The generation of MR-based synthetic CT using deep learning for brain radiotherapy

Abstract

This dissertation evaluates a conditional conditional Generative Adversarial Network (cGAN) algorithm for generating Synthetic Computed Tomography (sCT) images from Magnetic Resonance Imaging (MRI) data, aimed at enhancing Radiotherapy (RT) planning for brain cancer patients by testing a Maltese cohort of 16 patients. This research addresses the need for streamlined alternatives to dual-simulation processes involving both Computed Tomography (CT) and MRI, with the potential to reduce patient fatigue and improve treatment precision. The study’s primary objectives are to assess the algorithm’s accuracy in generating structurally and dosimetrically accurate sCT images, examine its effectiveness in replicating dose distributions, and mitigate registration errors that arise from dual modality simulation. This focus on a minority population, often underrepresented in existing literature, highlights the unique application of the algorithm in a Maltese context. An experimental methodology was employed, utilising quantitative evaluation metrics to test the algorithm on a local cohort. Image quality was assessed using Mean Absolute Error (MAE), Peak Signal-To-Noise Ratio (PSNR), and Structural Similarity Index Metric (SSIM), while dose accuracy was measured through Dose-Volume Histogram (DVH) and Gamma Pass Rate (GPR) metrics. Results indicate that the algorithm performs well in generating structurally accurate sCT images, particularly in replicating anatomical features. However, discrepancies in dose metrics were observed, underscoring the need for refinement in handling complex anatomical structures, such as the optic nerves and orbits. This research contributes to the advancement of MRI-based sCT generation, with implications for reducing registration errors and enhancing patient-specific RT planning. Future studies should focus on improving dose distribution accuracy, expanding the minority dataset and optimising the algorithm for broader clinical application.