Averaging and Metropolis iterations for positron emission tomography.

Area of Science:

• Medical Imaging
• Computational Science
• Nuclear Medicine

Background:

• Iterative positron emission tomography (PET) reconstruction involves complex multi-dimensional integrals.
• Monte Carlo (MC) quadrature approximates these integrals but requires on-the-fly computation during reconstruction.
• High dimensions and large datasets in PET lead to significant approximation errors in standard ML-EM schemes.

Purpose of the Study:

• To reduce reconstruction error in PET imaging caused by on-the-fly MC approximations.
• To enhance the accuracy and stability of iterative reconstruction solutions.
• To decrease the computational time and required number of MC samples.

Main Methods:

• Modified Maximum Likelihood Expectation Maximization (ML-EM) iteration schemes.
• Analysis of error behavior with on-the-fly MC projections.
• Development of averaging iteration and Metropolis iteration techniques.

Main Results:

• Averaging iteration averages forward projection estimates to stabilize results.
• Metropolis iteration selectively rejects estimates to ensure unbiased tracer density and improve stability.
• Both methods significantly reduce the number of required MC samples and reconstruction time.

Conclusions:

• Averaging and Metropolis iterations effectively mitigate errors from on-the-fly MC approximations in PET reconstruction.
• These methods enhance accuracy and stability while reducing computational demands.
• The proposed techniques are integrated into the Teratomo system for practical application.