WE-AB-202-05: Validation of Lung Stress Maps for CT-Ventilation Imaging.

PURPOSE: To date, lung CT-ventilation imaging has been based on quantification of local breathing-induced changes in Hounsfield Units (HU) or volume. This work investigates the use of a stress map resulting from a biomechanical deformable image registration (DIR) algorithm as a metric of the ventilation function.

METHOD: Eight lung cancer patients presenting different kinds of ventilation defects were retrospectively analyzed. Additionally, to the 4DCT acquired for radiotherapy planning, five of them had PET and three had SPECT imaging following inhalation of Ga-68 and Tc-99m, respectively. For each patient, the inhale phase of the 4DCT was registered to the exhale phase using Morfeus, a biomechanical DIR algorithm based on the determination of boundary conditions on the lung surfaces and vessel tree. To take into account the heterogeneity of the tissue stiffness in the stress map estimation, each tetrahedral element of the finite-element model was assigned a Young's modulus ranging from 60kPa to 12MPa, as a function of the HU in the inhale CT. The node displacements and element stresses resulting from the numerical simulation were used to generate three CT-ventilation maps based on: (i) volume changes (Jacobian determinant), (ii) changes in HU, (iii) the maximum principal stress. The voxel-wise correlation between each CT-ventilation map and the PET or SPECT V image was computed in a lung mask.

RESULTS: For patients with PET, the mean (min-max) Spearman correlation coefficients r were: 0.33 (0.19-0.45), 0.36 (0.16-0.51) and 0.42 (0.21-0.59) considering the Jacobian, changes in HU and maximum principal stress, respectively. For patients with SPECT V, the mean r were: 0.12 (-0.12-0.43), 0.29 (0.22-0.45) and 0.33 (0.25-0.39).

CONCLUSION: The maximum principal stress maps showed a stronger correlation with the ventilation images than the previously proposed Jacobian or change in HU maps. This metric thus appears promising for CT-ventilation imaging. This work was funded in part by NIH P01CA059827.