Abib Alimi

I am a research scientist currently transitioning into Artificial Intelligence (AI) research and developping a solide foundation in understanding AI Safety and Alignment challenges. My previous work delt mainly with Healthcare research. I was a Postdoctoral Research Associate in the Computer Science Department, at Princeton University where I worked on developing methods at the intersection of AI and Materials Science to tackle problems related to medical imaging.

I completed my Ph.D in Computational NeuroImaging, from Inria, University of Côte d'Azur, in France. I was jointly advised by Rachid Deriche and Samuel Deslauriers-Gauthier. My Ph.D research delt with the reconstruction of diffusion Magnetic Resonance Imaging (dMRI) signal of the brain, and its complementary analysis with 3D-Polarized Light Imaging (3D-PLI). I performed multiscale and multimodal image analysis between dMRI and 3D-PLI. In 2019 I received the ISMRM Magna Cum Laude Award.

CV  /  Github  /  LinkedIn  /  Email: abib dot alimi at gmail dot com

The main goal is to shed light on the generalizability of existing dMRI signal representations to different diffusion encoding parameters and brain tissue types. Our team propose a single-layer neural network to predict unobserved dMRI signal from different image acquistion strategies.

An analytical and fast method is introduced to elegantly compute the fiber orientation distribution function (fiber ODF) from high-resolution vector data provided by 3D-PLI of the human brain.

The rotational invariance of diffusion MRI signal is considered in a new spherical CNN model with fully spectral domain convolutional and non-linear layers to estimate white matter microstructure parameters.

From the analytical fiber ODFs reconstructed from 3D-PLI datasets, we compute probabilistic brain fiber tractography at different spatial resolution scales to close the resolution gap between 3D-PLI and diffusion MRI.

We propose an analytical method to compute the fiber orientation distribution function (fiber ODF) from high-resolution vector data provided by 3D-PLI.

A partial differential equations (PDE)-based mthod is proposed to reduce the noise from tilted measurements and disentangle the sign ambiguity of the inclination angle for a more accurate reconstruction and interpretation of the 3D-PLI fiber orientation.

We show how tractograms obtained at different spatial scales from 3D-PLI human brain datasets can be inspected using homology theory in order to perform a quantitative comparison between them.

We embed the specific diffusion MRI signal features (spherical representation, antipodal symmetry) to build new spherical convolutional neural netwoks and predict brain microstructure parameters and fiber FODs.

We present Diffusion Microstructure Imaging in Python (Dmipy), a diffusion MRI toolbox which allows accessing any multi-compartment-based model and robustly estimates these important features from single-shell, multi-shell, and multi-diffusion time, and multi-TE data.

This work explores, for the very first time to our knowledge, the potential of the time-dependent diffusion microstructure indices (qτ-indices) as accurate biomarkers in order to understand and efficiently treat myelin-related pathologies, in vivo and non invasively.

I have been a teaching assistant at University Côte d'Azur in:

Design and source code from Jon Barron's website