TITLE: Evaluation of Quantitative Image Reconstruction Methods for Maximum Lesion Detection in In-111 ProstaScint® Prostate SPECT

 

 

AUTHORS: Tsui, Benjamin M.W. 1; Chen, Si 1; Liu, Chi 1; Volokh, Lana 2

 

___________________________________________________________________

ABSTRACT

 

PURPOSE:

 The objective is to evaluate different quantitative image reconstruction methods for maximum lesion detection in In-111 ProstaScient® prostate SPECT using the channelized Hotelling observer (CHO) and ROC analysis methodology.

METHODS AND MATERIALS:

 

The computer generated 3D NCAT phantom which realistically models the anatomical structure and attenuation distribution of a typical patient was used in the study. Three distribution of uptake ratios in different organs in the pelvis region, two different elliptically-shaped lesion sizes (10x10x15 and 12x12x17 mm3) and two different lesion-to-background contrasts (5:1 and 8:1) were used to simulate variations found in a patient population.


A GE VG Millennium SPECT system with a GE
medium energy general purpose (MEGP) collimator was used in data acquisition. The SimSET Monte Carlo code for simulating photon transports within the 3D NCAT phantom and an efficient collimator detector response generator for simulating the MEGP collimator were used to generate "noise-free" projection data with 120 128x128 projections over 360o for each of the NCAT phantom variations. They are then scaled to two detected count levels and Poisson noise added to each to create eight separate datasets with different noise realizations. 

 

The projections datasets were reconstructed using OS-EM image reconstruction methods with 4 subsets and with different compensations: (1) no compensation, (2) with attenuation compensation (AC), (3) with AC and collimator-detector response (DC) compensation and (4) with AC, DC and scatter compensation (SC).  The CHO and ROC methodology were used to determine the optimum iteration numbers for the OS-EM methods, cutoff frequencies for the Butterworth postfiltering, and the OS-EM reconstruction methods for maximum lesion detectability in terms of maximum area under the ROC curve (AUC). 

 

RESULTS:

The CHO and ROC study showed the optimum iteration number at 1, 1, 3 and 4 and optimum Butterworth postfilter cutoff frequency at 0.08, 0.08, 0.1 and 0.1 pixel-1 for the OS-EM methods (1), (2), (3) and (4), respectively.  When these optimum iteration numbers and cutoff frequencies were used, the AUC values for the OS-EM methods (1), (2), (3) and (4) are 0.87, 0.81, 0.92 and 0.93, respectively.  Statistically significant differences are found between all AUC curves except for those between the reconstruction methods (3) and (4).

 

 

 

 

 

 

 

 

 

 

 

CONCLUSIONS:

 

 

 

 

 

FUNDING SOURCES: