Automated Peak Prominence-Based Iterative Dijkstra's Algorithm for Segmentation of B-Mode Echocardiograms.

Area of Science:

• Medical Imaging
• Cardiovascular Imaging
• Computational Anatomy

Background:

• Accurate segmentation of the left ventricle (LV) is crucial for diagnosing cardiac conditions using echocardiograms (echo).
• Existing automated segmentation methods often require extensive training data or exhibit variability.
• Manual segmentation is time-consuming and subject to inter-observer differences.

Purpose of the Study:

• To develop and validate a user-initialized, automated LV segmentation method for echocardiograms.
• To assess the accuracy, efficiency, and inter-operator variability reduction of the proposed method.

Main Methods:

• Introduced the Prominence Iterative Dijkstra's (ProID) algorithm, utilizing an iterative Dijkstra's algorithm with strategic node selection and a novel cost matrix.
• Initialized ProID with three user clicks per time-series scan.
• Validated ProID on artificial echocardiograms and the CAMUS dataset, followed by analysis of a clinical cohort of 66 pediatric patients.

Main Results:

• ProID demonstrated accurate LV contours and volume estimations compared to ground-truth across various artificial echo systems.
• Achieved Dice Similarity Scores (DSS) comparable to other automated methods on the CAMUS dataset.
• In a clinical cohort, ProID achieved an average DSS of 0.93 against manual segmentation and reduced inter-operator variability between expert readers (manual DSS 0.93, ProID-assisted DSS 0.95).

Conclusions:

• ProID provides accurate left ventricle segmentation across diverse age groups, disease states, and echocardiogram platforms.
• The method offers low computational cost and eliminates the need for training data.
• ProID shows potential to improve consistency and reduce variability in clinical echocardiogram analysis.