Fully-automatic left ventricular segmentation from long-axis cardiac cine MR scans.

Area of Science:

• Medical Imaging
• Cardiovascular Imaging
• Image Analysis

Background:

• Large-scale cardiac magnetic resonance (CMR) studies require efficient image analysis.
• Manual segmentation of cardiac cine MRI scans is time-consuming and labor-intensive.
• Automated pipelines are crucial for processing population-based imaging datasets.

Purpose of the Study:

• To investigate the feasibility of a fully-automatic pipeline for simultaneous segmentation of left ventricular endocardium and epicardium.
• To evaluate the pipeline's performance on vertical and horizontal long-axis cardiac cine MRI scans.
• To assess the clinical utility of automated segmentation for deriving volumetric parameters and risk stratification.

Main Methods:

• A multi-atlas-based segmentation approach combined with spatio-temporal registration was employed.
• The pipeline processed orthogonal long-axis cardiac cine MRI scans.
• Performance was assessed by comparing automated segmentations and derived clinical parameters against manual references.

Main Results:

• High Dice similarity coefficients (DSC) were achieved for endocardial (0.85-0.93) and epicardial (0.88-0.95) segmentation across diastolic and systolic phases.
• Strong correlations (R=0.84-0.97) were observed between automated and manual volumetric parameters (ejection fraction, stroke volume, etc.).
• Automated ejection fraction accurately classified 80% of subjects into appropriate risk categories.

Conclusions:

• The proposed automatic pipeline is a viable and cost-effective solution for CMR image analysis.
• Automated segmentation significantly reduces the burden of manual annotation in population-based studies.
• The pipeline demonstrates high accuracy in segmenting cardiac structures and calculating key clinical parameters.