Abstract
1701
Objectives: The goal was to find the best algorithm to discriminate diffuse heterogeneous perfusion levels without being influenced by focal defects. 7 methods were compared on a set of 210 brain SPECT simulations and on a set of 40 patients. The 7 methods were: Coefficient of Variation (CV), Entropy (E1), local Entropy (E2), Fractal Dimension [FD1 (box counting), FD2 (Fourier Power Spectrum)], Gray Level Co-occurrence Matrix (GLCM), Random Walk (RW). Methods: First, the 7 algorithms were tested on numerical phantoms. The numerical anthropomorphic Zubal head phantom was used to generate 42 (6×7) different brain SPECT. 7 diffuse cortical heterogeneity levels were simulated with an adjustable Gaussian function, as well as 6 focal temporo-parietal perfusion defects. For each simulation, 5 noise realizations were performed yielding a total of 210 datasets (42×5). For a given method, the influence of diffuse heterogeneity/focal defect was assessed by slope comparison of linear regression at a given focal defect/heterogeneity level. The heterogeneity discrimination power was made by slope comparison of mean regression lines for each method. Secondly, the 7 algorithms were run on 40 HMPAO SPECT performed on patients referred for memory impairment. Scans were blindly ranked by 2 physicians only regarding the degree of heterogeneity. Results: The slopes (ie discrimination power) for diffuse heterogeneity / focal defects were : (0.58; 0.01) for GLCM, (0.36; 0.001) for FD1, (0.31; 0.2) for RW, (0.14; 0.38) for CV; (0.07; 0.014) for E2; (0.019; 0.019) for FD2;(0.017; 0.073) for E1. The correlation coefficients between physicians ranking and methods were 0.826 for RW (p<0.0001), -0.55 for CV, 0.09 for E2, -0.09 for GLCM, -0.12 for E1, -0.26 for FD1 and -0.46 for FD2. Conclusions: The GLCM method discriminates the more of diffuse heterogeneity levels in numerical phantoms without being influenced by focal cortical defects. However, GLCM classification is not correlated to physicians ranking. Only RW presented a significant correlation with physicians ranking and a relatively good ability to classify heterogeneity.
- Society of Nuclear Medicine, Inc.