BrainGAN: Brain MRI Image Generation and Classification Framework Using GAN Architectures and CNN Models.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Deep learning models require adaptation for sensitive medical imaging applications.
• Privacy concerns limit the use of real medical data for machine learning.
• Lack of sufficient brain MRI images hinders accurate tumor classification.

Purpose of the Study:

• To propose a framework (BrainGAN) for generating and classifying brain MRI images using Generative Adversarial Network (GAN) architectures.
• To develop an automated method for assessing the quality of generated medical images.
• To evaluate the performance of deep learning models on GAN-generated brain MRI data.

Main Methods:

• Utilized Generative Adversarial Network (GAN) architectures, including Deep Convolutional GAN (DCGAN) and Vanilla GAN, for synthetic brain MRI image generation.
• Implemented and trained three deep learning models: Convolutional Neural Network (CNN), MobileNetV2, and ResNet152V2.
• Evaluated model performance on a test set of real brain MRI images after training on GAN-generated data.

Main Results:

• ResNet152V2 demonstrated superior performance compared to CNN and MobileNetV2.
• ResNet152V2 achieved 99.09% accuracy, 99.12% precision, 99.08% recall, and 99.51% AUC on DCGAN-generated images.
• The proposed framework successfully generated and classified brain MRI images with high accuracy.

Conclusions:

• GAN-based data augmentation effectively addresses the scarcity of brain MRI data for deep learning.
• The BrainGAN framework provides a viable solution for enhancing medical image classification tasks.
• ResNet152V2 shows significant potential for accurate brain tumor classification using synthetic MRI data.