Effect of filtering on the detection and localization of small Ga-67 lesions in thoracic single photon emission computed tomography images.

Tumor detection can be significantly affected by filtering so determining an optimal filter is an important aspect of establishing a clinical reconstruction protocol. The purpose of this study was to identify the cut-off frequency of a Butterworth filter used in a filtered backprojection (FBP) reconstruction that maximized the detection and localization accuracy of 1 cm spherical lesions in Ga-67 citrate, thoracic SPECT images. Image quality was evaluated by means of a localization receiver operating characteristic (LROC) study using computer simulated images. Projection data were generated using the mathematical cardiac-torso digital phantom with a clinically realistic background source distribution. The images were reconstructed using FBP with multiplicative Chang attenuation correction and fifth-order Butterworth filtering. The cut-off frequencies considered were 0.25, 0.32, 0.47, and 0.79 cm(-1) for the case of three-dimensional (3D) post-filtering and 0.25, 0.32, and 0.47 cm(-1) for two-dimensional (2D) post-filtering. The images were read by three research scientists and one board certified nuclear medicine clinician. The area under the LROC curve and the localization accuracy for all test conditions were compared using Scheffé's multiple comparisons test. It was found that 3D post-filtering using filters with cut-off frequencies of 0.32 and 0.47 cm(-1) resulted in the highest lesion detectability and localization accuracy. These two test conditions did not differ significantly from each other but were significantly better (p<0.05) than all of the 2D, and the 3D 0.79 cm(-1) cut-off frequency cases.