This project aims to develop new techniques for image quality improvement for diagnosis based on medical images, with a focus on images acquired with lower cost equipment, which generally result in lower quality images that may hinder the overall performance of the diagnostic tool. The developed methods will go beyond traditional image processing techniques, delving into Deep Learning methods, including Variational Autoencoders (VAE) and Generative Adversarial Networks (GAN), already shown to be useful for image processing, namely cleaning, denoising and even in super-resolution challenges.
When considering the goal of decentralized healthcare and portable diagnostic tools, there is usually a trade-off between complexity/cost and accuracy/performance. This work aims at mitigating this gap by improving image quality for diagnosis, especially when acquired with lower cost equipment.
Author: Beatriz Simões
Type: MSc thesis
Partner: FCT NOVA – Faculdade de Ciências e Tecnologia da Universidade