Image reconstruction algorithms for microtomography

Andrei V. Bronnikov

Bronnikov Algorithms, The Netherlands

Contents
Introduction Fundamentals of the algorithms State-of-the-art in 3D image reconstruction Phase-contrast image reconstruction Summary

BronnikovAlgorithms

Microtomography systems
Desktop systems
Image Intensifier Object Camera control X-Ray tube Camera Manipulation system

PC
Motor control

Rail system

NDT systems Dental CBCT

Small animal CT

Nano x-ray microscopy

Synchrotron setup

BronnikovAlgorithms

Micro CT images

Stampanoni et al Sterling et al Dental CBCT

BronnikovAlgorithms

Problems
Object preparation, fixation, irradiation, etc Polychromatic source, miscalibrations, etc Small object size: insufficient absorption contrast Limited field-of-view, limited data, incomplete geometry Large amount of digital data

BronnikovAlgorithms

Solutions
Region-of-interest reconstruction Fully 3D cone-beam scanning and reconstruction The use of phase contrast Software/hardware acceleration

BronnikovAlgorithms

Geometry

Parallel beam
Synchrotron The source is far away from the object

Cone beam
Microfocus tube, microscopy The source is close to the object: - Increased flux - Magnification - Fully 3D

BronnikovAlgorithms

I nverse problem

Projection data: g , s

f dl
Line ( , s )

Radon transform

Af

s
Object: f

To find f from g ?

BronnikovAlgorithms

Backprojection Integration of the projection data over the whole range of * A g

1
0

g d

BronnikovAlgorithms

Algorithms: classification
Radon transform Fourier algorithm Filtered backprojection (FBP) Backprojection and filtering (BPF) Iterative

Af

g
* 1 * *

Imaging equation

A A
*

A g
1

BPF

A AA
1

FBP

Fn Fn 1 g

Fourier

BronnikovAlgorithms

Parallel-beam geometry (Synchrotron)

BronnikovAlgorithms

Fourier slice theorem

BronnikovAlgorithms

I mage reconstruction w ith NFFT

F2 F1 g

Interpolation from the polar grid to the Cartesian is required

Linogram (pseudo-polar) grid

Nonequispaced Fast Fourier transform (NFFT) can be used

Potts et al, 2001
BronnikovAlgorithms

FBP and BPF algorithms

* 1 * * * 1

f
F2 A A
* 1

A A
2

A g
2

A AA

F1 AA

ramp filter

F2

2 0

g d

1
0

F1 1

g d

BronnikovAlgorithms

FBP algorithm

1D Filtering

Backprojection

f
0

q g d

BronnikovAlgorithms

Local ( Lambda) tomography

F1 1 g A* H

Local operator

Hilbert transform is non-local:

Hg ( s )

g (t ) dt s t
BronnikovAlgorithms

Cone-beam geometry (Microfocus x-ray tube)

BronnikovAlgorithms

Feldkamp algorithm w ith a circular orbit

Feldkamp, Davis, Kress, 1984

Filtering

Backprojection

f
0

q g d
BronnikovAlgorithms

Kirillov-Tuy condition
detector Exact 3D reconstruction is possible if every plane through the object intersects the source trajectory at least once s a( ) f u

g (a( ), u )
0

f (a( ) su )ds

x-ray source trajectory; parametrized as a( )

BronnikovAlgorithms

Circular source orbit: artifacts

Slices of 3D reconstruction of a phantom (cone angle 30 deg):

Bronnikov 1995, 2000

BronnikovAlgorithms

ROI reconstruction

Two-step data acquisition

Position 1 Source ROI Sample

Position 2

Sample

Detector

BronnikovAlgorithms

Non-planar source orbits

Non-planar 3D reconstructions of a phantom:

- two

orthogonal circles - two circles and line - helix (most feasible mechanically) - saddle

Non-planar orbits satisfy the Kirillov-Tuy condition, but special reconstruction algorithms are required
BronnikovAlgorithms

Katsevich algorithm for a non-planar source orbit

Katsevich, 2002

f
a( 2)

