Fuzzy guided ensemble inference system for brain tumor classification

The abnormal growth of cells inside or near the brain is called a brain tumor. Brain tumors can be benign (non-cancerous) or malignant (cancerous). Both these types can exert pressure on the surrounding brain tissue, increasing intracranial pressure. As the tumor grows, it presses on the nerves and brain tissues, causing symptoms like persistent headaches, seizures, vision or hearing issues and changes in personality, coordination and balance, these of which will completely ruin the normal life of people. Since treating these tumors is very difficult at the late stages, it is highly significant to find them at the early stages. Understanding the importance of early identification of brain tumors, a fuzzy logic-based ensemble method using Convolutional Neural Networks (CNN) named Fuzzy Guided Ensemble Inference System (FGEIS) is proposed. It is developed to identify tumors from MRI images with a high success rate. The FGEIS approach uses ensemble learning that encompasses variants of four different architectures - Densenet, Resnet, VGG, and Mobilenet. While Resnet's residual connections allow for effective hierarchical feature learning for a variety of tumor types, Densenet supports feature reuse by collecting fine-grained tumor textures. The model is suitable for clinical usage because VGG prioritizes local spatial details that are important for accurate tumor localization, while mobilenet provides computing efficiency. These high-performing models are then integrated and applied through a fuzzy logic system. The experiments show improved performance of ensemble models over individual models with higher classification accuracy of 99.85 percentage.