Automatic segmentation and online virtualCT in head-and-neck adaptive radiation therapy

Area of Science:

• Medical Imaging
• Radiation Oncology
• Computational Anatomy

Background:

• Adaptive radiation therapy (ART) for head-and-neck (HN) cancer requires frequent imaging updates.
• Generating new computed tomography (CT) scans for replanning can be time-consuming and resource-intensive.
• Virtual CT generation offers a potential solution to streamline the ART workflow.

Purpose of the Study:

• To develop and validate an efficient, automatic strategy for generating online virtual CT scans.
• To support adaptive radiation therapy (ART) in head-and-neck (HN) cancer treatment.
• To reduce the need for full CT-based replanning.

Main Methods:

• Retrospective analysis of 20 HN cancer patients treated with intensity modulated radiation therapy (IMRT).
• Generation of 28 virtual CT scans using nonrigid registration between simulation CT (CTsim) and cone beam CT (CBCT) images.
• Validation against real replanning CT (CTrepl) using Dice similarity coefficient (DSC), center of mass (COM) distance, and root mean square error (RMSE).

Main Results:

• Deformation between CTrepl and CBCT was less than one voxel.
• Median DSC was approximately 0.8 for mandible and parotid glands, and 0.55 for nodal gross tumor volume (nGTV).
• COM distance and RMSE were comparable to image resolution, with no significant correlation to deformation.

Conclusions:

• Deformable image registration can significantly reduce the need for full CT-based replanning in HN radiation therapy.
• The strategy supports swift and objective decision-making in clinical practice.
• Further research is required to enhance nGTV localization accuracy.