Towards adapting a normal patient database for SPECT brain perfusion imaging

Single-photon emission computerized tomography (SPECT) is a tool which can be used to image perfusion in the brain. Clinicians can use such images to help diagnose dementias such as Alzheimer's disease. Due to the intrinsic stochasticity in the photon imaging system, some form of statistical comparison of an individual image with a 'normal' patient database gives a clinician additional confidence in interpreting the image. Due to the variations between SPECT camera systems, ideally a normal patient database is required for each individual system. However, cost or ethical considerations often prohibit the collection of such a database for each new camera system. Some method of adapting existing normal patient databases to new camera systems would be beneficial. This paper introduces a method which may be regarded as a 'first-pass' attempt based on 2-norm regularization and a codebook of discrete spatially stationary convolutional kernels. Some preliminary illustrative results are presented, together with discussion on limitations and possible improvements.