GPU-based volume reconstruction for freehand 3D ultrasound imaging

Area of Science:

• Medical Imaging
• Computational Ultrasound
• Image Processing

Background:

• Volume reconstruction is crucial for enhancing image quality in freehand 3D ultrasound.
• Kernel regression is effective for 3D ultrasound volume reconstruction but computationally intensive.

Purpose of the Study:

• To propose a fast kernel regression method utilizing graphic processing units (GPUs) for freehand 3D ultrasound volume reconstruction.
• To leverage GPU's data-parallel computing for improved efficiency and investigate parameter settings for optimal image quality.

Main Methods:

• Implementation of a kernel regression algorithm on programmable GPUs.
• Exploration of various parameter settings (kernel window size, bandwidth) for enhanced image reconstruction.
• Comparative analysis of GPU-based versus CPU-based computational performance.

Main Results:

• The proposed GPU-based method achieves over 200 times faster computational performance compared to CPU-based methods.
• Optimal image quality, including speckle reduction and detail preservation, is achieved with a kernel window size of 5×5×5 and a kernel bandwidth of 1.0.
• The study demonstrates the effectiveness of GPU acceleration for 3D ultrasound volume reconstruction.

Conclusions:

• GPU-based fast kernel regression significantly accelerates freehand 3D ultrasound volume reconstruction.
• Specific parameter settings (5×5×5 window, 1.0 bandwidth) yield superior image quality.
• This method offers a computationally efficient solution for high-quality 3D ultrasound imaging.