Nuclear image sequence decomposition

This research concerns the problem of dis criminating between various radioactive objects on the basis of their differing dynamics (photon intensity as a function of time) in a sequence of images. Each object is assumed to be homogeneous dynamically, but to differ in intensity from point to point. Since the objects of interest may overlap and are transparent to the radiation, the dynamics at each pixel will be a linear combination of the dynamics of the pure objects. The dynamics of the pure objects are in general unknown and must be inferred from the data. Since the number of pure dynamics is generally less than the number of images in the sequence, eigen vector analysis was used to define a sub-space containing the data (except for sampling error). Techniques for inferring the pure dynamics have been applied in this sub-space. On the basis of the pure dynamics, the portions of the images in which the objects overlap have been decomposed, so that separate images of each object as a function of time could be obtained. These techniques have been applied in the field of Nuclear Medicine, where the images of various organs frequently overlap. Each organ generally differs in its dynamics of uptake and elimination of a radioactive substance which is Injected into the blood stream. For example, image sequences were successfully decomposed into contributions due to the liver, the gallbladder, the blood, and extravascular tissue. The dynamics of each of these types of tissue can then be analyzed for normality or to evaluate various disease states. These same techniques could be used to decompose other image sequences, such as satellite images of mixed vegetation, into their component contributions. They could also be used In cases where the index on the set of images is not time, but is some other feature such as frequency spec tral component, photon energy, or color, or in cases where there are several indices.