Annonce

Les commentaires sont clos.

PhD offer: robust prostate cancer detection in multiparametric MRI using deep learning approaches

9 Juin 2022


Catégorie : Doctorant


Robust prostate cancer detectionin multiparametric MRI using deep learning approaches

Scientific context

Multiparametric magnetic resonance imaging (mp-MRI), which combines T2-weighted imaging with diffusion-weighted, dynamic contrast material–enhanced, and/or MR spectroscopic imaging, has shown promising results in the detection of prostate cancer (PCa). However, characterizing focal prostate lesions in mp-MRI sequences is time demanding and challenging, even for experienced readers, especially when individual MR sequences yield conflicting findings. There has been a considerable effort, in the past decade, to develop computer aided detection and diagnosis systems (CAD) of PCa cancer as well as prostate segmentation.

The vast majority of developed CAD models focus on characterizing prostate lesions aggressiveness in mp-MRI sequences [1]. Despite an important improvement brought by CAD systems with recent deep learning approach [2], simple detection of clinically significant (CS) cancers is still impossible to use in clinical practice. For some patients it is difficult to have a robust diagnosis without being mistaken. In addition, deep learning methods raise the problem of data harmonization. The diagnosis of the CAD system will be less precise and robust when its will be deployed in different health centers.

In this work, we propose to investigate new robust methods for the detection of prostate cancer. The objective is to enable a better deployment of CAD solutions based on deep learning in clinical practice.

 

Job description and missions

First in this thesis, we want to explore new deep learning architecture that supervises the CAD system, itself based on a deep learning approach, to estimate the truthfulness of the diagnosis. This indicator will be used to define if a prediction made by the CAD system is reliable enough. If not, a proofreading by experienced readers will be required.

A second part of this thesis will be devoted to working on an innovative strategy to adapt an existing pre-trained model to the specificity of local data from another health institute. Especially considering images from different MRI system and acquisition quality. Methods such as transfer learning, few shot learning or style transferwill be explored. An additional goal is providing a scalable CAD system that continues to learn from new data, either from a different facility or by continuously adding newly diagnosed patients.

Finally, the last part of the thesis will be focused on the validation and the evaluation of the different proposed methods. This will be done by using public data(such as PI-CAI https://pi-cai.grand-challenge.org) but also using private datacoming from the Brest university hospital center and the Toulouse cancer institute.

 

Profile

PhD in computer science, image processing, AI, applied mathematics, data scientist. Good programming skills is an important requisite, especially in python. Autonomy, open-mindedness and motivation, as well as good English speaking/writing skills, are also expected. Some experience in deep learning is appreciated.

 

Position context

The thesis will join the INSERM UMR1101 Laboratory of Medical Information Processing (LaTIM, Brest, France, https://latim.univ-brest.fr). The future student will work in collaboration with different academic, hospital and company partnerswithin the context of a national projects. Access will be given to the computer cluster PLACIS (http://placis.univ-brest.fr/english) and to clinical data from our partners.

 

The thesis should ideally start on November or December 2022, for three years.

 

Contact and additional information

For application, a CV must be sent to the following e-mails:

Julien Bert (julien.bert@univ-brest.fr)

 

References

[1] A. Duran, G. Dussert, O. Rouvière, T. Jaouen, P.-M. Jodoin, and C. Lartizien, “ProstAttention-Net: A deep attention model for prostate cancer segmentation by aggressiveness in MRI scans,” Medical Image Analysis, vol. 77, p. 102347, Apr. 2022, doi: 10.1016/j.media.2021.102347.

[2] A. Saha, M. Hosseinzadeh, and H. Huisman, “End-to-end prostate cancer detection in bpMRI via 3D CNNs: Effects of attention mechanisms, clinical priori and decoupled false positive reduction,” Medical Image Analysis, vol. 73, p. 102155, Oct. 2021, doi: 10.1016/j.media.2021.102155.