SU-F-R-19: Novel Radiomics Quantifying Tumor Structural Evolution Using Deformation Vector Field: Application for Tumor Response Assessment




to quantify tumor volume/shape evolution due to chemoradiotherapy using structural evolution maps computed from deformation field. To extract radiomics from these maps for tumor response assessment.


In 20 patients with esophageal cancer, BSpline deformable registration was performed to register post-treatment CT image to pre-treatment CT image. The resulting deformation (vector) field represented tissue displacement at every voxel. We computed three structural evolution maps from the deformation field: (1) contraction map, to detect local tumor shrinking/growth, (2) Jacobian map, to characterize the ratio of local volumetric variation, and (3) Hessian map, to track dynamic evolving process. We then extracted intensity and textural radiomics from the evolution maps for both the GTV and the 1-cm ring surrounding the GTV, to characterize tumor structural evolution. ROC analysis and Wilcoxon rank sum test were employed to identify important predictors for tumor response.


We visually overlaid structural evolution maps on registered CT image. Local tumor shrinking/growth regions were identified with contraction map and Jacobian map. Hessian map dynamically tracked the evolving process of a tumor. Nine radiomics (four intensity, five texture) characterizing tumor structural evolution were found useful for tumor response assessment. A tumor was more likely to achieve complete pathologic response when skewness of the contraction map in the 1-cm ring was lower (AUC = 0.84, p = 0.03), or when Kurtosis of the Jacobian map was lower in the GTV (AUC = 0.77, p = 0.08), or when short run emphasis of the Hessian map was higher (AUC = 0.87, p = 0.01).


Structural evolution maps computed from the deformation field can quantify structural changes in a tumor. The radiomics extracted from these maps were useful predictors of tumor response.

This work was supported in part by the National Cancer Institute Grants R01CA172638.