Structure-Guided Target Volume Propagation for Adaptive Radiotherapy

The patent segments the deformation field into multiple local regions, each with its own deformation characteristics. By dividing the image space into zones with similar deformation properties and applying region-specific registration parameters, the method captures large structural motions more accurately while maintaining automation. This segmentation approach allows different parts of the anatomy to be registered with appropriate local transformations rather than a single global model.

The patent applies local quality by assigning different deformation properties to different regions of the image. Each local region is characterized by its own deformation parameters that reflect the specific tissue mechanics and motion patterns in that area. This enables the registration algorithm to adapt to varying deformation properties across the anatomy, improving accuracy for structures experiencing large motions while maintaining computational efficiency.

The patent performs preliminary deformation analysis and structure propagation before the actual treatment planning. By pre-calculating deformation vector fields and propagating structures automatically with region-specific parameters, the method prepares accurate initial contours that closely match the required anatomical accuracy. This preliminary automated registration reduces the subsequent manual refinement time significantly, as the pre-processed structures are already highly accurate.

The patent introduces an intermediary automated registration system that acts as a bridge between manual contouring and final treatment planning. This intermediary process generates highly accurate initial structures through region-specific deformable registration, which then serve as excellent starting points for minimal manual adjustment. The intermediary step captures most of the anatomical accuracy requirements automatically, leaving only fine-tuning for the physician.

The patent segments the deformation field into multiple local regions with distinct deformation characteristics. Each region is defined by spatial boundaries and assigned its own deformation parameters. This segmentation maintains relative algorithmic simplicity while dramatically improving accuracy, as each local model remains computationally tractable but collectively they capture complex varying deformation properties throughout the anatomy.

The patent applies dynamics by making the deformation model adaptive rather than static. The registration system dynamically selects and applies appropriate deformation parameters based on the local region being registered. This dynamic adaptation allows the algorithm to respond to varying tissue properties and motion patterns in different anatomical areas, achieving high precision without requiring a single overly complex global model.