Réseaux de neurones convolutifs en médecine nucléaire : applications à la segmentation automatique des tumeurs gliales et à la correction d’atténuation en TEP/IRM

Abstract : Convolutional neural networks group together a set of architectures whose elementary units are inspired by biological neurons. They are able to estimate and extract internal representations (filters) by learning from annotated data. They are then convoluted with our images to classify or predict. Our objective through two examples is to try to better understand the potential of the latter in nuclear medicine and to understand their limits. We will first study a segmentation procedure that can be assimilated to a classification problematic where we try to predict for each voxel a category. For this we will rely on a cohort of 37 patients with glial tumors explored within initial staging by 18-F-Fluoro-ethyl-thyrosine PET. In a second time, we will focus on estimating CT scans from ZTE MRI sequences and to evaluate its impact on attenuation correction during tomographic reconstruction in 47 patients referred for neurodegenerative diseases.
Document type :
Master Thesis
Complete list of metadatas

Cited literature [67 references]  Display  Hide  Download

https://dumas.ccsd.cnrs.fr/dumas-02116080
Contributor : Université Paris Descartes - Scd <>
Submitted on : Tuesday, April 30, 2019 - 4:38:38 PM
Last modification on : Friday, May 10, 2019 - 10:08:27 AM

File

ThExe_BLANC_DURAND_Paul_DUMAS....
Files produced by the author(s)

Licence


Distributed under a Creative Commons Attribution - NonCommercial - NoDerivatives 4.0 International License

Identifiers

  • HAL Id : dumas-02116080, version 1

Citation

Paul Blanc-Durand. Réseaux de neurones convolutifs en médecine nucléaire : applications à la segmentation automatique des tumeurs gliales et à la correction d’atténuation en TEP/IRM. Médecine humaine et pathologie. 2018. ⟨dumas-02116080⟩

Share

Metrics

Record views

48

Files downloads

166