以軟體基礎實現醫用陣列超音波影像

Software-Based Implementation of Ultrasonic Array Imaging

李彥鋒(Yen-Feng Li)

指導教授 ： 李百祺

國立臺灣大學/電機資訊學院/生醫電子與資訊學研究所/碩士(2010年)

https://doi.org/10.6342/NTU.2010.03259

摘要

本研究提出陣列超音波成像的軟體實現架構，以軟體方法取代傳統陣列超音波的硬體電路。其優勢在於可以減少系統複雜度，也降低開發所需時間與成本。以CPU為基礎的軟體成像系統，其系統表現受到CPU架構限制，平行能力較低。本研究提出以GPU為基礎的架構，利用GPU的高平行能力與影像處理的優勢，並使用 CUDA來設計成像程式。成像功能則包含調相陣列成像、彩色都普勒影像與合成發射孔徑成像。
本軟體系統實現在由NVIDIA GeForce GTX 260顯示卡 與CPU為Intel core i7 920組成的電腦系統上。其扇形掃描在使用128單元、2048點接收長度與由189條波束組成影像的測試資料可達到每秒42張的成像率。相同格式的測試資料在使用每張影像16次發射的合成發射孔徑技術可達每秒40張的成像率。彩色都普勒成像使用128單元、2048點接收長度與181條波束的B-mode測試資料與1024點長度，90條波束、每方向發射8次的流速資料。使用傳統方法的成像率為每秒14張，使用雙波束法則可達到每秒20張。
軟體系統大量的傳輸資料需求可能會成為成像速度的瓶頸，為了達到即時成像，本研究提出壓縮傳輸資料的方法來克服傳輸資料速度的限制。使用JPEG技術對傳輸資料壓縮後合成影像並評估影像品質，並以影像點平均誤差小於4dB做為指標。JPEG在壓縮10倍後仍能保有可接受品質，使用JPEG2000技術則可進一步壓縮到20倍。壓縮率太高則會造成信號失真，降低影像品質。
為了做到即時合成發射孔徑流速偵測，本研究也探討彩色都普勒影像應用在合成發射孔徑流速偵測的可能性，使用field II模擬超音波信號的方法並比較各種方法的偵測準確度。雖然合成發射孔徑流速偵測的可以每次發射都合成一張流速影像，且在無雜訊時可以達到比平面波或使用一半樣本數計算流速傳統方法低約1%內的誤差率，但在SNR為20dB時準確率會低於另外兩個方法。其最大可偵測流速亦會受限，為傳統方法的一半。低誤差耐受性與最大可偵測流速成為合成發射孔徑流速偵測的主要限制。

關鍵字

陣列超音波成像 ； 軟體平行技術 ； CUDA ； 合成發射孔徑成像 ； 合成發射孔徑流速偵測 ； 超音波信號壓縮

並列摘要

In this research, we investigate software-based ultrasonic array imaging. The advantages of a software-based architecture include lower hardware complexity, development time and cost. size. Compared to CPU-based systems, with which performance is often limited by its inadequate parallelism, the GPU-based architecture has the potential of increasing computation speed significantly. In this research, CUDA GPU-based parallel programming is used. The imaging tasks under investigation include phased array imaging, color Doppler, and synthetic transmit aperture (STA) imaging.
The software is implemented with NVIDIA GeForce GTX 260 and Intel core i7 920 CPU. Frame rate of this system is 42 fps with a 128 elements array transducer, 2048 samples per firing and 189 beams per image. Frame rate of STA imaging is 40 fps with 16 firings per image. Color Doppler imaging is tested with similar B-mode data, and flow data consisted of 1024 samples per firing, 90 beams and an ensemble of 8 firings per beam. The frame rate is 14 fps with single beam, and it is 20 fps with dual beams.
The large data transfer requirement of software-based system may be a bottleneck to achieve real time imaging. In this research, we also investigated channel data compression methods to overcome the limitation. In channel data compression, when the average error is under 4dB, the maximum acceptable compression ratio is 10 with JPEG and 20 with JPEG2000.
In order to realize real time STA flow estimation, accuracy of STA color Doppler is investigated by simulations. Field II is used to compare accuracy of various flow estimations. The frame rate of STA color Doppler is equal to PRF, and the error is about 1% lower than plane wave and conventional method with half autocorrelation samples. However, the accuracy of STA color Doppler is lower than the other methods when SNR is 20dB or lower. Also, the maximum detectable velocity with STA color Doppler is half of that with conventional method.

並列關鍵字

ultrasonic array imaging ； software parallel programming ； CUDA ； synthetic transmit aperture imaging ； synthetic transmit aperture flow estimation ； ultrasonic signal compression

參考文獻

[1] K. K. Shung, Diagnostic Ultrasound: Imaging and Blood Flow Measurements, 2006.

[2] T. L. Szabo, Diagnostic Ultrasound Imaging: Inside Out, 2004.

[5] M. J. Quinn, Parallel Programming in C with MPI and OpenMP: McGraw-Hill, 2004.

[6] S. Che, M. Boyer, J. Meng, D. Tarjan, J. W. Sheaffer, and K. Skadron, "A performance study of general-purpose applications on graphics processors using CUDA," Journal of Parallel and Distributed Computing, vol. 68, pp. 1370-1380, 2008.

[7] R. Shams, P. Sadeghi, R. Kennedy, and R. Hartley, "A Survey of Medical Image Registration on Multicore and the GPU," Signal Processing Magazine, IEEE, vol. 27, pp. 50-60, 2010.