Investigating the potential of diffusion tensor atlases to generate anisotropic clinical tumor volumes in glioblastoma patients

Abstract

Background and purpose: Diffusion tensor imaging (DTI) has been proposed to guide the anisotropic expansion from gross tumor volume to clinical target volume (CTV), aiming to integrate known tumor spread patterns into the CTV. This study investigate the potential of using a DTI atlas as an alternative to patient-specific DTI for generating anisotropic CTVs.

Materials and methods: The dataset consisted of twenty-eight newly diagnosed glioblastoma patients from a Danish national DTI protocol with post-operative T1-contrast and DTI imaging. Three different DTI atlases, spatially aligned to the patient images using deformable image registration, were considered as alternatives. Anisotropic CTVs were constructed to match the volume of a 15 mm isotropic expansion by generating 3D distance maps using either patient- or atlas-DTI as input to the shortest path solver. The degree of CTV anisotropy was controlled by the migration ratio, modeling tumor cell migration along the dominant white matter fiber direction extracted from DTI. The similarity between patient- and atlas-DTI CTVs was analyzed using the Dice Similarity Coefficient (DSC), with significance testing according to a Wilcoxon test.

Results: The median (range) DSC between anisotropic CTVs generated using patient-specific and atlas-based DTI was 0.96 (0.93-0.97), 0.96 (0.93-0.97), and 0.95 (0.93-0.97) for the three atlases, respectively (p $>$ 0.01), for a migration ratio of 10. The results remained consistent over the range of studied migration ratios (2 to 100).

Conclusion: The high degree of similarity between all anisotropic CTVs indicates that atlas-DTI is a viable replacement for patient-specific DTI for incorporating fiber direction into the CTV.

Keywords: Anisotropic margin expansion; CTV; DTI; Glioblastoma; Radiotherapy.

Fig. 1

Example of distance maps using a uniform expansion (left) and anisotropic expansion (right). 3D distance maps calculated with Eq. (1), using different choices of parameter$\sigma$in Eq. (3). A value of$\sigma =1$results in quasi isotropic iso-distance contours (left), while setting and$\sigma =10$in white matter voxels results in anisotropic iso-distance contours (right). Expansions were restricted by a barrier segmentation, which prevented migration across structures such as the falx cerebri and into the ventricles. Note that the colorbar units in each image are in mm. However, the distance metric used in the anisotropic expansion on the right operates in a non-Euclidean space, where anisotropic regions are effectively “shortened”. As a result, the iso-distance contours do not correspond to actual physical distances. GTV: Gross Tumor Volume. mm: millimeters.

Fig. 2

Three examples of patients with all four anisotropic CTVs and the clinical-CTV. The three patients differ by their location of the GTV, extent of resection and contrast enhancement. The anisotropic CTVs are generated with a migration ratio equal to 10. Barrier structure is shown in blue. A: Non-contarst enhancing tumor in the right frontal region distant from corpus callosum B: Resection cavity near splenium corporis callosi. Example of minor barrier violations by the clinical-CTV. C: Irregularly shaped contrast enahncing tumor growing through splenium corporis callosi. Instance of substantial barrier violations by the clinical-CTV. GTV: Gross Tumor Volume, CTV: Clinical Tumor Volume, HCP: Human Connectome Project, IIT: Illinois Institute of Technology, MIITRA: Multichannel Illinois Institute of Technology & Rush university Aging .

Fig. 3

Feasibility of using an atlas-DTI in place of a patient-DTI. Comparison of DSC between anisotropic CTVs using Atlas- and Patient-DTI across migrations ratios (1, 2, 5, 10, 50, 100) and atlas versions (HCP, IIT, MIITRA). The p-values for difference comparisons are shown above. DSC: Dice Similarity Coefficient, DTI: Diffusion Tensor Imaging, HCP: Human Connectome Project, IIT: Illinois Institute of Technology, MIITRA: Multichannel Illinois Institute of Technology & Rush university Aging .

Fig. 4

Similarity of the anisotropic CTVs and the clinical-CTV. Comparison of DSC between anisotropic CTVs and clinical CTV across migrations ratios (1, 2, 5, 10, 50, 100) and DTI sources (HCP, IIT, MIITRA, MR). The p-values for difference comparisons are show above. DSC: Dice Similarity Coefficient, DTI: Diffusion Tensor Imaging, HCP: Human Connectome Project, IIT: Illinois Institute of Technology, MIITRA: Multichannel Illinois Institute of Technology & Rush university Aging, MR: Magnetic Resonance.

Fig. 5

The influence of migration ratio on similarity between anisotropic CTV and the clinical-CTV. Bootstrap uncertainty of median DSC on comparison of MR anisotropic CTVs versus clinical-CTVs, across migrations ratios (1, 2, 5, 10, 50, 100). DSC: Dice Similarity Coefficient, MR: Magnetic Resonance, CTV: Clinical Target Volume.