A filter design for optimization of lesion detection in SPECT

Abstract

Recently, we proposed a filter formulation for effective lesion detection in an attempt to increase the local S/N ratio of a lesion corresponding to the SPECT lesion detection limit. The filter function is composed of (1) an exactly-derived theoretical filter (that is a multiplication of a GAUSS filter whose resolution is determined by the estimated minimum detectable lesion size (MDLS) and an inverse detector system MTF), and (2) a designed Butterworth filter for controlling the extent to which the theoretical filter is followed before our filter switches to noise suppression. In this work, a frequency cut-off design by the Butterworth filter was made as a function of (MDLS/system resolution) ratio, with respect to a MDLS-dependent frequency point that was determined based on a simple object model. The simulation studies demonstrated better performance of our filter on lesion detectability over the conventional SPECT filters. Especially with this Butterworth filter design, we considerably reduced the potential hazards of our filter, that were observed in large system resolution cases as a form of structured noise for high-count and high-contrast cases, and as a form of excessive smoothing of image contrast for low count and low-contrast cases.