An approach to generate synthetic 4DCT datasets to benchmark Mid-Position implementations.

Area of Science:

• Medical Imaging
• Radiotherapy Physics

Background:

• Four-dimensional computed tomography (4DCT) is crucial for radiation therapy planning, enabling motion management.
• Mid-Position (Mid-P) images, derived from 4DCT data via deformable image registration, offer potential for reduced planning target volume (PTV) margins.
• A lack of readily available Mid-P images for testing hinders the validation of Mid-P generation algorithms.

Purpose of the Study:

• To describe a novel approach for generating synthetic 4DCT datasets with corresponding Mid-P images.
• To create a benchmark dataset for validating Mid-P image generation techniques.
• To facilitate the development and standardization of Mid-P image applications in radiotherapy.

Main Methods:

• Twenty synthetic 4DCT datasets and their reference Mid-P images were generated from clinical 4DCT data.
• Deformable Vector Fields (DVFs) were computed by registering an anchor phase to other phases.
• DVFs were used to warp the anchor phase, generating synthetic 4DCT datasets and Mid-P images, along with corresponding tumor masks.

Main Results:

• Generated synthetic Mid-P images demonstrated high similarity to reference images, with minor discrepancies in one noisy dataset.
• The largest motion amplitude difference was observed in the Superior-Inferior direction (-2.6 mm).
• Statistical analysis revealed no significant performance differences among three tested Mid-P implementations.

Conclusions:

• The proposed method provides a reliable, independent approach for validating Mid-P image generation algorithms.
• The synthetic datasets and experimental framework support the ongoing development of Mid-P image applications.
• This work contributes to the advancement of motion-adaptive radiotherapy planning through robust image validation.