锥束CT硬件加速：多平台并行重建策略评析

需积分: 31 111 浏览量更新于2024-07-20 收藏 3.41MB PDF 举报

锥束CT三维重建是一种先进的医学成像技术，其核心在于利用锥形X射线束对患者进行断层扫描，以获取高精度的三维图像。在实际应用中，快速重建是提高效率和降低辐射剂量的关键因素。本文档《锥束CT重建中各平台的硬件加速评价》由Holger Scherl撰写，发表于《Medizintechnik – Medizinische Bildgebung, Bildverarbeitung und bildgeführte Interventionen》期刊，由包括多个知名教授在内的编辑团队编审。文章详细探讨了如何在不同的硬件平台上实现锥束CT的快速重建，这些平台包括：CBEA（定制化的生物电子学架构）、FPGA（现场可编程门阵列）、GPU（图形处理器）以及多核CPU（中央处理器）。作者深入剖析了这些硬件在算法计算复杂度上的优势与挑战，以及它们如何通过并行计算技术来优化重建过程。 CBEA提供高度定制化的优势，能够针对特定任务设计专门的电路，从而实现高效的数据处理。FPGA则因其灵活性和可编程性，能动态调整计算任务，适用于需要实时处理的场景。GPU的设计初衷本就适合并行处理大量数据，其大规模并行执行单元使得图像重建的计算负载分布更均匀，大大提升了速度。多核CPU的优势在于其处理能力的扩展性，通过增加核心数量，可以同时执行多个任务，适合于那些对CPU性能要求较高的应用场景。然而，这种并行方式可能受到内存带宽和数据传输限制的影响。文章不仅关注硬件本身，还涵盖了算法流程的实现，如迭代法、滤波反投影等方法，以及如何在硬件选择上平衡计算效率和成本。此外，文中还提到了物理信号与组织相互作用的原理，这对于理解图像重建中的能量传递和组织特性至关重要。总结来说，这份PDF为CT领域技术人员提供了宝贵的参考，他们可以根据不同硬件平台的特点，选择最适合自己的方法进行锥束CT的三维重建硬件加速，从而推动医学影像诊断的精度和效率提升。这对于临床实践、研究和新技术发展具有重要的指导意义。

