Soutenance de thèse de Amine Amyar le mardi 31 Août 2021 à 14 h à l'université de Rouen Normandie

Date :

...
La soutenance de thèse de Amine Amyar aura lieu le mardi 31 août à 14h. Cette thèse, réalisée à l'Université de Rouen Normandie au sein de l'équipe Quantif du LITIS, en collaboration avec General Electric Healthcare  s'intitule : 
"
Deep Learning for Outcome Prediction in Cancer using Positron Emission Tomography Images"

La soutenance aura lieu devant le jury composé de:

* M. Dimitris Visvikis Dr, Inserm UMR, UBO Président du Jury 
* M. John Lee Pr, Université catholique de Louvain, Rapporteur
* Mme. Diana Mateus Pr, Centrale Nantes, Rapporteur  
* M. Pierre Vera PUPH, Centre Henri Becquerel, Rouen, Examinateur
* M. Vincent Morard PhD, General Electric Medical Systems Examinateur 
* M. Baptiste Perrin Directeur R&D, General Electric Medical Systems Membre invité
* Mme. Su Ruan Pr, Université de Rouen Directrice
* M. Romain Modzelewski PhD, Centre Henri Becquerel, Encadrant

Du fait des conditions sanitaires actuelles, la soutenance s'effectuera en visioconférence. Vous êtes donc invités à assister à la soutenance de thèse via les identifiants Zoom suivants : 

https://harvard.zoom.us/j/96936249997?pwd=STFKK2RvQ0w4VlBwTUFTRnBqL2RjUT09

Password: 562208

Join by telephone (use any number to dial in)
        +1 929 436 2866
        +1 301 715 8592
        +1 312 626 6799
        +1 669 900 6833
        +1 253 215 8782
        +1 346 248 7799

International numbers available: https://harvard.zoom.us/u/aSXEylzZ5

One tap mobile: +19294362866,,96936249997# US (New York)
    Password: 562208

Join by SIP conference room system
Meeting ID: 969 3624 9997
96936249997@zoomcrc.com


Summary

Radiomics is a research field where images are used for their potential in precision medicine. It is defined as the analysis of a large number of extracted features from medical images such as Computed Tomography(CT),Magnetic Resonance Imaging(MRI) orPositron Emission Tomography(PET). These features are used to uncover disease characteristics that fail to be found or quantified by the naked eye. The first step in radiomic analysis in oncology is the lesion segmentation, which is the process of isolating a Region Of Interest (ROI) from other regions with contours. After segmentation, thousands of features can be extracted from the ROI , and then the most relevant ones are selected. Finally, a machine learning algorithm such as Random Forest (RF) or Support Vector Machine (SVM)  is applied to identify the best relevant features that predict the outcome. This classical workflow is limited for several reasons: segmentation requires a highly trainable physician, is time consuming and the defined ground truth is physician subjective and prone to error (intra and inter observer variability). Secondly, the handcrafted features defined from the ROI are limited since they are heavily influenced by many factors like the used segmentation method. Therefore they fail when the  ROI is altered. 

The goal of this thesis is to go beyond the current radiomic paradigm which requires manual extraction of characteristics and replace it by deep radiomics. In our new approach, features are learned along with the prediction of the outcome. To achieve this, we develop different Deep Learning (DL) algorithms to create end-to-end architectures that take an image as input, learn feature representation and outcome prediction.