I am working as a Data Scientist for the Brain Functional Genome Program at the Paris Brain Institute. My mission is to develop and maintain a pipeline to put together and analyze 3D lighsheet microscopy images of a brain sample and spatial omics data for slices of the same sample.
I developed a colocalization module for the ClearMap package. It allows to see which of the objects (typically cell nuclei) detected in different channels of a volumetric image are spatially related.
I am developing a 2D slice / 3D volumetric image alignment software in PyTorch (not released publicly, for the moment). It recovers the position of a slice of brain in an image of the whole brain. It is particularly challenging due to tissue deformation in-between the 3D imaging and the 2D imaging. A promising loss function we have is based on the SIFT detector/descriptor paradigm.
I also contributed to Kornia with a differentiable implementation of the Mutual Information Metric.
I use the optimization framework of PyTorch, with a custom Optimizer, based on the Riemannian geometry of a space of slice parametrizations.
The alignment is feasible only if the signals are comparable, this entails the need for style transfer.
Motivated by the need for assessment of the alignment quality, I developed a benchmarking package for predictive models.
Our team records the history of our tissue samples and of the underlying data in an integrated database through a Django web application I co-develop with Élise Liu. (not released publicly, for the moment)