Skip to Main content Skip to Navigation
Theses

Parallélisation hybride d’une application de détection de noyaux cellulaires

Abstract : The shared memory parallelized version of the Marked Point Process algorithm allows to speed-up cells nuclei detection. However, the limitations imposed by the number of CPU cores or the memory capacity of GPU cards do not allow to analyse an entire histological image (50,000 × 50,000 pixels). To achieve this, we propose to add a distributed dimension to this parallelization using the hybrid Ordered Read-Write Locks model. In order to guarantee the validity of the original algorithm, we have implemented different strategies to ensure that all nuclei are processed and that local processing is considered on a global scale. The tests carried out first validated the scalability of the application and then showed an acceleration factor of 40 on 64 CPU cores.
Complete list of metadatas

Cited literature [53 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-02511544
Contributor : Abes Star :  Contact
Submitted on : Wednesday, March 18, 2020 - 8:37:07 PM
Last modification on : Thursday, March 19, 2020 - 1:32:11 AM

File

Salas_Daniel_2019_ED269.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-02511544, version 1

Citation

Daniel Salas. Parallélisation hybride d’une application de détection de noyaux cellulaires. Imagerie médicale. Université de Strasbourg, 2019. Français. ⟨NNT : 2019STRAD018⟩. ⟨tel-02511544⟩

Share

Metrics

Record views

117

Files downloads

69