請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/64990
標題: | 適用於超音波都卜勒系統之特徵引擎演算法及架構設計 Algorithm and Architecture Design of Eigen-based Doppler Engine in Ultrasound Systems |
作者: | Cheng-Zhou Zhan 詹承洲 |
指導教授: | 吳安宇 |
關鍵字: | 超音波,血流,都卜勒, ultrasound,blood flow,Doppler, |
出版年 : | 2012 |
學位: | 博士 |
摘要: | 超音波影像(ultrasonic imaging)因為有低成本、非侵入性(non-invasiveness)、即時成像、與儀器較方便攜帶等優點,所以是一種被廣泛使用的造影方式。而其中彩色都卜勒(color Doppler)影像是一種在醫學界被廣泛使用的超音波成像模式。其主要的價值在於它可以即時(real-time)地呈現出體內流速影像,應用的對象包含血流及心搏等。然而目前超音波系統的可攜性仍明顯不足,不但會使病人必須親自到醫院才能作檢查,造成病人的痛苦,且使這樣的醫療資源難以普及化,尤其在一些需要第一線診斷的情況如救護車或戰場等,大型的醫療設備更無法提供立即的救助。
而傳統的都卜勒引擎需要複雜的參數設定,所以沒有辦法在環境變化的情形下適當的自動調節參數即時成像。所以我們希望可以不僅讓受過訓練的專業人員,使其繁複的操作步驟簡化,且降低儀器的操作介面面積以增加可攜性,這些都有利於可攜式系統之使用便利性與有效性。本論文的研究主題是針對我們正在發展的一套可攜性超音波影像系統,設計並實現智慧型彩色都卜勒的數位訊號處理引擎(Intelligent color Doppler DSP engine)。由於此超音波系統的目標為智慧型、良好的影像品質及高攜帶性。首先,本設計根據智慧型(Intelligent)的特色自動根據處理環境調整參數以及智慧地選擇所需的計算複雜度,其具有的彈性能根據目前環境需求變化動態自行調整係數,方便使用者對各種不同的目標環境都能夠掌握到實用的資訊,並兼顧良好的抗雜訊能力。 在超音波都卜勒信號處理用於顯示血流資訊時,由於血流信號相較於周圍的組織及背景雜訊微弱許多,需要各式的濾波器將雜訊濾除,其中障壁濾波器(clutter filter)便是用以濾除影響最大的組織雜訊,傳統上是以高通濾波器(high-pass filter)實現。近年有以奇異值或特徵值分解理論(eigen-based theory)為基礎的濾波演算法,具有信號可適性(signal-adaptive)特性可隨目前環境調整濾波特性,符合智慧型動態調適的特性,並可大幅增強障壁濾波器之效能,但在實際應用上仍有運算複雜度過高及運算速度受限等問題,在演算法上亦有可改進之處。 本論文的研究主題主要有三個部分,第一部分首先針用於障壁濾波器的特徵值演算法提出改進,利用所收到超音波信號於特徵值與頻率上之特性,使用聯合決策機制改善其效能;同時根據超音波觀測生理部位時其連續變化特性為基礎,提出回溯確認機制進一步增加信號判讀正確性,相較於目前文獻可改善2-5dB的處理效果。在本論文的第二部分中,進一步設計一套非線性軟決策閥值(nonlinear soft-decision thresholding)機制,相較於傳統的硬決策(hard decision)機制,本論文藉由組織與血流在時間與空間上不同的統計特性使用軟決策(soft decision)機制,並使用最大似然(maximum likelihood, ML)以及最大事後機率(maximum a posteriori, MAP)之方法,更準確地判斷血流信號並濾除信號通過障壁濾波器後殘餘的組織及背景雜訊,相較於文獻可改進5-10dB之效能。 本論文的第三部分對目前以奇異值或特徵值分解為基礎的高效能濾波引擎,在實際應用時會遭遇之運算速度較慢及運算複雜度過大之問題,提出一套超線性收斂(superlinear convergence)之快速奇異值分解演算法,同時兼具可快速運算及總運算複雜度低的優點,並以硬體實現其架構。其硬體實現後的吞吐量可達現有文獻之20-40倍,等效的單位面積吞吐量效率也能達到約5倍之多,並能同時支援數種不同大小矩陣之奇異值或特徵值分解運算。 Ultrasonic imaging is a well-established imaging modality that has the advantages of cost-effectiveness, non-invasiveness, rapid imaging, and portability. color Doppler imaging is a well-established Doppler ultrasound mode and very valuable for visualizing in the real-time distribution of blood flow in a specific region of interest. The blood velocity can be detected by measuring the Doppler frequency shift of the echoes back-scattered from the moving blood. Nevertheless, the portability of current high-frequency ultrasonic imaging systems is insufficient. This will be more and more unacceptable since it is inconvenient and sometimes harmful for patients to go to the hospital for diagnoses. The high cost also reduces the popularity of such medical devices. Moreover, in ambulances or battlefields, portable devices supporting immediate diagnosis can greatly increase the survival rate. Furthermore, the traditional Doppler engine needs complex parameter settings, it is inconvenient for the user, and it would not work when the users are not familiar with parameter settings for particular environment. Therefore, we simplify complex operations and enhance the efficiency. These features improve the convenience and efficiency of the portable ultrasound systems. The blood signals are generally much weaker than tissue and background noises, so that various filters are required to mitigate the noises. The clutter filtering in the Doppler flow imaging is utilized to suppress the great tissue noises. The performance of the clutter filtering will remarkably affect the results of the Doppler flow imaging. Compared with the traditional high-pass filters adopted as clutter filters, singular-value-based or eigenvalue-based filters are proposed for signal-adaptive properties and high de-noising performance. Nevertheless, the high computational complexity and long computational time are still the main limitations for real-time Doppler flow imaging. There are three main topics in this work. In the first part, the joint-decision mechanism with trace-back verification scheme is proposed to improve the performance of the eigen-based clutter filtering. After receiving the ultrasonic signals of the region of interest, the joint-decision mechanism is proposed to divide the signal spaces as clutter-noise space and blood-signal space according to the properties of eigenvalues and corresponding frequencies. In addition, the trace-back verification scheme is proposed according to the correlation of successive Doppler frames. The performance is 2-5 dB better than that of referenced works. In the second part of this work, a nonlinear soft-decision thresholding after clutter filtering is proposed. The thresholding scheme is conventionally utilized to eliminate the remaining tissue or background noises after the clutter filters. Compared with the traditional hard-decision thresholding, we use the statistical properties of bloods and tissues for nonlinear thresholding. We use the maximum likelihood (ML) and maximum a posteriori (MAP) tests for decision. There is about 5-10 dB performance improvements compared with that of the referenced works. In the third part, a superlinear-convergence singular-value-decomposition (SL-SVD) algorithm is proposed with properties of high-speed and low total computational complexity. The SL-SVD may be used to mitigate the problems of computational complexity and processing time in the singular-value-based or eigenvalue-based filters as clutter filters. The hardware is also implemented in 90 nm technology. The implementation results of SL-SVD has 20-40 times throughput compared with that of referenced works. The design is also capable of supporting different sizes of matrixes. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/64990 |
全文授權: | 有償授權 |
顯示於系所單位: | 電子工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-101-1.pdf 目前未授權公開取用 | 6.48 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。