Bootstrap methods for estimating PET image noise: experimental validation and an application to evaluation of image...

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Image Processing

Background:

• Accurate estimation of regional PET image noise is crucial for optimizing image processing and reconstruction.
• The bootstrap is a powerful data-driven simulation method for statistical inference, enabling noise estimation without requiring repeated measurements.

Purpose of the Study:

• To experimentally validate bootstrap-based methods for estimating PET image noise.
• To demonstrate the utility of bootstrap methods in evaluating the performance of different PET image reconstruction algorithms.

Main Methods:

• Two bootstrap methods, list-mode data bootstrap (LMBS) and sinogram bootstrap (SNBS), were implemented on a clinical PET scanner.
• A reference standard deviation (SD) map was generated from 60 independent measurements of a phantom, and bootstrap SD maps were calculated from 60 replicates of a single measurement.
• Brain (18)F-FDG data were analyzed, and three reconstruction algorithms (FBP 2D, DRAMA 2D, DRAMA 3D) were assessed.

Main Results:

• Bootstrap SD maps showed strong agreement with the reference SD map across all tested reconstruction algorithms, validating the bootstrap methods.
• The LMBS and SNBS methods demonstrated equivalent performance in estimating PET image noise.
• Bootstrap analysis of FDG data revealed a superior contrast-noise relationship for DRAMA 3D compared to DRAMA 2D and FBP 2D.

Conclusions:

• Bootstrap methods provide accurate PET image noise estimates for various reconstruction algorithms using only a single scan.
• These methods eliminate the need for repeated measurements, making them practical for human PET studies.
• The validated bootstrap approach aids in the selection and optimization of PET image reconstruction techniques.