Resampling estimates of precision in emission tomography

Area of Science:

• Medical Imaging
• Statistical Analysis
• Nuclear Medicine

Background:

• Emission tomography generates images from count data, inherently subject to Poisson noise.
• Accurate estimation of regional variance is crucial for reliable image interpretation and quantitative analysis.
• Existing methods may not fully capture the statistical uncertainties arising from raw data.

Purpose of the Study:

• To present and validate statistical resampling methods for estimating regional variance in emission tomography.
• To adapt these methods for time-of-flight PET and demonstrate their applicability to non-time-of-flight PET and SPECT.
• To provide computationally feasible and broadly applicable variance estimation techniques.

Main Methods:

• Application of bootstrap and jackknife resampling techniques to emission tomography data.
• Implementation of bootstrap methods in time-of-flight PET (positron emission tomography).
• Validation of estimates by comparison with emission scan repetitions.

Main Results:

• The bootstrap and jackknife methods provide reliable estimates of regional variance.
• The proposed techniques are applicable to various emission tomography modalities (PET, SPECT) and reconstruction methods.
• Simple expressions for the accuracy of the variance estimates are derived.

Conclusions:

• Statistical resampling offers a computationally feasible and accurate approach to regional variance estimation in emission tomography.
• These methods enhance the reliability of quantitative analysis in PET and SPECT imaging.
• The techniques are versatile, applicable to raw, uncorrected emission tomography data.