Synchrotron X-ray tomographic microscopy Theory vs. practice

Transcription

1 Synchrotron X-ray tomographic microscopy Theory vs. practice Federica Marone Swiss Light Source, Paul Scherrer Institut, Villigen, Switzerland

2 Theory Radon transform Rf x = Beer-Lambert law I E = I 0 (E)e I E ln I 0 E = Fourier Slice Theorem Filtered Back Projection L f(x) dx L μ E dl μ E dl P θ = f x, y dl y x f(x, y)

3 Practice Standard microtomography setup to digital camera Dust/dirt Response Thickness Zingers Pixels Dynamic range PSF X-ray beam Scintillator Light microscopy objective Mirror Vibrations Monochromaticity Sample size Strong absorption Stability

4 Practice Standard microtomography setup to digital camera Dust/dirt Response Thickness Zingers Pixels Dynamic range PSF X-ray beam Scintillator Light microscopy objective Mirror Vibrations Monochromaticity Sample size Strong absorption Stability

5 Overview Flat field correction Beam stability Scintillator response Ring artifacts Local tomography artifacts Background noise Cupping artifact Sample stability Iterative algorithms Importance of phase contrast

6 Flat field correction Beer-Lambert law I E = I 0 (E)e Measurement μ E dl Flat field correction ln I E I 0 E = L μ E dl

7 Flat field correction Mechanical vibrations Several different frequencies (100 Hz, Hz, 66 Hz, 16 Hz) Intensity instabilities 3 ms exposure time 251 projections Sampled every 3 Movie 25 fps Scan is actually faster

8 Flat field correction Scintillator decay time and afterglow Injection 40x, LSO:Tb 5.9 um, 1% 20x, LAG:Ce 20 um, 1%

9 Flat field correction Scintillator decay time and afterglow 2x, LAG:Ce 20 um, 3.5%

10 Flat field correction Resulting artifacts Ring/band artifacts Non-quantitative Resolution degradation Solutions Average flat-field or intermediate flat-fields Flat-field tracking Adaptive time-dependent normalization Titarenko et al., 2010 Dual camera acquisition Mokso et al., 2013

11 Ring artifacts Inconsistencies in single or multiple channels over an extended range of views

12 Ring artifacts Inhomogeneities in the individual pixel response of detector elements Damaged dirty scintillator screens Beam variations

13 Ring artifacts Correction algorithms Simple to use, minimum amount of tuning parameters Sinogram domain Lines easier to detect Integration in existing pipelines difficult Reconstruction domain (Cartesian/Polar coordinates) Circles/rings difficult to detect Integration straightforward

14 Ring artifacts Correction algorithms Low/high pass filtering Oimoen et al., 2000, Yousuf and Asaduzzaman, 2010 Median/mean/moving average filtering Rivers at al., Boin and Haibel, 2006, Sadi et al., 2010 Transform domain processing Raven, 2010, Münch et al., 2009 Artifact template vector Sijbers and Postnov, 2004, Brun et al., 2009 A priori information Titarenko et al., 2010

15 Ring artifacts Correction algorithms None fully suitable for different (weak, varying intensity, band) rings without distortion and blurring Difficult choice of the optimal parameters Parameters often slice dependent Combination of approaches

16 Local tomography artifacts Large object at low resolution insufficient resolution for features of interest Tesselation time consuming, computationally heavy Larger detectors expensive, inefficient Ill-posed reconstruction problem x-rays CCD

17 Local tomography artifacts

18 Local tomography artifacts

19 Local tomography artifacts Marone et al., SPIE, 2010 Kyrieleis et al., J. Microscopy, 2010

20 Local tomography artifacts

21 Local tomography artifacts

22 Local tomography artifacts 0.5 mm Marone et al., SPIE, 2010

23 Local Original Local tomography artifacts ZP=0 ZP=0.5 ZP=1.5

24 Local tomography artifacts Simple lateral sinogram extension Insensitive to truncation degree Insensitive to position of region of interest Good results for structural analysis Non quantitative Difference in absolute value from slice to slice Calibration points if sample comparison needed More involved algorithms if absolute values needed

25 Local tomography artifacts Many ideas present in the medical imaging community Specific for medical applications Patent protected Various technique for projection completion Smooth continuation Iterative methods (e.g. sparsity, statistical) Computationally heavy Back projection of the first (Hilbert) or second (Lambda) derivative of projections A priori information needed For specific geometries Zoom-in tomography Multiple scans needed

26 Local tomography artifacts 4 mm sample 0.7 mm FOV prj

27 Local tomography artifacts 1501 prj 2251 prj 4501 prj

28 Sample stability 20 mm 20 mm

29 Sample stability Dynamic processes Animal physiology (brain, lungs, insect) Foam rheology Mechanical testing Sintering Annealing Electrochemistry Dose limitation In-vivo experiments Liquid foams surface tension properties Metals temperature change Micro-particles thermal dilation Fuel cell degradation Iterative reconstruction algorithms Few projections Short exposures (noisy projections) Phase contrast

30 Importance of phase contrast Towards in-vivo lung physiology Absorption CNR=1.8 Phase CNR=15.0 Total scan time: 5.4 s G. Lovric et al.,2013

31 Importance of phase contrast Monochromatic Polychromatic Foam physics CNR=4.9 CNR=0.7 CNR=4 Total scan time: 330 ms Phase Edge- enhancement Mokso et al., 2013

32 Importance of phase contrast Distance 1 Distance 2 Phase Edge- enhancement Mixed Guigay et al., 2007 TIE Paganin et al., 2002 Mokso et al., 2013

33 Importance of phase contrast Mixed Distance 1 Phase Edge- enhancement Mokso et al., 2013

34 Summary More advanced algorithms useful for different artifacts: Ring artifacts Local tomography artifacts Flat-field correction Ultrafast tomography Noisy projections -> Advanced phase retrieval algorithms Few projections -> Iterative algorithms

35 Acknoledgments Rajmund TOMCAT group Marco Thank you for your attention!

36 Streaking artifacts Inconsistencies in isolated measurements

37 Streaking artifacts High flux, high energy experiments Scattered x-rays hitting the camera chip directly Large energy deposition relative to the visible light photons Perfectly straight lines at random orientations in the image

38 Streaking artifacts High flux, high energy experiments Scattered x-rays hitting the camera chip directly Large energy deposition relative to visible light photons Perfectly straight lines at random orientations in the image

39 Streaking artifacts Correction algorithm (after M. Rivers) Low pass filter for each line in the sinogram Division of the original sinogram by the smooth one Substitute pixels with a grey level value smaller than threshold with average of neighboring pixels