Non-rigid registration on histopathological breast cancer images using deep learning

Cancer is one of the leading causes of death in the world, in particular, breast cancer is the most frequent in women. Early detection of this disease can significantly increase the survival rate. However, the diagnosis is difficult and time-consuming. Hence, many artificial intelligence application...

Descripción completa

Detalles Bibliográficos
Autor: Cavallari, Lucia
Tipo de recurso: tesis de maestría
Fecha de publicación:2022
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/378097
Acceso en línea:https://hdl.handle.net/2117/378097
Access Level:acceso abierto
Palabra clave:Image registration
Breast--Cancer--Histopathology
image registration
non-rigid registration
histopathological images
Registre d'imatges
Mama--Càncer--Histopatologia
Àrees temàtiques de la UPC::Enginyeria de la telecomunicació::Processament del senyal::Processament de la imatge i del senyal vídeo
Descripción
Sumario:Cancer is one of the leading causes of death in the world, in particular, breast cancer is the most frequent in women. Early detection of this disease can significantly increase the survival rate. However, the diagnosis is difficult and time-consuming. Hence, many artificial intelligence applications have been deployed to speed up this procedure. In this MSc thesis, we propose an automatic framework that could help pathologists to improve and speed up the first step of the diagnosis of cancer. It will facilitate the cross-slide analysis of different tissue samples extracted from a selected area where cancer could be present. It will allow either pathologists to easily compare tissue structures to understand the disease's seriousness or the automatic analysis algorithms to work with several stains at once. The proposed method tries to align pairs of high-resolution histological images, curving and stretching part of the tissue by applying a deformation field to one image of the pair.