A COUNT-DEPENDENT RECONSTRUCTION FILTER OPTIMIZED TO MATCH THE LESION SIZE AND THE OBJECT CONTRAST IN SPECT

Abstract

We present a new formulation, based on a physical model of SPECT lesion detectability, for a reconstruction filter optimized to improve lesion detectability. To determine the optimal parameters for the filter, we focused on improving the local detectability of the lesion corresponding to the SPECT detection limit, or equivalently increasing the local S/N ratio of the threshold detectable lesion. The threshold lesion was determined by the minimum detectable lesion size (MDLS) from the SPECT lesion detectability contrast-detail curve. The proposed filter showed the characteristics of resolution recovery and smoothing for MDLS in the resolution-limited and photon-limited regions, respectively, compensating for the relative impacts of the main limitation factors on threshold detectability. In simulations which varied physical factors such as photon counts, object contrast and detector system resolution, the proposed filter produced a considerable improvement in SPECT image duality in terms of lesion detectability.