Item

Abstract

With the raising of deep learning methods on medical imaging tasks, super resolution(SR) was well investigated to improve image resolution to facilitate more applications One of them is segmentation task in Ultra high field MRI, served as a way of augmenting synthetic data, SR could help to not only improve model accuracy by expanding the trainset, but also was shown to perform better than traditionally interpolated images when fed to segmentation network Especially in Ultra high resolution(< 0 6 mm), lack of accuracy is problematic when a high precision segmentation is required such as for the estimation of cortical thickness In this work, we developed a Generative Adversarial Network(GAN) to perform SR task, which also generates images different resolution that was more accurately segmented.