A Comparative Analysis of Machine Learning and Deep Learning Models for Brain Tumor Survival Prediction

It is very important to predict the survival of patients with a brain tumour in order to provide better treatment and prognosis. The previously used techniques like the Cox Proportimal Hazards model are still used but with the recent development in deep learning, the prediction accuracy is better. This work considers the following survival prediction methods: KM estimators, Cox regression model, RSF, CNN-LSTM, and Vision Transformers on MRI scans. The C-index, the log-rank test, and the Brier score are the performance measures that are used. Also, post hoc explanation methods like Grad-CAM and SHAP are applied to aid in understanding the deep learning predictions. The results show that the integration of deep learning with statistical survival analysis methods offers the best balance between performance and interpretability, which can be useful for the future development of clinically feasible AI-based survival prediction models.