Ultrasound Image Despeckling Using Stochastic Distance-Based BM3D

Area of Science:

• Medical Imaging
• Signal Processing
• Statistical Modeling

Background:

• Ultrasound imaging is crucial in medical diagnostics, but speckle noise degrades image interpretability.
• Patch-based methods like Block-Matching Collaborative Filtering (BM3D) are advanced techniques for ultrasound despeckling.
• Existing methods often rely on Euclidean distance for patch comparison, which may not be optimal for all noise types.

Purpose of the Study:

• To develop and evaluate new stochastic distance measures for the Fisher-Tippett distribution.
• To integrate these novel distances into a modified BM3D algorithm for enhanced ultrasound image despeckling.
• To improve the filtering of noise in log-compressed ultrasound images.

Main Methods:

• Derivation of new stochastic distances for the Fisher-Tippett distribution using established statistical divergences.
• Implementation of a modified Block-Matching Collaborative Filtering (BM3D) algorithm incorporating the novel patch distance measures.
• Application and evaluation of the algorithm on simulated, synthetic, and real ultrasound images.

Main Results:

• The proposed stochastic distances effectively serve as patch distance measures within the modified BM3D framework.
• The developed algorithm demonstrates superior performance in filtering speckle noise compared to existing state-of-the-art methods.
• Significant improvements in the interpretability of processed ultrasound images were observed across various datasets.

Conclusions:

• The novel stochastic distances derived from statistical divergences offer a powerful alternative to Euclidean distance for ultrasound despeckling.
• The modified BM3D algorithm incorporating these distances achieves state-of-the-art results in reducing speckle noise.
• This approach holds significant potential for improving the quality and diagnostic value of ultrasound medical imaging.