Modeling the Volumetric Change of Head and Neck Tumor in Response to Radiation Therapy

Abstract

A large fraction of head-n-neck (HN) cancer patients experience tumor or lymph node volume change during the course of radiation therapy. Modeling the volumetric tumor kinetics is an essential step to better understand the therapeutic process and to compute the cumulative dose for offline/online adaptive radiation therapy. In this work we develop a novel method to model the tumor shrinkage and illustrate the performance by applying the technique to five clinical HN cases. Tumor shrinkage or growth cannot be described solely by an elasticity or similarity based deformable model because of the non-conservation of the tissue. The underlying hypothesis here is that, when a tumor grows or shrinks, some of the tissue features in the tumor volume or at the boundary region are preserved, despite of possible scale changes. Our algorithm focuses on auto-detecting the homologous tissue features shared by the planning CT and subsequent on-treatment CBCT images (typically, 3∼5 CBCT scans for each patient). Once the tissue features are identified, a voxel-to-voxel correspondence between two images for the rest of points can be established by basis spline interpolation. The tissue feature at a point of interest in an image is described by the local information using the Scale Invariance Feature Transformation (SIFT) method, which includes scale-space extrema detection, control point localization, orientation assignment and control point descriptor. If a feature region in the fixed image is mapped correctly to the moving image, it will necessarily be mapped back to the original region in the fixed image when we apply the inverse map to the corresponding region in the moving image. The bi-directional mapping procedure was included in our algorithm to increase the precision of tissue feature correspondence. The proposed method is applied to a number of synthetic cases and five clinical HN cases. A novel model describing the kinetics of tumor volume change in response to radiation therapy has been established. The identification of tissue features shared by the images acquired at different time points plays an essential role in the modeling process. Our bi-directional SIFT method further warrants the reliable association of the tissue features. In the testing cases based on synthetic images with artificially introduced “tumor” volume changes, the bi-directional SIFT was able to correctly identify majority of the features appeared before the introduction of artificial volumetric changes. Because the number of identifiable feature points are generally large in HN cases, the voxel-to-voxel correspondence resulted from the subsequent basis spline interpolation almost reproduced the “ground truth” solution provided that the tumor volume change is less than 15%∼20%. Generally, the model works well for a small or modest volume change and suggests that the time span between two consequent imaging sessions should not be unreasonably far away in order for the model to function properly. Tumor change kinetics can be derived reliably from consecutive imaging data by using the proposed SIFT-based model. The model is valuable to better understand the tumor response to radiation therapy, and to compute the cumulative dose distribution during the treatment. The latter makes adaptive therapy with consideration of dose delivery history practically feasible.