Advanced Brain Tumor Classification Via Transfer Learning From MRI With Deep Neural Networks

Abstract

An innovative approach to brain tumor classification using magnetic resonance imaging (MRI) coupled with deep learning methodologies. By harnessing the power of transfer learning, we employed renowned deep learning architectures like Xception, NasNet Large, DenseNet121, and InceptionResNetV2 to extract intricate features from MRI scans. Through meticulous preprocessing, data augmentation, and training with diverse optimization algorithms on benchmark datasets, our CNN model based on Xception emerged as the most effective, boasting superior accuracy, sensitivity, precision, specificity, and F1-score metrics. This model surpasses existing studies, highlighting its potential for swift and precise brain tumor identification, a pivotal factor in early diagnosis and treatment planning. Additionally, we explored the advanced MobileNetV2 algorithm, achieving a remarkable 100% accuracy when trained on an expansive dataset comprising 3000 images, encompassing both tumor and non-tumor samples.