Restoration of nuclear medicine images

Abstract

Restoration of single photon emission computed tomography (SPECT) images may be performed either on projection images (pre-reconstruction restoration), on two-dimensional (2-D) SPECT slices (post-reconstruction restoration), or on three-dimensional (3-D) SPECT blocks of images. This thesis presents a comparison among these different techniques and discusses the applicability of each in the restoration of various categories of SPECT images. The use of the shift-invariant Wiener, Power Spectrum Equalization (PSE), and Metz filters in 2-D and 3-D; pre-processing of projection images by geometric averaging to reduce the shift-variance of the blurring function; as well as a 2-D shift-variant implementation of the Kalman filter in restoration of SPECT images is investigated. The filters were used to restore SPECT images of two cylindrical phantoms, a cardiac phantom, a resolution phantom, and a truncated cone phantom containing two types of cold spots, a sphere, and a triangular prism. The filters were also applied to clinical images of the thyroid, heart, brain, and liver and spleen. The images were acquired on a Siemens Rota camera and an ADAC GENESYS camera. The 3-D Wiener, PSE, and Metz filter and the 2-D Kalman filter were seen to perform better than the other 2-D methods because the 3-D filter takes into consideration the interslice information in the filtering process and the Kalman filter uses a shift-variant blur function and assumes only local stationary of the noise. Quantitative analysis of the restored images performed through measurement of root mean squared (RMS) errors and contrast ratios showed a considerable reduction in error and increase in contrast of the 3-D Wiener, PSE, and Metz filter and the 2-D Kalman filtered images over images restored using other 2-D methods. However, 2-D methods are faster and showed comparable performances for small images. The thesis adds three new restoration methods to those already published in the literature for restoration of SPECT images, which perform better than previously reported methods, especially when the object being imaged is large. An entire suite of restoration methods has been developed to cater to the varying situations in different clinical applications of SPECT. The restoration provided by the methods should aid in improving the quality of clinical SPECT images, and thereby improve the efficiency of clinical diagnosis using SPECT.