ESTRO 2024 - Abstract Book

S5021

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

Three strategies were adopted to define the OS time: (i) continuous OS time which refers to the time between the first diagnosis and cancer-caused death, (ii) binary OS time which categorizes subjects into short and long-term survivors, and (iii) three class OS time which splits the subjects into short-, mid-, and long-term survivors. Conventional radiomics pipeline and end-to-end Deep Learning (DL) models were examined to predict the OS time by analyzing the GTV, clinical CTV (CTVclinic), and simulated CTV (CTVsim).

Results:

Table 1 shows a summary of the results quantified over repeated 5-fold cross-validation experiments in terms of the Area Under the Receiver Operating Characteristic Curve (AUROC) and Mean Square Error (MSE) metrics.

Table 1. Comparing the prognostic power between the radiomics and DL models for different tumor borders

In addition, conducting the Wilcoxon signed ranked test for MSE metrics and the Delong test for pairwise AUROC comparison resulted in significant differences between all examined tumoral regions (pvalue). Moreover, in addition to the potential of the simulated clinical target volume accounting for the cell invasion to guide the treatment planning optimization, the prediction power of the CTVsim outperformed the reported results of relevant studies conducted over GTV regions [5], [6].

Conclusion:

The overall prognostic performance achieved by the simulated clinical target volume demonstrated the potential of the employed tumor invasion model to estimate the distribution of the invisible parts of GBM, which in turn can improve the performance of OS time predictions.

Keywords: glioma, clinical target volume

References:

[1] A. Mang, S. Bakas, S. Subramanian, C. Davatzikos, and G. Biros, “Integrated Biophysical Modeling and Image Analysis: Application to Neuro-Oncology,” https://doi.org/10.1146/annurev-bioeng-062117-121105, vol. 22, pp. 309– 341, Jun. 2020, doi: 10.1146/ANNUREV-BIOENG-062117-121105. [2] W. Häger, M. Lazzeroni, M. Astaraki, and I. Toma-Daşu, “CTV Delineation for High-Grade Gliomas: Is There Agreement With Tumor Cell Invasion Models?,” Adv Radiat Oncol, vol. 7, no. 5, p. 100987, Sep. 2022, doi: 10.1016/J.ADRO.2022.100987.