Emission tomography from compressed data

In emission tomographic modalities such as SPECT or PET, various regions of the detector space yield different amounts of information about the emission source. A framework for exploring the tradeoffs between binning the detector space is developed to reduce memory storage and the associated loss in performance for image reconstruction tasks and detection tasks. We use high rate vector quantization theory to establish just how much relevant information can be preserved after compression of the emission measurement data. We illustrate our results for one dimensional deconvolution and two dimensional lesion detection in PET.