Enhanced Brain Tumor Classification Using Capsule Networks with Hybrid Activation Function

Abstract

Detection of life-threatening disease at an early stage is crucial for improving patient survival rates. The advancement in diagnostic technology has made disease identification faster and more accurate. Brain tumors are very critical and require prompt and precise detection for treatment. Deep learning algorithms can aid medical experts in diagnosis and treatment. Capsule Networks, a Deep Learning technique, have shown significant potential in medical diagnosis. The performance of capsule networks is driven mainly by the choice of activation functions, as Activation functions help in improving sensitivity to detect the spatial and orientational rotation. In this study, a hybrid activation function combining ReLU and Swish is proposed to make a more effective activation function. The proposed model achieves an accuracy of 96.83%, precision of 96.84%, recall of 96.83%, and a ROC value of 98.07%.  Comparative analysis shows that the proposed approach performs better than the other state-of-the-art techniques, highlighting its effectiveness in brain tumor detection.