Evaluation of Deep Learning-based Scatter Correction on a Long-axial Field-of-view PET scanner
Objective: Long-axial field-of-view (LAFOV) positron emission tomography(PET) systems allow higher sensitivity, with an increased number of detectedlines of response induced by a larger angle of acceptance. However, thisextended angle increases the number of multiple scatters and the scattercontribution within oblique planes. As scattering affects both quality andquantification of the reconstructed image, it is crucial to correct this effectwith more accurate methods than the state-of-the-art single scatter simulation(SSS) that can reach its limits with such an extended field-of-view (FOV). Inthis work, which is an extension of our previous assessment of deeplearning-based scatter estimation (DLSE) carried out on a conventional PETsystem, we aim to evaluate the DLSE method performance on LAFOV total-body PET. Approach: The proposed DLSE method based on a convolutional neural network(CNN) U-Net architecture uses emission and attenuation sinograms to estimatescatter sinogram. The network was trained from Monte-Carlo (MC) simulations ofXCAT phantoms [18F]-FDG PET acquisitions using a Siemens Biograph Vision Quadrascanner model, with multiple morphologies and dose distributions. We firstlyevaluated the method performance on simulated data in both sinogram and imagedomain by comparing it to the MC ground truth and SSS scatter sinograms. Wethen tested the method on seven [18F]-FDG and seven [18F]-PSMA clinicaldatasets, and compare it to SSS estimations. Results: DLSE showed superior accuracy on phantom data, greater robustness topatient size and dose variations compared to SSS, and better lesion contrastrecovery. It also yielded promising clinical results, improving lesioncontrasts in [18F]-FDG datasets and performing consistently with [18F]-PSMAdatasets despite no training with [18F]-PSMA.