Comparative Analysis of Overall Survival Prediction for Glioblastoma Patients: Neural Networks vs. Random Forests with Data Augmentation

Automated brain tumor segmentation is crucial for the accurate diagnosis and prognosis of patients. Additionally, clinical and radiomics features of brain tumors aid in predicting patients' overall survival. This paper primarily aims to segment tumors from the BraTS2020 benchmark dataset and utilize clinical and radiomics features to predict patients' overall survival. The proposed study used a balanced random forest classifier and an artificial neural network to separate clinical and radiomic datasets from BraTS 2020. The radiomics data was synthesized using the Gretel.ai tool as a data augmentation method. The proposed artificial neural network model achieves an overall survival accuracy of 90% for the clinical dataset and 95% for the radiomics dataset of BraTS 2020, and the proposed balanced random forest model achieves an overall survival accuracy of 92% for the clinical dataset and 92% for the radiomics dataset. The proposed study used a machine learning approach called a balanced random forest, a deep learning approach that is an artificial neural network for predicting overall survival along with data augmentation and achieving better accuracy than other state-of-the-art methods.