Predicting Short-Term Survival after Gross Total or Near Total Resection in Glioblastomas by Machine Learning-Based Radiomic Analysis of Preoperative MRI

SIMPLE SUMMARY: Identifying GBM patients with very short survival could contribute to adapting the therapeutic approach. According to our results, high-precision models can be elaborated using basic MRI sequences available at any center, combined with advanced image analysis. Although there are several previous publications related to this topic, a survival threshold that may be clinically relevant has not been proposed. The importance of our study lies in selecting patients with total or near-total resection, the short-term survival end-point applied of six months, and the employment of user-friendly software that allows clinicians to explore new statistical methodologies and carry out complex tasks such as the extraction of radiomic features. Promoting the use of these technological tools will motivate other clinical researchers to get involved and take advantage of radiomics and artificial intelligence, tools that have come to reinforce our analytical capacity. ABSTRACT: Radiomics, in combination with artificial intelligence, has emerged as a powerful tool for the development of predictive models in neuro-oncology. Our study aims to find an answer to a clinically relevant question: is there a radiomic profile that can identify glioblastoma (GBM) patients with short-term survival after complete tumor resection? A retrospective study of GBM patients who underwent surgery was conducted in two institutions between January 2019 and January 2020, along with cases from public databases. Cases with gross total or near total tumor resection were included. Preoperative structural multiparametric magnetic resonance imaging (mpMRI) sequences were pre-processed, and a total of 15,720 radiomic features were extracted. After feature reduction, machine learning-based classifiers were used to predict early mortality (<6 months). Additionally, a survival analysis was performed using the random survival forest (RSF) algorithm. A total of 203 patients were enrolled in this study. In the classification task, the naive Bayes classifier obtained the best results in the test data set, with an area under the curve (AUC) of 0.769 and classification accuracy of 80%. The RSF model allowed the stratification of patients into low- and high-risk groups. In the test data set, this model obtained values of C-Index = 0.61, IBS = 0.123 and integrated AUC at six months of 0.761. In this study, we developed a reliable predictive model of short-term survival in GBM by applying open-source and user-friendly computational means. These new tools will assist clinicians in adapting our therapeutic approach considering individual patient characteristics.