Exploiting anatomical information for PET image enhancement: A phantom experiment for algorithm validation

Multimodality imaging techniques (PET/CT and PET/MR) offer the opportunity of integrating functional and anatomical information to improve clinical diagnostic accuracy. High-resolution PET images can be obtained by exploiting structural information within- and post-image reconstruction. Although PET/CT is more utilised than PET/MR in routine clinical practice, the majority of multimodal technologies are validated either on simulations or clinically acquired PET/MR data. This work describes a PET/CT phantom experiment that provides realistic data for the validation of anatomy-based algorithms in a clinical setting. We performed a PET/CT phantom acquisition combining PET radiotracer concentration and CT Contrast Media (CM) to obtain images with contrast levels similar to clinical [18F]Fluoride bone scans. We performed three acquisitions to cover a wider range of possible clinical situations and evaluate the performance of image enhancement algorithms. On the one hand the CT was used as a prior to regularized the PET image reconstruction, on the other it was integrated with the functional data into a post reconstruction resolution recovery algorithm. Through the analysis of the CT acquisition we described the correlation between CM concentration and CT image contrast. We also quantified a 10-20% difference on the recovered PET radioactivity due to an incorrect attenuation correction. Furthermore, we were able to quantify the accuracy of the true activity estimation when integrating anatomical and functional information. Specifically, the improvement was of 9% when CT images were used as prior within the reconstruction and of 12% when used for resolution recovery in post reconstruction. This experimental procedure aimed to obtain PET/CT contrast similar to that of a patient acquisition. The results reported can be used to reproduce experiments mimicking a wider range of clinical studies and provide a solid ground truth for the validation of image enhancement algorithms based on the integration of anatomical and functional images.