Gaussian process regression for ultrasound scanline interpolation.

Area of Science:

• Medical imaging
• Ultrasound technology
• Signal processing

Background:

• Scanline interpolation is crucial for converting ultrasound scanline data into brightness-mode (B-mode) images.
• Conventional methods like bilinear interpolation have limitations in capturing spatial data dependencies, leading to inaccuracies.
• Deviations from the underlying probability distribution occur at interpolation points with standard methods.

Purpose of the Study:

• To introduce Gaussian process (GP) regression as an advanced technique for ultrasound scanline interpolation.
• To compare the accuracy of GP regression against traditional interpolation methods (nearest neighbor, bilinear, cubic spline).
• To evaluate the impact of interpolation window size on GP regression's accuracy and computational efficiency.

Main Methods:

• Utilized ultrasound scanlines from in vivo trials using two different ultrasound scanners.
• Applied Gaussian process regression and compared it with nearest neighbor, bilinear, and cubic spline interpolation.
• Quantified scanline conversion accuracy using Peak Signal-to-Noise Ratio (PSNR) and Mean Absolute Error (MAE).
• Investigated the effect of limiting the GP regression interpolation window size.

Main Results:

• Gaussian process regression demonstrated superior scanline conversion accuracy compared to all tested conventional methods for both datasets.
• PSNR and MAE scores confirmed the enhanced performance of GP regression.
• Reducing the GP regression interpolation window size to 15 minimized computational time with negligible impact on PSNR.

Conclusions:

• Gaussian process regression provides a quantitatively more accurate method for ultrasound scanline conversion.
• GP regression offers valuable uncertainty estimates at each interpolation point.
• The proposed windowing approach effectively reduces the computational cost associated with GP regression in scanline conversion.