Ultrasound k-nearest neighbor entropy imaging: Theory, algorithm, and applications

Area of Science:

• Medical Imaging
• Biophysics
• Quantitative Ultrasound

Background:

• Ultrasound backscattering analysis is crucial for tissue characterization.
• Shannon entropy imaging using probability distribution histograms (PDHs) is a common method, but sensitive to bin number.
• Limitations in PDH stability necessitate advanced entropy estimation techniques.

Purpose of the Study:

• Introduce k-nearest neighbor (KNN) algorithm for ultrasound entropy estimation.
• Propose ultrasound KNN entropy imaging for enhanced tissue characterization.
• Develop cumulative relative entropy (CRE) imaging for monitoring thermal lesions.

Main Methods:

• Implemented KNN algorithm to estimate entropy, leveraging Euclidean distance.
• Developed cumulative relative entropy (CRE) imaging for time-series radiofrequency signal analysis.
• Validated KNN entropy imaging against PDH and Nakagami-m imaging using simulations, clinical breast lesion data, and ex vivo porcine liver MWA experiments.

Main Results:

• KNN entropy estimation demonstrated reduced sensitivity to tuning parameters compared to PDH.
• KNN entropy imaging showed higher sensitivity to scatterer density changes and improved visualization.
• KNN-based Shannon entropy (KSE) imaging achieved higher accuracy in classifying breast tumors.
• KNN-based CRE imaging provided superior lesion-to-normal contrast during microwave ablation monitoring.

Conclusions:

• Ultrasound KNN entropy imaging is a robust quantitative ultrasound technique.
• KNN-based methods offer improved performance over conventional PDH and Nakagami-m imaging.
• This approach holds potential for accurate tissue characterization and thermal lesion monitoring.