Brain Metastasis Origin and Patient Mortality Predictions Using MRI with Clinical and Imaging Feature Information by Deep Learning Architectures

Cancer is one of the worst prevailing diseases that affects millions of lives each year. Cancers is the uncontrolled growth of a type of cells which when travel to other parts of the body cause metastatic cancers. This paper focuses on brain metastasis. This paper uses machine learning and deep learning algorithms to predict the origin of the metastasis and predict the mortality in brain cancers. The study uses MRI with clinical and imaging feature information which also contains lifestyle factors like smoking along with medical attributes like edema. To predict the origin of the metastasis, LightGBM performed the best out of the boosting algorithms and predicted with an accuracy of 97.12%. Furthermore, 11 machine learning and deep learning algorithms were implemented to compare and find the optimum algorithm for mortality prediction in brain metastasis patients. Recurrent Neural Network predicted mortality with an accuracy of 99.50% among all the algorithms. Feature importance scores are also extracted for the best performing machine learning algorithm, XGBoost and it shows that edema volume has the highest importance in mortality, with smoking being the third most influential factor.