Automatic adaptive parameterization in local phase feature-based bone segmentation in ultrasound.

Area of Science:

• Medical imaging
• Signal processing
• Biomedical engineering

Background:

• Log-Gabor filters provide accurate bone surface localization in 3D ultrasound.
• Current methods use fixed, empirically chosen filter parameters, limiting accuracy.
• Parameter selection critically impacts the quality of extracted local phase features.

Purpose of the Study:

• To develop a novel method for contextual parameter selection of Log-Gabor filters.
• To autonomously adapt filter parameters to specific image content for optimized feature extraction.
• To improve the accuracy of bone surface localization in 3D ultrasound imaging.

Main Methods:

• Developed a contextual parameter selection technique for Log-Gabor filters.
• Integrated principal curvature from the Hessian matrix and directional filter banks.
• Utilized a phase scale-space framework for parameter optimization.
• Automated selection of scale, bandwidth, and orientation parameters.

Main Results:

• Achieved a 35% improvement in bone surface localization accuracy in vitro.
• Demonstrated similar accuracy improvements in a pilot in vivo study on human subjects.
• Validated the effectiveness of the proposed contextual parameter selection method.

Conclusions:

• Contextual parameter selection significantly enhances Log-Gabor filter performance for bone localization.
• The novel method offers a more robust and accurate approach compared to empirical parameter setting.
• This technique holds promise for improving surgical navigation and medical diagnosis using 3D ultrasound.