2086 - AI-Based Synthetic Simulation CT Generation from Diagnostic CT for Simulation-Free Radiotherapy Workflow in a Low Resource Setting

Purpose/Objective(s): CT simulation (sCT), which is time-consuming and requires patients to visit in person, is one of the major slowdowns in the radiotherapy (RT) workflow, particularly in a limited resource setting. Diagnostic CT (dCT) has been proposed to be used for treatment planning for an expedited workflow without the CT simulation process. However, differences in couch curvatures between dCT and sCT pose challenges in imaging-guided patient setup and introduce treatment planning uncertainties. This study demonstrates an innovative AI-based synthetic sCT generation from dCT for spine palliative RT patients, eliminating the need for sCT and enabling a faster treatment process.

Materials/Methods: The synthetic CT simulation (ssCT) was generated from two neural network models to predict patient anatomy on a flat couch and making patient geometry compatible for RT planning. The neural network models were trained using a dataset of 32 spine patients with dCT and sCT pair, with 22 cases allocated for training and 10 for testing. To assess the accuracy of the ssCT for treatment planning, the ssCT and dCT were registered to the sCT with the same frame of reference using rigid registration on bone windows. Then, the dose differences between dCT and sCT were compared with the values between the ssCT and sCT. Dmax, D95, D99, V100, and V107 and root-mean-square (RMS) differences were calculated to quantify deviations in the dose-volume histogram (DVH).

Results: The proposed method using synthetic simulation generated from dCT demonstrated significant improvements in accuracy compared to dCT for RT planning, using sCT as the reference. Key dosimetric metrics showed reduced differences for ssCT versus dCT. The metrics are shown in the following table.

Conclusion: By integrating a neural network for spine position estimation and an innovative curvature correction algorithm, our method significantly improves the accuracy of dCT-based treatment planning for dosimetric consistency with sCT. This approach has the potential to streamline the radiotherapy workflow, reduce treatment uncertainties, and accelerate treatment initiation.