Accurate Detection and Classification of Brain Tumors using U-Net and Extreme Learning Module

Abnormal cell growth in the brain can lead to the formation of brain tumors. Gliomas, in particular, are highly dangerous due to their low survival rate and their irregular shape and complex boundaries, which make physical detection challenging. These characteristics render gliomas the most perilous type of brain tumor. Magnetic resonance imaging (MRI) is the most commonly used imaging technique, allowing radiologists to examine the brain using radio waves and a magnet. However, manually identifying the tumor region is both time-consuming and difficult. Therefore, an accurate and automated method is crucial for effectively segmenting and predicting brain tumors. This study aims to propose a method utilizing U-Net for brain tumor segmentation and an ensemble learning model to predict patient survival chances. In this work, U-Net and Extreme Learning Machine (ELM) are employed. Local Binary Patterns (LBP) is utilized to extract statistical and shape characteristics. The BRATS 2020 dataset is used for both training and testing the model. The images are segmented using U-Net, and the brain tumors are classified using ELM. This approach achieves approximately 99.87% accuracy, 98.96% specificity, 99.58% sensitivity, and 98.75% precision.