Robust unsupervised texture segmentation for motion analysis in ultrasound images

Area of Science:

• Medical imaging
• Image analysis
• Biomedical engineering

Background:

• Ultrasound (US) imaging is a cost-effective, portable, and non-ionizing diagnostic tool.
• Analyzing motion in US images requires robust segmentation and tracking of anatomical structures.
• Challenges in US imaging include low contrast and blurry boundaries, hindering accurate analysis.

Purpose of the Study:

• To present a robust descriptor inspired by fractal dimension for characterizing gray-level variations in ultrasound images.
• To develop a method for accurate segmentation and motion tracking in challenging ultrasound data.
• To improve the reliability of analyzing anatomical structures and tissue deformation in medical imaging.

Main Methods:

• A novel descriptor based on fractal dimension is introduced to characterize local image gray-level variations.
• The descriptor utilizes an adaptive grid pattern whose scale adjusts to local gray-level changes.
• Robust features are identified based on gray-level variations for consistent temporal tracking.

Main Results:

• The method achieved accurate segmentation of the left ventricle in simulated echocardiography (Dice coefficient: $$).
• It demonstrated robust tracking of diaphragm motion in healthy subjects (Mean Sum of Distances: $$ mm).
• The technique accurately tracked diaphragm motion in a scoliosis patient (Root Mean Square Error: $$ mm).

Conclusions:

• The developed method can segment structures based on texture in an unsupervised manner.
• It shows potential for analyzing tissue deformation and aiding in disease diagnosis.
• The principle is applicable to other medical imaging modalities like MRI and CT scans.