研究目的
To implement median filtering with asymmetric kernel sizes on intermediate ultrasound data to reduce speckle noise and improve computational efficiency for real-time imaging.
研究成果
Asymmetric median filtering on intermediate ultrasound data improves image quality with edge preservation. Optimal kernel sizes are around 3x9 to 3x23. GPU implementation reduces computational time by approximately half compared to CPU, making it feasible for real-time imaging at 25 fps.
研究不足
The study is limited to specific ultrasound data with high axial sample counts; kernel sizes are small (up to 11x23), and improvements may not generalize to larger kernels or other imaging systems. GPU optimization is dependent on register usage and compiler behavior.
1：Experimental Design and Method Selection:
The study designed median filtering with asymmetric kernel sizes for ultrasound data, using median selection networks and compare-and-swap stages optimized for CPU and GPU. Theoretical models include sorting networks and CUDA programming.
2：Sample Selection and Data Sources:
Ultrasound data were acquired from a phantom using a linear array probe, with 81 scanlines and 8,192 samples per scanline, converted to 16-bit intermediate envelope data.
3：List of Experimental Equipment and Materials:
Laptop computer (Asus G750JW) with Intel Core i7-4700HQ CPU, NVIDIA GeForce GTX 765M GPU, Windows 8 OS, Microsoft Visual Studio 2012, CUDA library v6.5.14, linear array probe (Sonostar L7I40), phantom (Gammex 403GS LE).
4：14, linear array probe (Sonostar L7I40), phantom (Gammex 403GS LE).
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Data acquisition, beamforming, envelope detection, median filtering implementation on CPU and GPU using SSE2 and CUDA, with kernel sizes varied in axial and lateral directions. Computational time and CNR were measured over 500 repetitions.
5：Data Analysis Methods:
Computational time measured using clock() for CPU and CUDA events for GPU; CNR calculated using mean and standard deviation of pixel values in ROIs.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
