Dynamic regional cerebral blood flow SPECT processing using the Karhunen-Loeve transform for the reduction of noise and computational load

The Karhunen-Loeve (K-L) transform can be used to achieve a reduction in noise and computational load for dynamic acquisitions. This study evaluated the use of the K-L transform in processing Xe-133 dynamic brain SPECT images using both simulated and clinical data. A simple 3D dynamic brain phantom, consisting of five different regions that represents high and low regional cerebral blood flow (RCBF) embedded in a low flow background were used in the simulation study. An analytical 3D projector was used to realistically simulate 24 time sequences over a 6 minute period (15 sec/projection set) to represent dynamic brain blood flow. Poisson noise was added. Two implementations of the K-L transform were investigated. Method 1 consisted of selecting the first two principle components in K-L space before inverse transformation and conventional OSEM reconstruction. Method 2 consisted of retaining the first three principle components in K-L space, reconstructing using MAP AB-EMML with Gibbs regularization. Three patients referred to the clinic undergone routine Xe-133 brain blood flow SPECT studies. In comparison with conventional processing K-L Method 1 gives similar RCBF values in all the regions while a bias was introduced with K-L Method 2. Noise reduction and an order of magnitude decrease in computational load were obtained. In patient data, K-L Method 1 showed similar results as the conventional method for high blood flow values and have improved contrast of low blood flow areas