Key words: mice behavior modeling and recognition, computer vision, machine learning, autism spectrum disorder, neuroscience.
Location: LIS, IBDM, INMED / Marseille
Supervisors: Séverine DUBUISSON, Laurent FASANO, Françoise MUSCATELLI
Summary:Machine learning and computer vision (CV) tools allow automatic detection and quantification of mice behaviors, including mouse models of autism spectrum disorder (ASD). The project aims at improving the Live Mouse Tracker system (LMT-https://livemousetracker.org) to develop new, objective behavioral metrics, which is key for understanding neural systems and behavior relationship. Although LMT allowsthe automated monitoring of individual mice housed socially and reporting of individual behavior, a combined effort between neurobiologists and CV scientists is needed for acquiring, processing, analyzing and extract new information from images. The successful candidate with background in programming knowledge and CV is expected to develop and implement new modules allowing robust measurement of i) repetitive behaviors and ii) complex social interactions between multiple pairs of wild type and ASD animals, in particular during infancy.
The goal of our work is to analyze top-view videos acquired by a Kinect 2 (providing depth and color images) of mice interacting each other. Indeed, the current need in phenotyping is to measure automatically in a detailed way, social and non-social behaviors of individuals within a group. For that purpose, we will use Live Mouse Tracker (LMT), that integrates a framework for the behavior analysis of a small group of mice (1-3), and we will extend its functionalities to larger mice group analysis. We also will provide a model for repetitive motion that will be used to detect stereotypic behavior characterizing ASD. This study will include a statistical analysis of the quality of interactions between mice as well as individual events, group events by exploiting trajectories. Finally, by learning the spatio-temporal information of extracted trajectories, we will identify and study differences in the phenotypes expressed by individuals in group-housed mice. This will improve our understanding of the neurobiological basis of behavioral abnormalities in neurodevelopmental diseases.
Objectives: Our goal is to automated extraction of core ASD behavior, social interaction deficits and stereotypies, in the context of animal group interaction. While challenges still remain, we believe that the development of new modules will significantly advance our knowledge on behavioral differences between wild type and ASD-like mice. The availability of 4 mouse models of ASD will be the ground of this study. The candidate will assemble the LMT system and develop new Python analysis scripts that will be shared with the LMT community.
This collaborative project brings together researchers in computer science specialized in image processing and shape recognition (LIS research laboratory) and neurobiologists expert in brain development, circuitry development and function (IBDM & INMED). The three labs are interested in ASD; Scientists at the LIS, use ML and CV technology to aid better ASD diagnosis and scientists at the IBDM & INMED used mouse models of ASD. The three labs are already equipped with computers for big data storage and processing during longterm experiments. With the expertise of the LIS, we will be able to develop software to provide additional extracted features to document behaviors in an environment allowing maintenance of social interactions over a long period of time. This project will also train the PhD candidate to develop an interdisciplinary expertise in computer science applied to neurobehavioral imaging.
Expected profile : We expect a candidate with a background in computer science (a high level in Python programming is required) and strong skills in image and video processing for tracking and behavior analysis. The candidate will be involved from LMT building to development of imaging software.
Contacts: Send your application with a CV and cover letter to:firstname.lastname@example.org
(c) GdR 720 ISIS - CNRS - 2011-2020.