Brain Tumor Recognition from MRI Using Deep Learning with Data Balancing Methods and Its Explainability with AI

Abstract

The abnormal growth of brain cells is a disease that can damage human brains and lead to malignant brain cancer. This abnormal growth in the brain is called a brain tumor. In the past few years, the number of brain tumor-related patients has increased at an extraordinary rate, due to which brain tumor is at the tenth position in the list of common tumors. But if a brain tumor is detected very early, it is possible to increase the way to get rid of this disease. Researchers are working tirelessly to develop automatic systems for brain tumor recognition. Hence, this paper proposes an automated system for brain tumor recognition based on deep learning and image processing. The dataset used in this article was collected from Kaggle and was not balanced. The dataset was preprocessed using a high boost filter and non-local means denoising technique, and then augmentation and SMOTE were applied to balance the dataset. Finally, different deep learning architectures were utilized. Among them, the ResNet50 model achieved 99.62% accuracy for 15-class recognition. The experimental result showed the reliability of the proposed model. This research will enable the research community to explore further brain tumor recognition from MRI using deep learning. It will also help the physician to detect brain tumors earlier.