Ultrasound Images Resolution Enhancement using Generative Adversarial Network

Abstract

This thesis focuses on enhancing ultrasound scan resolution and image quality, particularly for the detection of breast cancer. During pregnancy, diagnostic ultrasound is frequently used to examine internal organs and track fetal progress. The presence of noise in ultrasonic imaging, however, makes diagnosis and interpretation difficult. A generative adversarial network (GAN) is suggested as a deep learning method to address this problem. The GAN is made up of a generator network that has been trained to transform low-resolution ultrasonic in puts into high-resolution outputs and a discriminator network that can tell real images from fake ones. Convolutional layers, skip connections, Multi resolution Convolution blocks (MRCB), and loss functions are all incor porated into the GAN model’s architectural choices, which are tailored to the peculiarities of breast cancer. For iterative optimization throughout the training phase, backpropagation and gradient descent techniques are used. The peak signal-to-noise ratio (PSNR) and structural similarity in dex (SSIM) are quantitative metrics that are used to assess the effectiveness of the proposed Ultrasound Images Resolution Enhancement GAN (UIRE GAN) approach. The findings show considerable increases in image quality, increasing the capability of breast cancer ultrasound imaging as a diagnostic tool. This study advances the area of breast cancer imaging and has the potential to help patients receive better healthcare.