1 * A H 2

a( 1)

1. Differentiation of data 2. Hilbert transform along the filtration lines inside the Tam-Danielson window 3. Backprojection

PI-line (segment, chord) between a( 1) and a( 2):

1=2

Helix:

h R cos , R sin , 2

BronnikovAlgorithms

BPF algorithms for ROI reconstruction

f 1 H A* 2 g

1. Differentiation of data 2. Backprojection onto the PI chord (locality!) 3. Hilbert transform along the PI chord

Using that f has the finite support and

HHg
a( )

Zou, Pan, Sidky, 2005 derived:

1 H 2

1 A* ,a1 a 2

BronnikovAlgorithms

Ultra-fast implementation
Graphic card (GPU) CPU
Reconstruction of a 512x512x512 image from 360 projections:

CPU (~2 GHz) : Time :

Single ~80 sec

Dual core ~40 sec

Quad core ~20 sec

Twin quad-core ~10 sec

Reconstruction of a 1024x1024x1024 image from 800 projections:

CPU (~2 GHz) : Time :

Dual core ~480 sec

Twin quad-core ~120sec

BronnikovAlgorithms

Phase-contrast microtomography (Free propagation mode)

BronnikovAlgorithms

Phase contrast
Interference of the phase-shifted wave with the unrefracted waves

BronnikovAlgorithms

I nline phase-contrast imaging

Snigirev et al, 1995

BronnikovAlgorithms

Polychromatic x-ray phase contrast

Wilkins et al, 1996

BronnikovAlgorithms

Phase-contrast tomography w ith Radon inversion: edges

BronnikovAlgorithms

I nverse problem of phasecontrast microtomography

Object function: f = n 1

find

f ( x1 , x2 , x3 ) from

I ( x, y ), 0

CTF (Cloetens et al, 1999) TIE (Paganin and Nugent, 1998)

Phase retrieval, more than one detection plane

Weak-absorption TIE (Bronnikov, 1999)

FBP, single detection plane

BronnikovAlgorithms

Radon transform solution of TI E

I ( x, y )
Bronnikov, 1999
d

d 2

( x, y )

g ( x, y )

I d / Ii 1

1 4
2

g ( s, ) d d

g Object f d
BronnikovAlgorithms

Phase-contrast reconstruction in the form of the FBP algorithm

Bronnikov, 1999, 2002, 2006

q
0

g d

y x2 y2
2D Filtering

Gureyev et al, 2004: choice of for linearly dependent absorption and refraction

Backprojection
BronnikovAlgorithms

I mplementation at SLS
MBA: Modified Bronnikov Algorithm Groso, Abela, Stampanoni, 2006

Phase tomography reconstruction (a) and the 3D rendering (b) of a 350 microns thin wood sample using modified filter given in the Eq. (8). The length of the scale bar is 50 m.

Validation of the MBA method: (a) Phase tomographic reconstruction of sample consisting of polyacrylate, starch and cross-linked rubber matrix obtained using DPC and (b) using MBA. The length of the scale bar is 100 m.
BronnikovAlgorithms

I mplementation at Ghent University

De Witte, Boone, Vlassenbroeck, Dierick, and Van Hoorebeke, 2009

Radon inversion

Modified Bronnikov Algorithm Bronnikov-Aided Correction

BronnikovAlgorithms

Polychromatic source, mixed phase and amplitude object

Air bubbles in epoxy, relatively strong absorption:

6 mm

Reconstruction by the Bronnikov Filter with correction

Data provided by Xradia

BronnikovAlgorithms

Summary
New developments in theory: parallel-beam CT (synchrotron): the use of NFFT cone-beam CT (microfocus): exact reconstruction with nonplanar orbits; exact ROI reconstruction (Katsevich formula, PI line, Hilbert transform on chords) New developments in implementation: ultra-fast 3D reconstruction on multicore processors (10243 voxels within one-two minutes on a PC) New developments in coherent methods: robust algorithms for 3D phase reconstruction correction for the phase component

BronnikovAlgorithms