Introduction

  1. Motivation
  1. X-ray computed tomography
  1. Experiment design
  1. Automated data analysis
  1. Hypotheses and open questions
  1. Objectives and thesis outline
  • Salvino D'Armate 13th century, "flea glasses", 6x - 10x
  • 1590's, two Dutch spectacle makers, Zacharias Jansen and his father Hans; "compound microscope"; Compound microscopes feature two or more lenses, connected by a hollow cylinder (tube).
  • in the very beginning, the optical microscope was used for studying sample with magnification
  • optical microscopes, limited penetration depth by visible light, very small magnification
  • very limited application, imperfections of construction
  • studying animals, insects, plants with very small throughput
  • no way for non-invasive studies, samples should be cutted
  • resolution, magnification limits
  • http://www.impactscan.org/CThistory.htm
  • Wilhelm Röntgen is usually credited as the discoverer of X-rays in 1895
  • Thomas Edison, March 1896, the fluoroscope he developed became the standard for medical X-ray examinations.
  • 14 February 1896 Hall-Edwards was also the first to use X-rays in a surgical operation
  • about 1906, the physicist Charles Barkla discovered that X-rays could be scattered by gases, and that each element had a characteristic X-ray spectrum
  • the 1920s through to the 1950s, x-ray machines were developed to assist in the fitting of shoes and were sold to commercial shoe stores
  • opening x-ray allowed for investigation of things which were hidden before
  • medical imaging, radiography
  • more efficient treatment of broken bones, and discovering pathologies
  • for diagnostic and therapeutic purposes in various medical disciplines
  • Since its introduction in the 1970s, CT has become an important tool in medical imaging to supplement X-rays and medical ultrasonography.
  • more recently been used for preventive medicine or screening for disease, example CT colonography for people with a high risk of colon cancer, or full-motion heart scans for people with high risk of heart disease
  • the potential to identify disease (e.g. cancer) in early stages, and early identification can improve the success of curative efforts
  • Nondestructive testing or non-destructive testing (NDT)
  • rediscovering of x-ray at particle accelerators allowed to build synchrotrons - the particle accelerators dedicated to x-ray imaging in various ways
  • non-destructive control, industrial imaging, fragile samples for invasive imaging (e.g. fossils)
  • archaeological uses such as imaging the contents of sarcophagi
  • higher resolution, speed, energy
  • the emergence of rapid and effective imaging methods allowed for studying of series of similar samples, that was required for correlative analysis amoung studied samples
  • such kind of studies require dedicated processing workflows since a large number of samples cannot be processed manually as it was in case of single samples
  • many approaches were proposed, some of them in recent time
  • the development of detectors contributed to increasing the size of produced data, which grows very rapidly
  • the produced data should be placed in a reliable and persistent storage
  • the processing workflows should be efficient in memory, processing requirements, and execution time

image

  • Nuclear magnetic resonance was first described and measured in molecular beams by Isidor Rabi in 1938
  • NMR is also routinely used in advanced medical imaging techniques, such as in magnetic resonance imaging (MRI)
  • In the late 1970s, physicists Dr. Peter Mansfield and Dr. Paul Lauterbur, developed MRI-related techniques, like the echo-planar imaging (EPI) technique.[
  • Ultrasound imaging or sonography is often used in medicine. In the nondestructive testing of products and structures, ultrasound is used to detect invisible flaws
  • Echolocation in bats was discovered by Lazzaro Spallanzani in 1794, when he demonstrated that bats hunted and navigated by inaudible sound and not vision
  • The first technological application of ultrasound was an attempt to detect submarines by Paul Langevin in 1917.
  • Ultrasonic testing is a type of nondestructive testing commonly used to find flaws in materials and to measure the thickness of objects.
  • The potential for ultrasonic imaging of objects, with a 3 GHZ sound wave producing resolution comparable to an optical image, was recognized by Sokolov in 1939 but techniques of the time produced relatively low-contrast images with poor sensitivity. Ultrasound has been used by radiologists and sonographers to image the human body for at least 50 years and has become a widely used diagnostic tool.
  • Single-photon emission computed tomography (SPECT
  • https://hal-insu.archives-ouvertes.fr/hal-00101634/document
  • high-resolution X-ray detectors based on a CCD camera exhibiting simultaneously a high dynamic range and a fast readout
  • http://www.esrf.eu/UsersAndScience/Publications/Highlights/2002/Imaging/IMA1
  • In many cases the scientific data have to be extracted from a statistically relevant
    number of samples. This is the case, for instance, for the quantification of the action
    of a drug on osteoporosis, where a series of bones of animals submitted to the drug are imaged to obtain a result going beyond the variability of reaction of the
    individuals. For this kind of experiments it is important to automate as much as
    possible the acquisition. An automation device has been developed and was used,
    for instance, when analysing paper microstructure (Thibault et al., 2005). A large
    number of different samples were investigated without having to re-enter into the
    hutch, thus without changing the environmental conditions of the beamline optics
    and the specimen. For such high-resolution experiments on samples that are very
    sensitive to small hygrometry changes, image quality is highly improved by
    controlling in this way the experimental conditions.

