[An Error Evaluation of Iterative Image Reconstruction Methods Using Chi-Square (χ²) Statistic Minimization for...

Area of Science:

• Nuclear Medicine Imaging
• Medical Physics
• Image Reconstruction Algorithms

Background:

• Iterative reconstruction (IR) methods in nuclear medicine commonly use Neyman's (χ²N) or Pearson's (χ²P) chi-square statistics.
• Mighell's chi-square statistic (χ²M) has been integrated into commercial SPECT systems, but its comparative accuracy requires evaluation.
• Previous studies lacked error assessment for χ²M, creating uncertainty regarding its superiority.

Purpose of the Study:

• To investigate and compare the accuracy of chi-square statistic-based iterative image reconstruction methods.
• To evaluate the performance of Mighell's chi-square statistic (χ²M) against Neyman's (χ²N) and Pearson's (χ²P) statistics.
• To assess noise propagation and image reconstruction accuracy using computer simulations.

Main Methods:

• Utilized two numerical phantoms (Phantom A and B) with varying count densities.
• Simulated Poisson noise in projection data, excluding attenuation and scatter effects.
• Employed the conjugate gradient method for minimizing chi-square statistics in iterative image reconstruction.
• Evaluated accuracy using root mean square error (RMSE) over multiple trials.

Main Results:

• All chi-square methods showed noise suppression when incorporating projection data variance.
• Pearson's chi-square (χ²P) demonstrated insufficient noise suppression compared to χ²N and χ²M.
• Mighell's chi-square (χ²M) consistently yielded lower RMSEs across different count densities and phantoms, indicating superior accuracy.

Conclusions:

• Mighell's chi-square statistic (χ²M) offers improved accuracy in iterative image reconstruction for nuclear medicine.
• χ²M demonstrates better noise handling and reduced root mean square error compared to χ²N and χ²P.
• The findings support the clinical utility of χ²M-based reconstruction, particularly for low-count projection data.