Postdoctoral Position in Federated Deep Learning for Stroke Imaging (CHUV)
Monday 30th November 2020
The Lausanne University Hospital (CHUV) is one of five Swiss university hospitals. Through its collaboration with the Faculty of Biology and Medicine of the University of Lausanne and the EPFL, CHUV plays a leading role in the areas of medical care, medical research and training.
The Radiology Department has a strong research focus, with several groups dedicated to advancing MR acquisition, improving image processing and machine learning for radiology, as well as radiologists that are very active in clinical research. The department is also part of the Center for Biomedical Imaging, a joint undertaking of the EPFL, University of Geneva, University of Lausanne, and Geneva University Hospital, and enjoys regular collaborations with these institutions.
The three-year Advanced Stroke Analytics Platform (ASAP) project aims at developing novel deep learning techniques to drive forward the utility and use of magnetic resonance imaging for acute stroke, and will be led in close collaboration with experts in deep learning for radiology and stroke at the Support Center for Advanced Neuroimaging, Bern University Hospital, and the Advanced Clinical Imaging Technology group of Siemens Healthineers, itself a joint R&D group with the CHUV Radiology Department.
Deep learning algorithms are now a staple of medical imaging. While some network architectures can generalize surprisingly well even in the low-sample regime, the key to high performance and convincing evaluation, in particular in a clinical context, remains dataset size. However, acquisition in medical imaging is costly, and data sharing remains a difficult proposition due to ethical and privacy constraints. In this project, we will leverage recent advances in algorithms and tooling to develop federated deep learning algorithms that enable joint training of high-capacity deep neural networks while allowing each hospital to keep their data private. Our focus will be on developing robust and explainable 3D/4D segmentation models and GANs applied to stroke imaging, where both the CHUV and Bern hospitals have very high expertise and available data (several thousands of patients). Because stroke imaging workups rely on several protocols, an important and exciting challenge will be to develop algorithms that can deal with spatio-temporal differences in images between hospitals in terms of contrast, voxel size, slice thickness, repetition time, etc. Effective solutions to these challenging domain gap problems would also constitute important contributions to machine learning at large.
The candidate will be expected to develop and implement new algorithms, present work at conferences and in journal publications, collaborate with local and international researchers, help supervise junior researchers including PhDs and Master students, participate in grant submissions, and to interact fruitfully with radiologists, clinicians, and our industrial R&D partners.
PhD in computer science, electrical engineering, biomedical engineering, statistical physics, statistics, or related field
- Good training in linear algebra, calculus, statistics
-Good knowledge of Python and relevant deep learning packages such as PyTorch or TensorFlow
- English proficiency, French knowledge an asset
- Demonstrated previous experience in machine learning and deep learning is required Experience in medical imaging, optimization, distributed computing, signal processing, statistics are all an advantage.
You will be part of a team of scientists from multiple partners, which requires a strong team spirit and professionalism. Excellent communication an inter-personal skills are as important as technical skills. This project will also require creative spirit and the ability to work autonomously.
Nous offrons :
If you become an employee at the Centre Hospitalier Universitaire Vaudois, we will offer you the following :
- High social benefits
- Regular salary progression according to job responsibilities
- Three days of training per year
- 25 working days of vacation per year
- Very good restaurants with preferential rates.
Contact for information on the function : Monsieur Jonas Richiardi, Responsable de recherche : 021 314 15 27 or firstname.lastname@example.org
|Title||Postdoctoral Position in Federated Deep Learning for Stroke Imaging|