In 1981, Elliott and Dover designed an X-ray microtomography scanner as a means of measuring the local mineral
concentration in teeth. Although slow, this first generation system gave accurate measurements of the X-ray linear
attenuation coefficient (LAC) due to its use of energy dispersive photon counting apparatus. Attaining such accuracy
with integrating detectors in third generation scanners is difficult, but has been the goal of our ongoing development.
The current "MuCat 2" system uses a 6cm square CCD chip with a parallel fibre-optic faceplate coupled to a CsI
scintillator. Time delay integration readout (with sliding camera) is used to eliminate ring artefacts and enable high
dynamic range X-ray projections to be acquired. The beam is collimated with a moving aperture (tracking the camera) to
reduce X-ray scatter. Beam hardening is reduced by the use of filtering and corrected using data from an aluminium step
wedge to optimise a model of polychromatic X-ray generation, attenuation and detection. Adjustments can be made to
the model to allow for known specimen composition. Projections are corrected for distortion and repeatable wobble in
the rotation stage. Where high absolute accuracy of the LAC is required, a pure aluminium wire is included in the scan
and used to "fine-tune" the grey level after reconstruction.
Beam hardening in X-ray tomography is often corrected using an arbitrary polynomial whose coefficients are subjectively selected. A better approach is to model X-ray generation, transmission and detection and to use step wedge transmission measurements to fit the model parameters. This allows for extrapolation of linearization curves beyond the range of the step wedge and it allows this curve to be adjusted according to the specimen composition without changing
the composition of the step wedge. This paper presents the principles behind beam-hardening and the model used for correction. Initial tests of this method have shown very good results where a priori knowledge of the specimen composition is available.
The way in which microtomography developed in the authors' laboratory in the early 1980s is described, together with some background material. Later developments in scanning geometries and detectors, mainly in other laboratories, are described. Some present problems and possible future directions will be considered.
Optical coupling between the X-ray scintillator and digital camera (typically CCD) is a major design consideration in X-ray microtomography. Previously, we used a pair of 50mm f1.2 lenses, which we determined to be approximately 60 % efficient, that is, the signal to noise ratio is that which would occur if 60 % of the X-ray photons absorbed by the scintillator were directly detected. For larger CCDs, lenses become excessively large, heavy and expensive. For our 60 x 60 mm time-delay integration CCD camera, we used parallel fibre-optic coupling, giving greater efficiency. A problem with this is the scattering of light through the fibre cladding, which reduces image contrast, adding a very blurred image to the sharp image transmitted through the fibres. This problem is ideally suited to solution by deconvolution. Since the high frequency image components are present (direct fibre image) deconvolution can be used to eliminate the low frequency scatter image, without the problems normally associated with de-blurring. The point spread function was assumed to be rotationally symmetrical and was determined from an edge image of a lead plate positioned close to the scintillator. In frequency space, the mid frequency portion was extrapolated into the low frequency portion using a parabolic fit. The difference between the extrapolated and measured low frequency portions was deemed to be the scatter response. This was then added to the frequency response for a perfect delta function to obtain the frequency response used for deconvolution. The results showed excellent correction of the X-ray microtomographic images.