Magnetic Resonance Image Enhancement For Early Brain Tumor Diagnosis Using Wasserstein Generative Adversarial Network With Gradient Penalty

Abstract

Early detection of brain tumours is really important since symptoms are often vague, which can delay diagnosis. MRI scans can be affected by patient movement, bodily changes, and the need for expert interpretation. Generative Adversarial Networks have become a powerful tool for improving the quality of medical images for brain cancer diagnosis and filling in missing data. So, in this work, we aim to improve the quality of MRI images of brain tumours using an advanced deep learning model based on GAN, specifically the Wasserstein GAN with Gradient Penalty (WGAN-GP) with a generator G and Dual Discriminator D (Dual D). This method is used to make grayscale images from the Br35H (2020) dataset, which includes medical brain tumour images, clearer and more accurate. We conducted many experiments with different network architectures and learning parameters in this research. Our results show an improvement in generated MRI Brain Tumor images with an accuracy of 99.634% and a PSNR of 42.42 dB. This research is important because it helps doctors and researchers get clearer images, making it easier to diagnose Brain Tumors accurately. It also helps improve medical treatment plans and outcomes since original images often have noise or distortions that can affect accuracy.