Publication

Abstract : Advancements in medical technology have brought a significant change in healthcare by enabling accurate diagnosis and personalized treatments. However, machine learning algorithms' achievement in medical image analysis depends on the amount and quality of available data. Data augmentation, which is a process to diversify the data set, has emerged as a crucial tool to address the medical image shortage. A new method has been presented in this paper to enhance brain tumor MRI images using Generative Adversarial Networks (GANs). It was trained using pre-processed images, where a generator creates images from random noise and a discriminator distinguishes real from generated images. The post-processing involved clamping and smoothing to ensure pixel values and visual quality. The quantitative evaluations of the method included a Structural Similarity Index (SSIM) for image similarity, and a high pixel accuracy of 86%, along with Intersection over Union (IoU) for segmented image quality assessment and we added the generated image to a pretrained model which had an accuracy of 84% and we improve it to 91% by adding some generated images.