PhD student Daniël Pelt of CWI has developed new algorithms for computer tomography (CT) for fast computation of high-quality three-dimensional images. This increases the practical use of this technology in areas such as material sciences and biomedical research. Pelt defends his thesis ‘Filter-based reconstruction methods for tomography’ on Tuesday 3 May at Leiden University.
In many applications, it is useful to look inside an object without destroying it, such as medical examination or product quality assessment in industry. This is possible with a CT-scan, which involves a radiation source and detector which are rotated around the object. This results in several X-ray images with varying angles. A mathematical algorithm uses these two-dimensional images (projections) to create a three-dimensional image of the object’s internal structure. CT is often used in medical applications, but also in material sciences, palaeontology, biomedical research and assessing and optimizing industrial production.
An important aspect of every application is the mathematical reconstruction algorithm that computes the three-dimensional image using the projections. There are roughly two types of algorithms: analytical and algebraic. Analytical algorithms transform the two-dimensional images using a filter, allowing for faster computation. An important disadvantage of this method is errors in the case of a small number of projections, or in noisy conditions, both of which are very common in modern advanced tomography applications. Algebraic algorithms are better at computing images in such conditions by incorporating prior knowledge of the object and scanning equipment. Main disadvantage is that they are very slow, often to the point of being unusable for most applications.
In his thesis, Pelt develops new reconstruction algorithms that combine the benefits of analytical and algebraic algorithms, so-called filter-based algorithms. The speed of these new algorithms is comparable to the analytical algorithms, and the image quality to the algebraic algorithms. This is possible by using analytical algorithms with a filter based on characteristics such as the scanning set-up or the properties of the scanned object. Filters can also be determined by a neural network which has learned to recognize the best filter for specific applications. Ultimate goal of this research line is to produce real-time high-quality 3D images that can be viewed during scanning.
This research is funded by the 2011 NWO Vidi grant of Prof. Joost Batenburg.