The DIVA (Data Integration and Visualization in Virtual and Augmented Environments) project aims to create environments for data treatment that leverages virtual reality (VR), human-data interaction and automated algorithms. It is a joint initiative between the Institut Pasteur and the Institut Curie. The proposed post-doctoral fellowship involves developing machine-aided segmentation approaches to volumetric imaging data. Key to our segmentation approach is the relation between user intervention and data treatment. It is tailored to handle data in limited quantities, possibly very noisy, and to allow rapid and robust learning of the required procedures to extract relevant information from data.
Development of the proposed segmentation tool will be based on an in-house software platform that is designed to load any type of image stack (e.g. microscopy image or MRI/CT scan) into VR environments. The platform generates entire scenes allowing real-time interaction with data (i.e. representation modifications, annotations and projections of data). Currently the platform is available in alpha version at both DIVA host institutes.
More information on applications of the DIVA project is available here.
We are proposing a one-year contract (renewable for an additional year) financed by the DIM ELICIT program for a motivated candidate. They will be expected to contribute to our efforts of designing segmentation procedures for noisy data coming from microscopy and biomedical images. The successful candidate will employ Bayesian pixel feature learning (within the VR environment) and amortized active membrane segmentation to user-data interactions. Strong experience in programming is required and previous experience in pixel classification and segmentation is recommended. The approach will be tested on large-scale multimodal images of neurons (e.g. via light-sheet and two-photons microscopy techniques) and, if times permits, on CT scans of breast and hepatic cancers. The successful candidate will be working directly with both designers and coders of the DIVA platform: Mohamed El Beheiry and Sébastien Doutreligne. He or she will work directly with various members of the team of Jean-Marie Lledo to validate performance on experimental data.
Figure 1: Demonstration tracing of a one month old adult-born neuron in the complete reconstructed hippocampus using an early prototype of the DIVA platform. The tissue was imaged using serial end-block imaging (SEBI) with a two-photon microscope (dimensions 2.5 × 2.3 × 3.1 mm).
The successful candidate will join a highly multidisciplinary lab of physicists and computer scientists who use computational approaches to study random walks of biomolecules in living cells and study the relationship between neural architecture and behavior in drosophila larva. Localized in the world-renowned Institut Pasteur, the successful candidate will have privileged access to multiple GPU clusters and custom platforms in the lab.
- C, C++ or C#
- Scripting (e.g. Python, Bash)
- Statistical Physics
- Inference and Numerical Optimization
- Image Processing
- Machine Learning Frameworks (e.g. PyTorch, TensorFlow)
- Computer Graphics (e.g. OpenGL, GLSL)
- General-Purpose GPU Programming (e.g. CUDA, OpenCL)
Institut Pasteur (Paris, France)
Interested candidates should contact Jean-Baptiste Masson (email@example.com) with their CV, a brief motivation letter and two letters of recommendation.