Evaluation of task-oriented performance of several fully 3D PET reconstruction algorithms

Area of Science:

• Medical Imaging
• Image Reconstruction
• Positron Emission Tomography (PET)

Background:

• Accurate image reconstruction is crucial for quantitative analysis in 3D PET imaging.
• Iterative reconstruction algorithms offer potential advantages over traditional filtered backprojection (FBP).
• The choice of basis functions (voxels vs. blobs) can impact reconstruction performance.

Purpose of the Study:

• To compare the performance of five 3D PET reconstruction algorithms.
• To evaluate the efficacy of using 'blobs' versus voxels as basis functions in iterative algorithms.
• To assess algorithm performance for specific medical tasks like activity estimation and lesion detection.

Main Methods:

• Evaluated filtered backprojection (FBP), EM-ML, and ART iterative methods.
• Compared voxel-based and blob-based basis functions for iterative algorithms.
• Utilized randomly generated phantoms and projection data for training and comparative evaluation.
• Determined algorithm parameters by maximizing a technical figure of merit.

Main Results:

• Using blobs as basis functions in iterative algorithms demonstrated statistically significant advantages over voxels.
• No clear overall performance difference was found between FBP, EM-ML with blobs, and ART with blobs.
• ART with blobs suggested potentially the highest efficacy among the compared methods.

Conclusions:

• Blob-based basis functions are superior for iterative 3D PET reconstruction.
• Further investigation is needed to differentiate the performance of leading iterative algorithms like ART and EM-ML with blobs.
• The study provides a statistically rigorous framework for evaluating PET reconstruction algorithms.