Computed Tomography (CT). Computed tomography, also known as computer
assisted tomography (CAT), is the name of an imaging system, where the
role of mathematics is evident. With the large number of collected data the
calculation of the desired image information can be achieved only by means of
powerful computers. Fundamental for the use of computers is the development
of efficient algorithms, based on a precise mathematical model of the complex
connection between measured data and image information to be determined.
The measured data, i.e. the attenuation of the X-ray intensity after travelling
through the body, is related to the X-ray attenuation coefficient, interpreted
mostly as the density of the tissue, along the path of the rays. In mathematical
terms, the observed attenuation is related to the line integral of the X-ray attenuation
coefficient along the ray path. The arising integral equation is, in the
2D case, named as Radon transform, after the Austrian mathematician Johann
Radon. In the first commercial scanners, Godfrey Hounsfield solved this integral
equation by standard discretization methods. He projected the solution on a
pixel basis, resulting in large, unstructured systems of linear equations that he
solved iteratively. Only in the mid-seventies the integral equation was actually
recognized as the Radon transform, for which Radon had derived an analytical
inversion formula already in 1917.
However, it has been a long way to go from his mathematical formula to
working algorithms, questions to be answered first concerned image resolution,
non-uniqueness for finitely many data, and optimization of parameters and stepsizes. In addition, a phenomenon typical for all imaging methods showed up:
These problems are so-called inverse problems, where unavoidable input data
errors, due to photon scattering, beam hardening, or miscounts in the detector,
are extremely amplified in the solution. In order to avoid this unwanted effect,
the problem has to be regularized, such that a balance between best possible
resolution in the image and maximal damping of the noise is obtained. In the