Cerca nel sito
DiaporamaDiaporamaDiaporamaDiaporamaDiaporamaDiaporamaDiaporamaDiaporamaDiaporamaDiaporamaDiaporamaDiaporamaDiaporamaDiaporamaDiaporamaDiaporamaDiaporamaDiaporama
Accueil
ICube   >   Agenda : Thèse : Jelica Vasiljević : Generative Adversarial Networks in Digital Histopathology : Stain Transfert and Deep Learning Model Invariance to Stain Variation

Thèse : Jelica Vasiljević : Generative Adversarial Networks in Digital Histopathology : Stain Transfert and Deep Learning Model Invariance to Stain Variation

22 settembre 2022
13h30
Campus d'Illkirch - A301

Jelica Vasiljević invite you to the defence of his PhD thesis entitled “Generative Adversarial Networks in Digital Histopathology: Stain Transfer and Deep Learning Model Invariance to Stain Variation”.  It will be held at 13:30h on 22nd September in room A301 (Campus d'Illkirch), followed by a reception in the cafeteria.

The jury will be composed of the following:

Thesis Supervisors

Cédric WEMMERT, Professor, University of Strasbourg, France

Srdjan STANKOVIĆ, Professor Emeritus, University of Belgrade, Serbia

Rapporteurs 

            Xavier DESCOMBES, Research Director, INRIA Sophia Antipolis, France

Maja TEMERINAC OTT, Professor, Hochschule Furtwangen, Germany

Examinators

            Odyssée MERVEILLE, Lecturer, INSA Lyon, France

Sarah LECLERC, Lecturer, University of Bourgogne, France

Invites

Thomas LAMPERT, HDR, University of Strasbourg, France

 

For those who cannot physically attend, it will be possible to follow the defence by videoconference via the Zoom application:

https://cnrs.zoom.us/j/94206932195?pwd=UitFYjlzQU9RVDVWNFFSSkk0RHE1Zz09

Meeting ID: 942 0693 2195

Password: xkn11A

 

Title: Generative Adversarial Networks in Digital Histopathology: Stain Transfer and Deep Learning Model Invariance to Stain Variation

Abstract: Current state-of-the-art deep learning methods are data-hungry approaches which require huge annotated data collections to perform well. Nevertheless, digital histopathology, like the other fields of the medical domain, is known for its scarcity of data. Moreover, considering the variations that can occur due to the staining process and staining protocols, already collected and annotated datasets can only be reused with limited success. Such stain variation represents a source of domain shift and significantly affects deep learning-based solutions in practice.  This thesis investigates the potential of Generative Adversarial Networks (GANs) in two directions for addressing these problems --- stain transfer to enable reusing already available data collections; and developing stain invariant solutions which would alleviate the need for additional data acquisition or annotations.

Keywords: deep learning, Generative Adversarial Networks (GANs), segmentation, digital histopathology

Date: 22nd September, 13:30h, A301 + cafeteria.

À la une

Le dépôt des candidatures pour les postes d’enseignants-chercheur est ouvert. Les offres sont...

Flux RSS

Flux RSS