Regularization Methods for Tomographic Image Reconstruction

Regents Of The University Of Michigan - Ann Arbor, Ann Arbor MI

Investigators

Abstract

9982349 Fessler Tomographic imaging is used in a variety of scientific, industrial, national security, and medical applications. Since tomographic imaging problems generally are ill-posed, the reconstructed images can be severely degraded by measurement noise, unless one employs adequate regularization methods. Conventional quadratic regularization methods have the undesirable property of producing nonuniform and anisotropic spatial resolution properties when applied to photon-limited emission computed tomography (SPECT). These properties cause noticeable distortion of object shape in reconstructed images, such as elongation of tumors in PET scans, and can cause nonuniform contrast in thin structures such as the myocardium in cardiac SPECT scans. These deficiencies may impact the diagnostic accuracy of PET and SPECT imaging. The goals of this research project are to develop, analyze, and evaluate improved regularization methods of tomographic image reconstruction, particularly in the context of PET and SPECT imaging. Current commercially available iterative image reconstruction algorithms for SPECT and PET are all based on unregularized methods. The methods developed in this project will improve the suitability of regularized methods for routine use in tomographic imaging by providing uniform spatial resolution where appropriate, and by improving the bias-variance tradeoffs of regularized image reconstruction methods.