ESTRO 2024 - Abstract Book

S5121

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

1 Stockholm University, Departmen of Physics, Stockholm, Sweden. 2 Karolinska Institute, Department of Oncology and Pathology, Stockholm, Sweden. 3 Royal Institute of Technology, Department of Biomedical Engineering and Health Systems, Huddinge, Sweden

Purpose/Objective:

Prognosis of glioblastoma (GBM) remains poor in spite of the progress in radiotherapy and imaging techniques. It has been suggested that aggressive and widespread tumor invasion of normal tissue is responsible for the tumor recurrence following treatment. Since the complete extension of invasion is undetectable on conventional imaging, it is not deliberately treated. To improve treatment of GBMs, models of tumor invasion have been developed to simulate tumor invasion from standard imaging data. A previous study performed in our group showed that the model for tumor invasion could predict the overall survival time better than more common metrics such as the size of the gross target volume (GTV). The purpose of this study was to investigate whether the model for tumor invasion could also predict the tumor progression at different time points after the end of the treatment.

Material/Methods:

A tumor invasion model based on a Fisher-Kolmogorov differential equation for cell diffusion and proliferation was applied to 56 cases of treated GBMs. Tumor progression was assessed at 20±10 days and 90±20 days after the end of radiotherapy. A binary classifier for tumor progression was provided for each case. Tumor invasion was simulated using the segmented gross tumor volumes (GTVs) as basis and the distributions of the cells into the normal brain tissue were obtained. To quantitatively describe the extent of tumor invasion, volumes encompassed by cell number isocontours (in cells/mm 3 ) were defined. Correlation between the observable tumor volume, GTV, as well as the simulated tumor invasion described by the volumes encompassed by different cell number isocontours and progression was determined from ROC curves.

Results:

The results showed that the size of the GTV did not correlate with the progression at any time point of the follow up. The volume of the tumor invasion into the normal brain was not able to predict the local control at the late follow up time. However, the volume defined by the isocontour of 100 cells/mm 3 (V 100 ), correlated with the observation of progression at the early time point for follow up. Figure 1 shows the ROC curves for: a) the GTV; and b) V 100 . As it could be observed, correlation between GTV and tumor progression was not significant (p = 0.684) but the correlation between V 100 and progression was however significant, with an area-under-the-curve (AUC) of 0.7 (p = 0.023), which indicates its ability to predict the response at the very early time point after the end of the treatment.