Hybrid GAN-CNN Model for Brain Tumor Detecting and Classifying Diseases Based on MRI Images.

Abstract

Brain cancer diagnosis using MRI scans is aided by the advancement of artificial intelligence, a promising tool in medical imaging, but further optimisation is needed due to privacy constraints and limited medical imaging data availability.

To address these challenges, this study proposes a hybrid algorithm that integrates Convolutional Neural Networks (CNNs) with Generative Adversarial Networks (GANs) for data augmentation and improved classification accuracy. To enhance training efficiency, an additional 15 epochs are incorporated into each network and introduced to optimise brain cancer classification. Two deep learning models—CNN and GANs—are trained on synthetic MRI Dataset Kaggle (2022) images generated by GAN architectures and evaluated using real brain MRI scans.

Experimental results demonstrate that CNN outperforms the other models, achieving a loss of 0.2301, accuracy of 98.90%, validation loss of 0.4756, and validation accuracy of 84.56% when trained on brain cancer images generated by GANs with a generator loss of 1.4502, discriminator loss of 0.8163, and fake image accuracy of 83.12%. Real Image Accuracy: 92.34%

These findings confirm that the proposed hybrid algorithm 98.85% accuracy significantly enhances brain cancer classification using deep learning techniques.