Parametrical modelling for texture characterization-A novel approach applied to ultrasound thyroid segmentation

Area of Science:

• Medical imaging
• Signal processing
• Biomedical engineering

Background:

• Texture analysis is crucial for ultrasound (US) image interpretation, aiding structure segmentation and tissue classification.
• Existing methods often require substantial data and are sensitive to noise, limiting localized analysis.
• Parametric modeling offers a potential avenue for more robust and efficient texture feature extraction.

Purpose of the Study:

• To present a novel parametric modeling approach for ultrasound image texture feature extraction.
• To improve robustness to low signal-to-noise ratio and enable more localized image analysis.
• To demonstrate the effectiveness of the proposed method for thyroid segmentation in ultrasound images.

Main Methods:

• Parametric modeling of ultrasound image signals as data from a dynamical process.
• Extraction of texture features based on the predictive characteristics of the parametric model.
• Application of k-means clustering for differentiating thyroid and non-thyroid image patches.
• Validation using two distinct ultrasound datasets for thyroid segmentation.

Main Results:

• The proposed method extracts texture features with high predictive accuracy using less data than conventional techniques.
• Features derived from energy ratios across modeled texture frequency bands effectively distinguish thyroid tissue.
• Thyroid segmentation using the novel approach achieved Dice coefficients exceeding 85% on two datasets.
• The method demonstrated enhanced robustness to low signal-to-noise ratios and improved localization.

Conclusions:

• The novel parametric modeling approach provides an effective and robust method for ultrasound image texture analysis.
• This technique significantly improves the accuracy and localization of thyroid segmentation in ultrasound images.
• The findings suggest broader applicability of this method for texture-based analysis in medical imaging.