利用資料壓縮以改善即時軟體超音波成像系統

Improving Performance of Real Time Software-Based Beamformers Using Data Compression

陸裕威(U-Wai Lok)

指導教授 ： 李百褀

國立臺灣大學/電機資訊學院/生醫電子與資訊學研究所/博士(2017年)

https://doi.org/10.6342/NTU201700396

摘要

超音波陣列影像系統由於成像所需資料量龐大，過往在硬體電路上利用高度平行架構達成即時影像處理。但使用硬體架構的缺點之一為電路的功能固定因此較難有高度之可程式化設計，演算法的開發因此亦需較長之時間。然而隨著軟體運算能力進步，以圖形處理器(GPU)為基礎的軟體陣列成像系統能有效地降低硬體資源和系統開發速度。但實現軟體成像系統的主要瓶頸之一是大量超音波射頻訊號需要從硬體的記憶體中，藉由傳輸介面傳送到軟體端。因此資料壓縮成為軟體成像之主要工作。但目前的資料解壓縮方法不適合平行處理，解壓縮成為即時軟體陣列成像系統的另一瓶頸。本研究主要提出快速壓縮-解壓縮演算法來改善軟體陣列系統的資料傳輸問題。本研究的第一部份主要介紹二維成像系統的壓縮-解壓縮演算法，其主要分為兩個方法。第一種是提出無失真壓縮-解壓縮演算法，這方法能在GPU上能有效達成平行解壓縮資料，因此2-3毫秒的時間即可完成解壓縮一張影像所需的資料，而此方法的壓縮率約為1.7。第二種是導入沃爾什轉換 (Walsh transform)壓縮方法，針對振幅資訊實現接近無失真資料壓縮，以增加資料壓縮率。此方法能使壓縮率從1.7上升為接近2.2，並且也保留快速解壓縮能力。為了實際驗證我們的研究在有限傳輸介面的可行性，我們將沃爾什轉換的壓縮方法應用在64 通道的系統、並且利用單一USB 3.0 傳輸線作為傳輸介面。結果顯示超音波訊號經過壓縮後能利用此傳輸介面作即時傳輸。最後，我們提出微波束成像以及錯誤補償演算法來實現資料壓縮。此方法亦可在類比訊號實現，因此可用於使用二維陣列之三維超音波影像，並降低類比-數位轉換器以及線路連接的複雜度。在這部份可達到5-6倍的壓縮率。而微波束成像方法也能應用在數位系統上作資料壓縮。此方法能跟沃爾什轉換方法相容並且提供約3.1的壓縮率，硬體資源利用率只需多約2-3%。本研究實現了平行壓縮-解壓縮演算法以滿足2-D軟體超音波的傳輸速率之需求，而在使用二維陣列之三維超音波影像中也實現了錯誤補償演算法以減少系統複雜度並維持成像品質。未來工作將會專注於減少三維成像系統之計算複雜度以及最佳化錯誤補償演算法之電路設計與實現。

關鍵字

GPU ； 資訊壓縮 ； 波束成像 ； 微波束成像 ； 軟體陣列成像系統

並列摘要

Graphics processing unit (GPU)-based software beamforming has advantages over hardware-based beamforming such as higher programmability, channel data preservation and shorter design cycles. However, the need for a high data transmission rate when transferring ultrasound radiofrequency (RF) data from the front end to the back end is a major technical challenge. Data compression methods can be applied in the front end to mitigate this issue. Nevertheless, many decompression processes cannot be performed efficiently on a GPU in real time, thus presenting another bottleneck for real-time imaging. The aim of this dissertation is to develop algorithms enabling parallel compression and decompression of ultrasound channel data. With the presented methods, the data transfer rate can be reduced and decompression can be performed on the GPU in real time. In the first part of this dissertation, a real-time lossless compression-decompression algorithm is presented. This method exploits low hardware resource requirements, fast compression-decompression speed as well as high image quality. The compression ratio of this approach is approximately 1.7. In the second part of the dissertation, Walsh transform is integrated with the previous lossless compression method to further enhance the compression efficiency. With this integrated approach, the compression ratio can be improved from 1.7 to 2.2, while preserving the compression-decompression speed at the expense of slightly increased hardware utilization. With these, the overall data compression rate can reach to 5-6 with demodulation and down-sampling. To test the performance of the proposed methods, the Walsh-transform compression method is implemented on a 64-channel system with limited data transfer rate via a single USB 3.0 cable. For the third part of the dissertation, analog micro-beamforming (MBF) method is investigated to reduce the system complexity of the 3-D ultrasound imaging system. To improve the image quality of MBF approach, the error compensation method is proposed. The compensation approach can be applied in the analog domain to reduce system complexity of a 2D array system by reducing the number of cables and analog-to-digital converters. Furthermore, this approach can also be used digitally to suppress channel amplitude data as a lossy data compression method. This approach is compatible to the Walsh-transform-based compression algorithm with only 2-3 % increment of hardware resources. The compression ratio of this approach is nearly 3.1 as well. In summary, in the thesis fully parallel compression-decompression algorithms was realized to mitigate the transmission rate requirements for software-based beamformers for 2-D imaging system. The system complexity for 2D array system can also be reduced with the MBF with the compensation method. Future work will focus on developing a real-time 3-D ultrasound imaging system with reasonable computational complexity and high image quality.

並列關鍵字

GPU ； software beamforming ； data compression ； micro-beamforming ； error compensation ； Walsh transform

參考文獻

[1] M. Karaman, E. Kolagasioglu, and A. Atalar, "A VLSI receive beamformer for digital ultrasound imaging," IEEE Int Conf. on Acous., Speech, and Signal Processing, vol. 5, pp. 657 - 660, Mar 1992.

[2] K. E. Thomenius, "Evolution of ultrasound beamformers," IEEE Int. Ultrason. Symp., vol. 2, pp. 1615-1622, Nov 1996.

[3] M. Karaman, A. Atalar, and H. Koymen, "VLSI circuits for adaptive digital beamforming in ultrasound imaging," IEEE Trans. Med. Imaging, vol. 12, pp. 711-20, Aug 1993.

[4] H. Li and D. C. Liu, "An Embedded high performance ultrasonic signal processing subsystem," IEEE Int. Conf. Embed. Software System, pp. 125-130, May 2009.

[5] F. K. Schneider, A. Agarwal, Y. M. Yoo, T. Fukuoka, and Y. Kim, "A fully programmable computing architecture for medical ultrasound machines," IEEE Trans Inf Technol Biomed, vol. 14, pp. 538-40, Mar 2010.