Comparison of heterogeneity quantification algorithms for brain SPECT perfusion images

Area of Science:

• Neuroimaging
• Medical Physics
• Quantitative Analysis

Background:

• Brain perfusion heterogeneity quantification is crucial for diagnosing neurological disorders.
• Existing algorithms require rigorous comparison for clinical utility.
• Single Photon Emission Computed Tomography (SPECT) is a key imaging modality for this assessment.

Purpose of the Study:

• To compare the performance of several algorithms against the original random walk (RW) algorithm for brain perfusion heterogeneity quantification.
• To evaluate the ability of these algorithms to differentiate various levels of heterogeneity in simulated and real patient data.
• To assess the influence of focal defects on algorithm performance.

Main Methods:

• Five algorithms (coefficient of variation, entropy, fractal dimension, GLCM, RW) were tested on 210 brain SPECT simulations using the Zubal head phantom with varying diffuse and focal heterogeneity levels.
• SPECT images were reconstructed using filtered back projection.
• Algorithm performance was evaluated by their ability to discriminate heterogeneity levels (linear regression slope) and by comparing their rankings to physician consensus in 40 patient SPECT exams.

Main Results:

• The GLCM method (slope=58.5) and fractal dimension (35.9) demonstrated superior differentiation of diffuse heterogeneity compared to the RW method (31.6).
• The GLCM method was unaffected by focal defects, unlike fractal dimension and RW methods.
• The RW method showed a significant correlation with physician classification (r=0.86), but GLCM did not (Rho=-0.099).

Conclusions:

• The GLCM method accurately quantifies diffuse brain SPECT heterogeneity, unaffected by focal defects.
• The RW method aligns with physician perception of heterogeneity but is sensitive to focal defects.
• Algorithm selection depends on the specific clinical application and the presence of focal abnormalities.