Development of Virtual 4DCT for Image Guided Proton Therapy

This study reports development of virtual 4D-CT workflow that can be used during image guided proton therapy to evaluate patient anatomy changes and if motion mitigation with abdominal compression has been properly setup for liver and lung tumor patients treated with pencil beam scanning proton therapy (PBSPT). A total of ten (5 lung and 5 liver) patients were studied with one planning 4D-CT (p4D-CT) and 2-3 CBCT. CBCT projections were sorted into 8 breathing phases using the Amsterdam Shroud method.. Two different 4D-CBCT reconstruction algorithms were then compared. The MA-ROOSTER method uses a priori knowledge from 4D-CT to improve the 4D-CBCT reconstruction, while the conjugate gradient iterative reconstruction only uses the sorted projections. The velocity fields (Vel4D-CT and Vel4D-CBCT) that characterize the motion between each phase were generated with deformable image registration (DIR). This allowed to create Mid-Position 3D images (MP-CT and MP-CBCT) from p4D-CT and 4D-CBCTs, respectively. The virtual MP-CT (vMP-CT) is then generated by deforming MP-CT on MP-CBCT with DIR. This vMP-CT can then be deformed to each phase using Vel4D-CBCT in order to generate a virtual 4D-CT (v4D-CT). As a result, The conjugate gradient method is found to be sufficient quality and about 5-10x faster than MAROOSTER. For patients with irregular breathing, 4DCBCT is found to be of less artifact therefore more robust than 4DCT. However, streak artifacts can limit left-right and ante-post motion accuracy in the conjugate gradient CBCTs especially for lower contrast liver tumors.