ESTRO 2021 Abstract Book

S1521

ESTRO 2021

Purpose or Objective The microscopic spread of clonogenic cells outside of the GTV is thought to be the responsible for tumour recurrence in case of high-grade gliomas. Since the visibility threshold of MRI is limited to about ~8000 cells/mm3, models for glioma cells infiltration into the normal tissue have been proposed for guiding the definition of the target. Their clinical implementation, however, is still lacking one of the main reasons being the limited knowledge regarding their parameters. The aim of this study was to investigate the robustness of a diffusion-based model of tumour cell invasion with respect to its parameters variability. Materials and Methods A diffusion-based tumour growth model based on a time differential Fisher-Kolmogorov equation and describing the tumour cell concentration ( c ) as a function of the cell diffusion ( D ) and the cell proliferation ( ρ ) was considered in this study. The parameter D assumes different values in white and gray matter ( D W , D G ) and is 0 elsewhere, while the parameter ρ is limited by the tissue carrying capacity of tumour cells ( K ). The model was implemented on 126 glioblastoma multiforme cases. A full-brain T1 weighted MR image and a pre-segmented GTV ( V 0 ) were available. Image segmentation into white and gray matter was performed using the FAST algorithm. An initial tumour cell concentration ( c 0 ) was assigned to the GTV center-of-mass and the simulation iterated until each voxel of the GTV contour reached c ≥ 8000 cells/mm 3 (i.e. MRI visibility threshold). The obtained volume constituted the simulated GTV ( V ). Simulations were run for various D/ρ , D W /D G , K , and c 0 values, encompassing the broad range of figures proposed in the literature for these different parameters, and the V/V 0 ratio was scored for each of the cases. The optimal set of parameters, which would render a simulated volume ( V opt ) in closest agreement with the contoured one, was considered Results An example of a simulated tumour is shown in Fig. 1, with the segmented GTV in white. The mean value of V for all the considered cases as a function of D/ρ and D W /D G is shown in Fig. 2.a. The largest deviation between the simulated volumes corresponded to V/V opt equal to 1.16, with an average value of 1.08 indicating the robustness of the model with respect to D/ρ and D W /D G . Fig.2.b shows V opt / V 0 for all the 126 cases. A few outliers having V opt / V 0 > 5 (20 cases, 15.9%) and V opt / V 0 < 1 (four cases, 3.2%) are seen. Omitting these cases, the mean V opt / V 0 value was 2.21±0.95 and the median V opt / V 0 value of all cases was 1.95. The robustness analysis in terms of K and c 0 parameters resulted in a V opt / V 0 equal to 1.06 as worst case.