Contents

1 Introduction 1

1.1 Reconstruction Algorithms ........................ 1

1.2 HardwareVariants ............................ 4

1.3 RelatedWork ............................... 6

1.4 Scope and Main Contribution of this Thesis .............. 6

1.5 Outline ................................... 8

2 Algorithms for Cone-Beam Image Reconstruction 9

2.1 Prerequisites ................................ 9

2.2 The Feldkamp Algorithm ......................... 10

2.2.1 Algorithmic Steps ......................... 11

2.2.2 Implementation Strategies .................... 13

2.2.2.1 Ramp Filtering ..................... 13

2.2.2.2 Back-Projection ..................... 14

2.2.3 Complexity Analysis ....................... 17

2.2.3.1 Filtering ......................... 18

2.2.3.2 Back-Projection ..................... 19

2.2.3.3 Comparison of Computation Times .......... 20

2.3 Selected Alternatives to the Feldkamp Algorithm . . ......... 20

2.3.1 The M-Line Method ....................... 20

2.3.1.1 Algorithmic Steps ................... 23

2.3.1.2 Implementation Strategies ............... 25

2.3.2 Iterative Reconstruction ..................... 30

2.3.2.1 Simultaneous Algebraic Reconstruction Technique . . 31

2.3.2.2 Implementation Strategies ............... 32

2.3.2.3 Other Iterative Algorithms .............. 34

2.4 Summary ................................. 34

3 Design and Implementation of a General Reconstruction Frame-

work 37

3.1 Motivation ................................. 37

3.1.1 Cone-Beam CT Reconstruction ................. 38

3.1.2 Target Hardware Platforms ................... 39

3.1.2.1 Multi-Core and Multi-Processor Systems ....... 39

3.1.2.2 Cell Broadband Engine Architecture ......... 40

3.1.2.3 Common Uniﬁed Device Architecture ......... 40

xvi Contents

3.2 Reconstruction System Design ...................... 41

3.2.1 Reconstruction Pipeline ..................... 41

3.2.1.1 Pipeline Design Pattern ................ 41

3.2.1.2 Concurrency ...................... 42

3.2.2 Parallelization Strategy ...................... 42

3.2.2.1 Master/Worker Design Pattern ............ 42

3.2.2.2 Combination with the Pipeline Design Pattern .... 43

3.2.2.3 HardwareAbstraction ................. 44

3.2.3 Resource Management ...................... 44

3.2.3.1 Processing Elements .................. 44

3.2.3.2 Data Buﬀers ...................... 45

3.3 Implementation .............................. 45

3.3.1 Structure ............................. 46

3.3.2 Participants ............................ 46

3.3.3 Sample Code and Usage ..................... 48

3.4 Summary ................................. 51

4 Cell Broadband Engine Architecture 53

4.1 Architecture ................................ 53

4.1.1 The Power Processor Element .................. 54

4.1.2 The Synergistic Processor Elements ............... 55

4.1.3 The Element Interconnect Bus .................. 56

4.2 Feldkamp Algorithm ........................... 56

4.2.1 Filtering .............................. 57

4.2.2 Back-Projection .......................... 57

4.2.2.1 Problem Partitioning .................. 57

4.2.2.2 Main Memory Access .................. 63

4.2.2.3 Code Optimization ................... 63

4.3 Results ................................... 65

4.3.1 Filtering .............................. 65

4.3.2 Back-Projection .......................... 66

4.4 Summary ................................. 68

5 Standard Multi-Core Processors 71

5.1 Architecture ................................ 71

5.2 Feldkamp Algorithm ........................... 72

5.2.1 Filtering .............................. 72

5.2.2 Back-Projection .......................... 73

5.3 Results ................................... 74

5.4 Summary ................................. 75

6 Graphics Accelerator Boards 77

6.1 Architecture ................................ 77

6.2 Feldkamp Algorithm ........................... 80

6.2.1 Filtering .............................. 80

6.2.2 Back-Projection .......................... 80

6.3 Results ................................... 85

Chapter 1

Introduction

Computed tomography (CT) is a widely used imaging technique in the ﬁeld of med-

ical diagnosis and industrial non-destructive testing (NDT) applications. Using a

suﬃciently large series of X-ray images taken from diﬀerent views around the object

or patient a digital computer is used to generate slice images of the inside of the

considered object or patient.

In modern CT devices three-dimensional (3-D) reconstruction techniques are em-

ployed that are able to deal with the cone-beam geometry resulting from X-ray pro-

jections that are measured with a two-dimensional detector array. This imaging

approach enables the fast acquisition of the projection images and at the same time

delivers high image quality. The computational complexity of the applied 3-D recon-

struction methods prohibits their use in many medical applications without hardware

acceleration. The use of hardware acceleration is therefore an important area of re-

search in the ﬁeld of cone-beam reconstruction. It is very important that advances

in the theory of 3-D reconstruction go hand in hand with the development of suit-

able parallelization techniques that map them eﬃciently to appropriate hardware

acceleration platforms.

1.1 Reconstruction Algorithms

A computationally eﬃcient way to solve the reconstruction task is a technique which

is called ﬁltered back-projection (FBP). FBP methods achieve the reconstruction of

an object by ﬁrst ﬁltering the cone-beam projections and then computing the back-

projection of the ﬁltered projections into the 3-D space.

The most successful full 3-D FBP algorithm was developed by Feldkamp, Davis,

and Kress, which is commonly referred to as the FDK method [Feld 84]. The algo-

rithm deals with a circular acquisition trajectory, where the X-ray source and the

detector rotate around the center of the object to be scanned. It is equally possible

to ﬁx the X-ray source and detector while rotating the object, which is often the case

in NDT applications.

The FDK method is used in most of today’s cone-beam CT scanners such as

C-arm devices, radiation therapy devices, dental CT devices, and in a modiﬁed way

also in 3-D digital mammography devices and helical CT scanners (see Figure 1.1)

as the standard image reconstruction approach.

H. Scherl, Evaluation of State-of-the-Art Hardware Architectures for Fast

Cone-Beam CT Reconstruction, DOI 10.1007/978-3-8348-8259-2_1,

剩余156页未读，继续阅读

北冥有鱼1989

粉丝: 27
资源: 13

锥束CT硬件加速：多平台并行重建策略评析

GPU加速的锥束三维重建FDK实现

三维锥形束CT成像FDK重建算法发展综述

三维锥形束CT解析重建算法发展综述

FPGA加速三维CT图像重建

FPGA加速三维CT图像重建.pdf

GPU多重纹理加速三维CT图像重建.pdf

利用GPU实现单层螺旋CT的三维图像重建.pdf

CBCT.zip_FDK_FDK CBCT 算法_GPU加速_cbct的算法_三维图像重建

ct-算法重建-马建-fdk重建算法.pdf

CT重建中投影矩阵模型研究综述

最新资源