Book all in one: https://drive.google.com/open?id=1RrP5t1sJCrT2MO2I68pW62NYMfQ1zEgs

Animal husbandry

Sample preparation

Sample mounting

Perhabs useless

Resolution, contrast:

Staining

Animals, anesteshy: https://www.sciencedirect.com/science/article/pii/S1046202309001492

Image representation

Image noise and artifacts

Data analysis workflow

When data is reconstructed, what do voxels represent?

Typical data size

How are voxels represented (data types)?

How is it stored?

How is access organized?

Dimensionality? 2D? 3D? 4D?

How to treat such big datasets?

What the main artifacts and noise in tomography?

Ring artifacts

Sample movement

CNR and SNR

Why it occurs?

Overview of general workflow

Posture normalization

Localization and extraction

Segmentation

Morphometric analysis

Preprocessing

SR mu-CT for insects: https://www.annualreviews.org/doi/pdf/10.1146/annurev.physiol.70.113006.100434
Mice: https://www.annualreviews.org/doi/pdf/10.1146/annurev-bioeng-071910-124717
Bones: http://onlinelibrary.wiley.com/doi/10.1118/1.3697525/full

Мoreover, the time required to image a volume element or voxel with a certain statistical confidence increases drastically as the size of the voxel decreases. An object of smaller cross-section will absorb fewer photons and therefore requires longer exposure time to assure acceptable counting statistics. Consequently, increasing spatial resolutions require larger incident photon intensity or longer integration times. Because tube X-ray sources emit only a small fraction of their dissipated power as X-rays, obtaining high spatial resolution with these types of sources is often obtained at the cost of counting statistics and the ability to distinguish more subtle low-contrast features in an object. Synchrotron-based radiation on the other hand is well suited for high-resolution imaging because of the extremely high photon flux available. However, because it is difficult to produce energies above approximately 50 keV with synchrotron radiation sources, maximum sample size is generally limited to a few centimeters to assure that the beam can penetrate the sample, whereas larger samples can be examined in conventional systems that generally use higher energies. As a common rule, one can expect a spatial resolution on the order of 200–500 μm for medical CT systems, between 50 and 100 μm for industrial systems (no dose restrictions) designed to examine small samples (Kinney and Nichols, 1992), and from 50 μm down to approximately 1 μm for synchrotron based CT systems.

Streak artifacts

⚠ Firstly, the monochromatic beam yields reconstructed CT images free of beam hardening artifacts (Tafforeau et al 2006), and allows image calibration in order to evaluate linear attenuation coefficient distribution within the sample. Secondly, due to the small source dimensions and limited divergence of SR, spatial resolution in SR CT is limited mainly by the detector (Baruchel et al 2006). Thirdly, the high flux available at SR sources allows rapid CT data acquisition with high spatial resolution, resulting in precise mapping of sample internal structures (Trtik et al 2007). Fourthly, the high degree of (lateral) coherence of the SR sources allows implementation of phase-sensitive imaging techniques (Cloetens et al 1996).

https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7862905

http://iopscience.iop.org/article/10.1088/1361-6560/aab4b3/meta

2. Hierarchical analysis


  • a lot of segmentation approaches
  • mainly developed and applied for medical imaging

  • do not provide in-depth/hierarchial analysis of high-res data
  • they offer only limited applicability to high-res muCT data

  • leads to missing of potentially useful substructural information

  • medical CT data is very limited on spatial resolution
  • potentially useful substructural information is missed

1. High-throughtput analysis workflow

  • many automated analysis workflows exist
  • provide ad-hoc solutions for the specific cases
  • when the studied specimen is changed, the workflow is also changed
  • are not adapted for high-throughput experiments

4. New method for orientation analysis

  • existing methods are usually based on 2nd-order structure tensor
  • due to high computational burden have limited applicability to large datasets
  • either require downscale dataset or process specifid ROIs

6. Quanfima package

  • many orientation analysis methods were developed for analysis of 2D / 3D data
  • implementations of these methods are not easily available
  • written in different languages
  • cannot be integrated into analysis workflows

5. Workflow implementation

  • existing analysis workflows used in high-throughput experiments composed of different software packages
  • usage and improving of such workflow is complicated due to dependencies on third-party software packages

Design an automated analysis workflow for high-throughput micro-CT experimets

The development an efficient approach for hierarchial analysis of studied speciemens

Develop a rapid and reliable 2D/3D orientation analysis method applicable to large datasets of tens gigabytes

The eshaustive performance analysis of all stages of the proposed workflow on the manually segmented dataset provided by the expert

The accuracy and speed analysis of the proposed orientation estimation method with comparison to another widespread method

The implementation of all stages of the proposed workflow as independent modules using the state-of-the-art programming libraries

The implementation of fiber analysis module including all orientation methods involved into the evaluation

Develop the algorithm to generate a synthetic dataset composed of straight fibers for the performance evaluation of the fibrous analysis module

Apply the developed workflow to several real world application cases from the life and material sciences

3. 3D-to-2D PCA

  • the registration algorithms are broadly used for data alignment, when there is a reference dataset and the size of dataset is relatively small
  • thus, they hardly can be applied in cases very large datasets of hundreds gigabytes and the sample having specific geometric shape

Examples of structures