Quantitative CT-less PET Imaging in the Presence of Prior Attenuation Coefficient Distribution

In this study, we explore quantitative PET imaging without CT as a solution to true low-dose PET scan. The proposed reconstruction algorithm is particularly suitable when prior attenuation information, i.e. previous CT data, is available. Gaussian mixture models (GMM) were used to define the distribution of attenuation coefficients within the patient body using the prior CT data. Incorporating the background radiation from lutetium in LSO as ‘transmission data’ along with predefined GMM, a so-called TOF-MLAA-GMM was developed. In the proposed algorithm, attenuation maps from LSO background radiation are employed to provide information about non-radioactive objects in the FOV (such as patient bed), create suitable initial conditions and scatter data, and minimize the scaling problem in TOF-MLAA. A NEMA NU-2-like phantom and ¹⁸ F-FDG patient data from a long-axial FOV scanner, Biograph Vision Quadra PET/CT (Siemens Healthineers), were used to evaluate the developed algorithm. For both the phantom and patient data, PET images from TOF-MLAA-GMM were compared against those from TOF-MLAA and OSEM using CT-based attenuation and scatter corrections. Studying both the phantom and patient PET images, it was seen that the developed TOF-MLAA-GMM algorithm outperforms TOF-MLAA. Analysis of the mean SUV across various organs of the patient showed a mean quantification error of -14.80% (range: -27.20 to -6.93%) in the PET images using TOF-MLAA and 10.64% (range: 2.20% to 14.55%) when using TOF-MLAA-GMM attenuation maps. The proposed method can therefore be used in low-dose PET studies where previous attenuation data are available to avoid repeated CT scans.