Activation Functions Evaluation to Improve Performance of Convolutional Neural Network in Brain Disease Classification Based on Magnetic Resonance Images

Early detection and treatment of brain disease are essential. However, brain disease diagnosis used to be challenging, on the other hand imaging techniques such as MRI make it easier. For the past years, many researchers have used several methods of Machine Learning and Deep Learning to diagnose brain abnormalities without any human help. Convolutional Neural Network is the best method to extract features of images automatically. In this study, a Deep Learning model of Convolutional Neural Network algorithm is applied to classify brain MR Images into normal and abnormal classes. The constructed network architecture was evaluated based on several activation functions and numbers of epoch. The experiment results achieved a significant performance with the best accuracy of 99.12% and dice score of 98.17% using ELU activation function at epoch 50. This result indicates that the proposed method has improved the performance of Convolutional Neural Network in brain disease classification.