Information gain from count corrections in SPECT image reconstruction and classification

The authors present a method for quantifying the impact of count correction side information on the capability of a SPECT (single-photon-emission computed tomography) projective tomography system to perform image (emitter) reconstruction and feature classification. The method involves computing the image or feature related information gain which results from the presence of count correction side information at the detector. For image reconstruction this information gain is computed using Shannon´s mutual information, while for feature classification the cutoff rate of the SPECT information channel is used. For reconstruction the gain is proportional to the information divergence between the spatially dependent probability of deletion of a γ-ray originating at a particular emitter location and the spatially independent average deletion probability. For classification the gain is proportional to the difference between the arithmetic mean and the geometric mean of the average number of γ-ray deletions for each of the image classes. Results of analysis and numerical study are presented which indicate that the information gain associated with using count correction data is much more significant for reconstruction of the emitters than for classification of the emitter density when the total detected fluence is low