Comparison of full-size and patches-based learning approaches for aneurysm segmentation in TOF-MRI data

Abstract

The paper is interested in segmentation of intracranial aneurysms. Intracranial aneurysms are life-threatening issue. In this paper there are proposed two methods for this segmentation problem. First one is segmentation with use of full size images, the other one uses patches of the image, which could help decrease the ration between pixels representing background and pixels representing aneurysms. Data from ADAM challenge 2020 are used to train and evaluate these approaches. Using full images show better results in dice coefficient, which is 0.16 greater, then patched image approach.