Left Ventricle Segmentation in Cardiac MR Images Using Fully Convolutional Network.

Area of Science:

• Medical image analysis
• Cardiovascular imaging
• Artificial intelligence in healthcare

Background:

• Accurate segmentation of cardiac structures, particularly the left ventricle, is crucial for diagnosing heart abnormalities using cardiac magnetic resonance (MR) imaging.
• Challenges in left ventricle segmentation include intensity/shape similarity with adjacent organs, imprecise boundaries, and image noise.
• Clinical decision support systems rely on precise medical image analysis for effective diagnosis.

Purpose of the Study:

• To propose an automated method for segmenting the left ventricle in cardiac MR images.
• To overcome challenges associated with segmentation accuracy, including noise and anatomical similarities.
• To develop a robust algorithm for improved cardiac image analysis.

Main Methods:

• An automated region of interest extraction followed by input into a fully convolutional network (FCN).
• Training the FCN to handle datasets with a small proportion of left ventricle pixels.
• Implementing a post-processing phase involving thresholding and region selection based on roundness.

Main Results:

• The proposed automated method achieved a Dice score of 87.24% on the York heart image dataset.
• The method successfully segmented the left ventricle despite inherent image complexities.
• The combination of FCN and post-processing techniques proved effective.

Conclusions:

• The developed automated method offers a promising solution for left ventricle segmentation in cardiac MR images.
• The approach demonstrates potential for enhancing clinical decision support systems in cardiology.
• Further validation on diverse datasets is warranted to confirm generalizability.