Smart and robust segmentation of medical images using neural networks

From LRDE

Revision as of 17:22, 9 November 2020 by Bot (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Abstract

Gliomas are a category of brain tumors that have different degrees of malignancy, shapes and textures. Manual segmentation by experts is a challenging task because of the heterogeneity of these tumors. Several methods of automated gliomas segmentation have been studied at MICCAI 2019 BraTS Challenge. We want to improve the segmentation results submitted last year by LRDE's team, using a 2-step VGG architecture. This convolutional neural networkclassically used for natural image categorization, has been adapted for medical image segmentation through transfert learning and pseudo-3D techniques. Improvements are done through robustness assessment, study of image features, and new method.