Sci—Fri AM(2): Brachy—04: Spatial Correspondence Metrics for Assessing Longitudinal Images of Evolving Tumor

In radiotherapy, accurate response assessment is necessary for determining treatment efficacy and identifying patients that would benefit from additional or alternate modes of therapy. Conventional response assessment methods rely on uni‐dimensional measurements of tumor morphology and function, and are incapable of quantifying heterogeneity of response within the tumor. A voxel‐to‐voxel analysis of image signal to evaluate the disease state may be more sensitive to change, and would permit the assessment of the spatial distribution of tumor response to therapy. A challenge to voxel‐based approaches is the uncertainty in the spatio‐temporal correspondence between assessed voxels and the underlying biological tissue element that they represent. This uncertainty arises from inter‐fraction motion, deformation, and evolution of pathology. These changes can result in an apparent signal intensity change that complicates the identification of signal changes due to altered tumor function. A framework for measuring the strength of voxel to voxel correspondences between serial images has been developed and was tested on MR images of glioblastoma acquired pre and post‐chemoradiotherapy. Strength of correspondence was determined by characterizing each voxel with a feature vector derived from a multi‐resolution local neighbourhood computation of histogram moments, image gradients and mutual information. The metric provides a measure of confidence in the spatial‐temporal correspondence between assessed voxels. A minimum requirement for strength of correspondence can then be defined and used to identify a sub‐population of voxels with sufficient anatomic support to inform the selection of functional image voxel pairs for signal change analysis and quantification of response.