Room: Davidson Ballroom B
Purpose: To develop a methodology for characterizing longitudinal changes in intracranial vessel diameter using Magnetic Resonance Angiography as they relate to proton radiotherapy.
Methods: Time-of-flight Magnetic Resonance Angiography scans were acquired in 106 pediatric patients with Craniopharyngioma at 481 longitudinal time points (median 4 scans). Images were resampled to an isotropic resolution of 0.5x0.5x0.5 mm3. Intensity normalization, noise reduction with Gaussian filtering followed by multi-scale vessel filtering was applied to obtain a 3D binary vessel mask. Morphological closing was applied to ensure the 3D binary vasculature mask was fully closed. Connected components analysis was used to remove extra-cranial and unconnected vessels. Finally, a fast-marching algorithm was used to calculate diameter measurements and centerline spatial coordinates of the cerebral vasculature. For longitudinal diameter comparison, combinatorial utilization of rigid and deformable registration was performed using pre-therapy to post-therapy images. Eclipse (Varian) calculated proton physical dose was interpolated at the computed centerline coordinates along with distance to nearest bifurcation. Similar analysis was performed using Monte Carlo (TOPAS) scored, dose weighted linear energy transfer (LET) and dose multiplied by LET.
Results: Diameter measurements were averaged over 1 mm segments with computed vessel diameters ranging from 0.01 to 6.22, with median of 1.25 mm. Diameter distortion due to registration was assessed by sequential application of forward and inverse transformations. Distortions were within 0.5 mm for 95% vessel segments, larger distortions were due to missing vessels across scans. Outlier detection using Bland Altman threshold with 95% confidence interval detected 4.69%, 6.57% outliers using absolute difference, percentage difference respectively.
Conclusion: Using the proposed method, objective changes in vessel diameter can be acquired non-invasively from routine imaging. High throughput analyses of imaging derived vascular trees combined with clinical and treatment parameters will allow for rigorous modeling of dose, volume, and vessel diameter changes.
Segmentation, Vascular Imaging, MR