Photon counting detectors used in spectroscopic CT are often based on small pixels and therefore offer only limited space to include energy discriminators and their associated counters in each pixel cell. For this reason, it is important to make efficient use of the available energy discriminators in order to achieve an optimized material contrast at a radiation dose as low as possible. Unfortunately, the complexity of evaluating every possible combination of energy thresholds, given a fixed number of counters, rapidly increases with the resolution at which this search is performed, and makes brute-force approaches to this problem infeasible. In this work, we introduce methods from machine learning, in particular sparse regression, to perform a feature selection to determine optimal combinations of energy thresholds. We will demonstrate how methods enforcing row-sparsity on a linear regression’s coefficient matrix can be applied to the multiple response problem in spectroscopic CT, i.e. the case in which a single set of energy thresholds is sought to simultaneously retrieve concentrations pertaining to a multitude of materials in an optimal way. These methods are applied to CT images experimentally obtained with a Medipix3RX detector operated in charge summing mode and with a CdTe sensor at a pixel pitch of 110μm. We show that the least absolute shrinkage and selection operator (lasso), generalized to the multiple response case, chooses four out of 20 possible threshold positions that allow discriminating PMMA, iodine and gadolinium in a contrast agent phantom at a higher accuracy than with equally spaced thresholds. Finally, we illustrate why it might be unwise to use a higher number of energy thresholds than absolutely necessary.

Date Published: 19 March 2014
PDF: 12 pages
Proc. SPIE 9033, Medical Imaging 2014: Physics of Medical Imaging, 90331A (19 March 2014); doi: 10.1117/12.2043400

Published in SPIE Proceedings Vol. 9033:
Medical Imaging 2014: Physics of Medical Imaging
Bruce R. Whiting; Christoph Hoeschen, Editor(s)