Tomographic reconstruction requires availability of sufficient number of measurements of the underlying data. However, multiple constraints like dosage of permitted radiation, time allocated for sampling and limitation on movement of data puts a constraint on the sampling density. The challenge is then to recover detailed anatomy of the data under measurement deficient conditions. Towards this, powerful iterative techniques like those in Compressive Sensing (CS) exploit the sparsity of data under some mathematical transform. In this talk, I will cover two techniques that I have been working on as part of my PhD thesis, to further improve the accuracy of reconstruction within the CS framework.
The first technique aims to “pool” together measurements from similar slices (cross-sections of a 3D volume) to increasing the data sparsity. This involves identifying the degree of similarity among slices directly from their measurements. This is followed by joint reconstruction of only the similar slices together. The second technique uses external information from prior data that may be available in the form of already reconstructed structurally similar slices. I will be discussing the use of a global prior to improve both the speed and quality of tomographic reconstruction. Specifically, a set of potential representative 2D slices, referred to as templates, is chosen to build an eigenspace, which is then used to guide the iterative reconstruction of a similar slice from its sparse acquisition data.