ESTRO 2024 - Abstract Book

S5080

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

Clin. (1996). 4. Fan, X. et al. Multiparametric MRI and machine learning based radiomic models for preoperative prediction of multiple biological characteristics in prostate cancer. Frontiers in Oncology 12, 839621 (2022). Cancer Res. 2, 1445-1451

1744

Digital Poster

Prognostic Prediction of High-grade Glioma Patients After Radiotherapy Using Radiomics Analysis

Nobuki Imano 1 , Daisuke Kawahara 2 , Riku Nishioka 1 , Yuzuha Kadooka 1 , Sawane Fukumoto 1 , Takashi Sadatoki 1 , Tsuyoshi Katsuta 1 , Junichi Hirokawa 1 , Ikuno Nishibuchi 1 , Yuji Murakami 1 1 Hiroshima university hospital, Department of Radiation Oncology, Hiroshima, Japan. 2 Hiroshima uiversity hospital, Department of Radiation Oncology, Hiroshima, Japan

Purpose/Objective:

Prognostic prediction of glioma has traditionally been based on clinical factors like age, performance status (PS), histology, and surgical resection. In recent years, prognostic factors based on molecular pathology, like the presence or absence of Isocitrate dehydrogenase (IDH) mutations and Methylation of the O6-methylguanine DNA methyltransferase (MGMT) promoter, have been reported. Although these factors correlate to some extent with prognosis, the prognosis for high-grade gliomas remains poor, and an established method for accurate prognosis remains lacking. The primary aim of this study is to establish a prognostic prediction model for high-grade glioma patients after radiotherapy using multi-region radiomics analysis from MRI images and dose distribution. Additionally, we seek to compare its predictive accuracy with a model solely created from traditional clinical data.

Material/Methods:

We conducted a retrospective analysis of 114 high-grade glioma patients who underwent radical radiotherapy at our institution from 2009 to 2022. For prognostic prediction based on clinical factors, we analyzed age, gender, Eastern Cooperative Oncology Group PS, histology, presence or absence of IDH mutation, presence or absence of MGMT promoter methylation, (Mindbomb Homolog-1) MIB-1 value, and surgical resection extent. Clinical factors were analyzed using univariate/multivariate analysis to select risk factors. To establish radiomics model, we performed multi-region radiomics analysis. We first created 44 analysis regions by modulating the gross tumor volume, clinical target volume, and planning target volume on treatment planning CT. The 44 regions of interest created from the planning CT were fused with both the MRI and the dose distribution. For the analysis, we used the following types of MRI images: T1-weighted imaging, T2-weighted imaging, T2-Fluid Attenuated Inversion Recovery, Gadolinium enhanced T1-weighted imaging, Diffusion-weighted imaging with a b-value of 1000 s/mm2 (DWI b=1000), and DWI b=4000. For each patient and each image, 844 radiomic features were analyzed across 44 regions of interest. The aforementioned six types of MRI images were analyzed, resulting in the analysis of 222,816 radiomic features per patient. Dimension reduction was performed using the least absolute shrinkage and selection operator (LASSO) analysis to select the relevant features for prognosis prediction. Radiomics-score was generated using Cox regression analysis based on these selected features. These models were created from 70% of the cases, and the accuracy of the model was evaluated using the concordance index (C-index) for the remaining 30%. Both the clinical and radiomics