Estimation of local statistical noise in PET images induced by attenuation inside the body

In positron emission tomography/computed tomography (PET/CT) examinations, the standardized uptake value (SUV) is a commonly used index to evaluate the activity of cancer. The precision of SUV is directly affected by the local statistical noise in PET images because SUV is calculated based on the counts on PET image data. The purpose of this study was to estimate the local statistical noise in the PET image caused by attenuation of annihilation photons inside the body.

To estimate the local statistical noise, we defined the number proportional to the standard deviation of a voxel caused by attenuation using CT image data. The conversion table between the number obtained from the CT image data and the actually measured standard deviation of a voxel was made by performing experiments with different forms or sizes of phantoms and scattering bodies. We compared the standard deviation of a voxel estimated by the method we proposed with actually measured ones.

The standard deviation of a voxel in the PET image varied according to the size of the phantom. The effect of attenuation on the standard deviation of a voxel was more significant in the 3D mode than in the 2D mode. The correlation coefficients between the estimated standard deviation and actually measured standard deviation were 0.98 and 0.99 in the 2D mode and 3D mode, respectively.

Our algorithm favorably estimated the local statistical noise in PET image data using CT image data, regardless of object size, although the consideration was limited for phantoms with homogeneous interiors.