Pipeline development for realistic synthetic database generation of SPECT neuroimaging

Abstract

This Final Degree Project introduces a simulation pipeline for generating realistic brain SPECT images using Monte Carlo methods. Built on real anatomical MRI and CT data from patients with diagnosed or suspected Parkinson’s disease, the pipeline creates synthetic images by first generating activity and attenuation maps that replicate radiotracer distribution and tissue densities. These are then used as inputs for the SimSET simulation engine to produce synthetic SPECT projections. A core feature of the pipeline is its iterative framework. Simulated reconstructions are compared to real clinical SPECT images using anatomical atlases to quantify regional differences. These discrepancies inform successive updates to the activity map, gradually refining image realism across iterations. The approach enables the generation of synthetic SPECT studies that closely resemble real data, while maintaining known ground truth—key for evaluating and validating quantification methods in nuclear medicine. The pipeline is modular, reproducible, and scalable, with integrated quality controls and standardized preprocessing steps. It lays the groundwork for creating a database of synthetic realistic neuroimaging studies. The project contributes to the advancement of validation for quantification imaging tools, particularly for Parkinson’s disease research and clinical validation.