A Hybrid CNN–ELM Approach for Accurate Brain Tumor Classification

Abstract

These days, brain tumors are among the most common health issues in many countries due to factors including career, genetics, and global warming.  Gliomas are the most aggressive type of brain tumor, resulting in a high-grade life.  A brain tumor can be saved if it is found and diagnosed early. Brain tumors are becoming one of the most common health problems in many countries throughout the world as a result of global warming, heredity, career, and other factors. Gliomas are the most aggressive type of brain tumor, and they have a long life. Early detection and diagnosis of a brain tumor can be life-saving. Magnetic Resonance Imaging (MRI) is a simple method for determining tumor location. However, the large number of data generated by MRI makes manual segmentation difficult to complete within a reasonable time frame, limiting the use of reliable quantitative metrics in clinical practice. An autonomous and dependable approach is required for effectively segmenting tumors. This paper provides a novel technique for brain tumor segmentation based on cascaded UNET architecture. Initially, the photos are downsized into 128 × 128 pixels to optimize computing time. Additionally, increased adaptive gamma correction is applied to the input photos to improve pixel quality. Training and validation are performed on selected slices including tumor areas. The suggested model has been validated using the BraTS 2020 dataset. The proposed work had an accuracy of 99.70%, 99.46% specificity, 96.14% sensitivity, and 98.25% precision. The proposed method performs better than earlier methods in terms of accuracy.