Flora ESTERMANN : Détection et segmentation d’anomalies en imagerie ultrasonore 3D par des approches faiblement supervisées ou non supervisées

from October 1, 2021 to September 30, 2024

Laboratories : CREATIS et LVA thesis director : Philippe Delachartre (CREATIS) supervisors : Philippe Guy (LVA) et Valérie KAFTANDJIAN (LVA)

Summary :
In some areas such as the medical field or in industry, we have to search for the presence of internal anomalies (not visible on the surface) and to characterize the anomalies detected. One way to access these anomalies is the imaging which can be 2D or 3D and use different modalities (X-ray, ultrasound, MRI ...).

We are interested in 3D ultrasound imaging applied to the medical field for the quantification of white matter lesions and to the industrial field for the diagnosis of faults.

The goal is to develop deep learning methods combining unsupervised and supervised approaches that take into account label uncertainty for the detection and segmentation of anomalies in ultrasound data. Besides, we will also try to verify that the developed methods can work on different modalities such as X-rays for instance (where anomalies are also characterized by a low contrast in a noisy environment).