Detection of Lesions in Positron Emission Tomography

Abstract

Observer studies are often used to compare the quality of Positron Emission Tomography (PET) images. One of the purposes of observer studies is to assess whether a lesion is detectable in an image or not (i.e., determine the detection threshold). However, observer studies are time consuming and the results may depend on the specific observer. Computer models may also be used to predict the detection threshold. The purpose of this study was to develop a novel method to quantify the detection threshold of a lesion, based on the distribution of the signal in the background. PET acquisitions of a phantom with six spheres were used. The distribution of the signal in the background was assessed by looking at the histogram of the contrast-to-noise ratio (CNR) in the background. The shape of the histogram was fitted and normalized to obtain a probability density function. This was tried with both a gamma distribution and a skew normal distribution. This probability density function was used to determine whether a fixed probability could be calculated that would predict the detection threshold. The results showed that for the two smallest spheres, a fixed threshold probability could be found when fitting a gamma distribution. The skew normal distribution fit was subject to numerical errors and therefore a slightly different approach was tried, in which the CNR corresponding to a fixed probabilty p=0.0001 was compared to the reference values. A linear relationship was found, which suggested that it may be possible to express the detection threshold in terms of a fixed probability, although this study did not succeed in quantifying this probability. In conclusion, the novel method that was investigated in this study did not result in an accurate quantification of the detection threshold. Improvements can be made in relation to numerical errors, the choice of probability density function, the method for determining the threshold scans and the measure used to describe how well the human eye can detect lesions.