Segmentation of left atrium using CT images and a deep learning model.

Area of Science:

• Cardiology
• Medical Imaging
• Computational Anatomy

Background:

• The left atrium (LA) is a complex cardiac structure vital for diagnosing conditions such as atrial fibrillation.
• Atrial fibrillation can lead to thrombogenesis within the left atrial appendage.
• Accurate segmentation of the LA is challenging due to anatomical complexity and patient-specific variations.

Purpose of the Study:

• To develop an unbiased computational pipeline for segmenting the left atrial cavity from CT images.
• To provide an automated solution for a demanding task in cardiac imaging analysis.

Main Methods:

• Development of an automated segmentation pipeline for the left atrium.
• Utilizing computed tomography (CT) imaging data.
• Evaluation using state-of-the-art segmentation metrics.

Main Results:

• The automated pipeline demonstrated effective segmentation of the left atrial cavity.
• Key performance metrics included a Dice score of 80%, Hausdorff distance of 11.78mm, average surface distance of 2.24mm, and Rand error index of 0.2.

Conclusions:

• The developed automated pipeline shows potential for accurate left atrium segmentation from CT images.
• This approach can assist in clinical diagnosis and patient management, particularly for conditions related to the left atrium.