An unsupervised image segmentation algorithm for coronary angiography

Area of Science:

• Medical image analysis
• Computer vision
• Biomedical engineering

Background:

• Computer visual systems offer advantages for automated medical image analysis.
• Precise image segmentation is crucial for accurate medical image processing.
• Coronary angiographs present segmentation challenges due to complex backgrounds and blurry edges.

Purpose of the Study:

• To develop an effective unsupervised learning algorithm for segmenting blood vessels in coronary angiographs.
• To establish coronary arteries and estimate vessel diameter using novel algorithms.

Main Methods:

• Proposed an unsupervised learning algorithm based on regional parameter expansion (RPE), derived from the flood fill algorithm.
• Introduced an optimal cover tree (OCT) algorithm for coronary artery establishment and vessel diameter estimation.
• Utilized region growing and spanning trees to record connection lengths for vessel path establishment.

Main Results:

• The RPE algorithm effectively segments blood vessel areas in coronary angiographs, even with complex backgrounds and uneven lighting.
• The OCT algorithm successfully establishes coronary artery paths and estimates vessel diameter.
• The proposed methods enhance the precision of medical image segmentation and analysis.

Conclusions:

• Unsupervised learning algorithms, specifically RPE and OCT, offer robust solutions for coronary angiograph analysis.
• These methods improve the accuracy of blood vessel segmentation and diameter measurement, crucial for cardiovascular diagnostics.
• The study demonstrates the potential of advanced computer vision techniques in enhancing medical imaging interpretation.