A Semi-Automatic Method To Segment The Left Atrium in MR Volumes With Varying Slice Numbers

Area of Science:

• Medical Imaging
• Cardiology
• Artificial Intelligence

Background:

• Atrial fibrillation (AF) is a common arrhythmia with significant health impacts.
• Cardiac MRI is crucial for AF management, but segmenting the left atrium (LA) is challenging.
• Existing segmentation tools struggle with atrial cine MR image characteristics and data limitations.

Purpose of the Study:

• To develop a semi-automatic method for accurate left atrium segmentation in cardiac MRI.
• To address challenges posed by image quality and data scarcity in LA segmentation.
• To improve the efficiency and accuracy of LA segmentation for AF patient management.

Main Methods:

• A semi-automatic approach requiring a single user click per volume to initiate segmentation.
• Utilizing a deep learning network with a chamber location map for LA identification.
• Implementing a tracking method to refine segmentation across image slices and remove artifacts.

Main Results:

• The proposed method significantly outperforms U-Net in accuracy, achieving superior Hausdorff distance and Dice scores.
• Demonstrated improved segmentation performance on an in-house MRI dataset.
• Requires limited manual interaction, enhancing usability for clinicians.

Conclusions:

• The developed semi-automatic method offers a robust solution for LA segmentation in cardiac MRI.
• This technique shows promise for improving the clinical workflow in managing atrial fibrillation.
• The method's accuracy and efficiency represent a significant advancement over current segmentation tools.