was evaluated on Dice Similarity Coefficient (DSC), Hausdorff distance (HD), and Intersection over Union (IoU) for the left and right atrium cavity and atrium walls. The BAt-RSFR deep learning model achieved high segmentation accuracy for bi-atrium cavities and walls in LGE-MRI, with Dice Similarity Coefficients of 0.9975, 0.8328, and 0.8727.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: