The post holder will carry out developments (research project) in deep learning for medical image analysis, to push the performance limits of deep diagnosis and prognosis models of brain pathologies based on multimodality imaging we have designed so far.
Performance of such architectures, indeed are highly impacted by the high dimensional, low sample, imbalance datasets whose annotation levels may be weak and uncertain. Specific studies may include -evaluating atypical loss terms, that may leverage more information from the data than standard loss terms, such as cross-entropy, -leveraging curriculum learning, indeed shown promising results in the domain of medical image analysis, -implementing interpretation tools such as attention networks to provide visual insights about the origin of the model predictions.
The post holder will combine responsibilities for proposing innovative methods leveraging the state of the art method as well as developing processing pipelines. Applicants should be proficient in the theoretical foundations of deep learning as well as their implementation with standards libraries such as PyTorch. The finer objectives of the project will be defined and prioritized according to prevailing methodological challenges to tackle at the beginning of the project and according to the candidate's experience.
This position is embedded within an innovative clinical project funded by the french National Research Agency (ANR) whose aim is to understand the physiopathological mechanisms of consciousness disorder and design efficient diagnosis and prognosis tools of patients being in acute coma based on multimodality imaging. The successful candidate will be part of the project team and have access to a unique patient database.
Please see attached file for a complete description.
Deadline for application is July 9 2021.