Clinical validation of an edge detection algorithm for two-dimensional echocardiographic short-axis images

Area of Science:

• Cardiovascular Imaging
• Medical Image Analysis
• Echocardiography

Background:

• Accurate delineation of endocardial borders in echocardiography is crucial for quantitative analysis.
• Variability in manual border tracing can affect diagnostic accuracy and reproducibility.
• Automated edge detection algorithms offer potential to improve consistency and efficiency.

Purpose of the Study:

• To validate an edge detection algorithm for short-axis two-dimensional echocardiographic studies.
• To assess the algorithm's performance across multiple clinical laboratories and varying image quality.
• To compare computer-generated borders with expert manual tracings and evaluate variability.

Main Methods:

• Collected 30 short-axis echocardiographic studies from five clinical laboratories.
• Expert observers manually traced endocardial borders on separate days.
• A computer algorithm generated borders based on observer-defined regions of search.
• Quantified accuracy using correlation coefficients and variability using mean percent area difference.

Main Results:

• High correlation (0.985) between computer and manual borders in excellent quality studies.
• Significant reductions in interobserver variability (e.g., +/-9.8% to +/-5.3% in excellent studies).
• Significant reductions in intraobserver variability (e.g., +/-6.5% to +/-4.5% in excellent studies).
• High observer acceptance rates (93-97%) across image quality levels.
• Computer processing time averaged approximately 4 minutes per cardiac cycle.

Conclusions:

• The validated edge detection algorithm produces accurate and reliable endocardial borders.
• The algorithm significantly decreases both interobserver and intraobserver variability in echocardiographic border delineation.
• The computer-generated borders are acceptable to expert observers, demonstrating clinical utility.