Improving Diagnostic Accuracy of Brain Tumor MRI Classification Using Generative AI and Deep Learning Techniques

Abstract

Brain Tumors are one of the most aggressive and deadly medical diseases with living individuals requiring a quick and highly sensitive diagnostic approaches to improve patient outcomes. Classification of brain tumors from magnetic resonance imaging (MRI) is still a difficult problem for medical imaging because the problem is highly complex, varies and the differences between the tumor types are subtle. Diagnostic accuracy and consistency are recently improved through the advances in Generative Artificial Intelligence (AI) and Deep Learning (DL) techniques. However, this paper presents a robust and efficient model based on the combination of the Generative AI approaches i.e. Generative Adversarial Networks (GANs) and Convolutional Neural Networks (CNNs) for the multi-classification of brain tumors from MRI images. With medical image dataset being limited, the GAN model is trained to augment the dataset which solves the problem of limited dataset and also that of the class imbalance. It trains and evaluates the CNN architecture on both original and artificially augmented datasets and shows that increased accuracy and generalization. The experimental results demonstrate that proposed combined approach significantly performs better than the traditional methods, thus allowing radiologists to reach higher precision, sensitivity and specificity in tumor identification and classification. The proposed methodology is a significant improvement in speeding, reliability, and precision in clinical diagnostics for Brain Tumors.