Quantitative comparison of automated PET volume delineation methodologies using simulated tumor lesions

Robust tumor activity quantification recently finds application in challenging medical scenarios like early therapy response detection, radiotherapy treatment planning, etc. This paper targets a quantitative comparison of existing state of the art Positron Emission Tomography (PET) volume delineation methodologies. The different methods evaluated include adaptive threshold based, gradient based and stochastic image segmentations. For that purpose, spherical and non-spherical tumor lesions were simulated and studied. PET images were reconstructed with Maximum Likelihood Expectation-Maximization (MLEM) and Maximum A Posteriori (MAP) algorithms. All schemes were evaluated with reference to the ground truth knowledge. The spherical lesions were best segmented with the adaptive threshold based method, whereas the stochastic methods were slightly better for the non-spherical lesion.