Real-time analysis of the behaviour of groups of mice via a depth-sensing camera and machine learning

Abstract

Preclinical studies of psychiatric disorders require the use of animal models to investigate the impact of environmental factors or genetic mutations on complex traits such as decision-making and social interactions. Here, we present a real-time method for behavior analysis of mice housed in groups that couples computer vision, machine learning and Triggered-RFID identification to track and monitor animals over several days in enriched environments. The system extracts a thorough list of individual and collective behavioral traits and provides a unique phenotypic profile for each animal. On mouse models, we study the impact of mutations of genes Shank2 and Shank3 involved in autism. Characterization and integration of data from behavioral profiles of mutated female mice reveals distinctive activity levels and involvement in complex social configuration.

Simple Behavioral Analysis (SimBA) – an open source toolkit for computer classification of complex social behaviors in experimental animals

“…The classifiers generated for this manuscript use the DeepLabCut 27 default network architectures for poseestimation positional data, which the subsequent machine learning features are derived from. There are, however, many other effective open-source solutions for animal tracking and pose-estimation 20,22,69 and SimBA is agnostic to the tools used to extrapolate positional coordinates. SimBA also has an interface for DeepPoseKit 26 that permits a range of alternative neural network architectures for pose-estimation that may offer speed and accuracy advantages 70 .…”

“…To overcome this, a range of elegant open-source tools ( Table 1) that use various forms of computer vision with synchronized RFID-tracking data and/or depth camera or multi-camera 3D systems have been developed for automated and precise classifications of animal behaviour [19][20][21][22][23][24] . Such methods can permit online classifications in semi-natural environments and are a foundation for impending closed-loop monitoring and forecasting systems 25 in behavioural neuroscience, but do require significant investment in specialized hardware and a working knowledge of computer science approaches.…”

Deep learning based behavioral analysis enables high precision rodent tracking and is capable of outperforming commercial solutions

“…As behavioral analysis moves more toward video tracking as opposed to reliance on beam grids, recent developments in unsupervised behavioral identification approaches have widened the horizons of what was previously thought possible. Approaches that focus on the unsupervised identification and separation of behavioral patterns are beginning to reveal the true complexity and richness of animal behavior [9,10,13] , However the interpretation of the findings from unsupervised machine learning techniques are more difficult. Although impressive, the implementation and use of many of these unsupervised behavior recognition approaches is out of reach of many basic science labs that lack the necessary programming and machine learning know-how.…”

“…Recently, major advances in machine learning have given rise to the first descriptions of unsupervised analyses of behavior, which reveal the stunning temporal and structural complexity of rodent behavior [9][10][11][12][13] . However, these advanced analyses are challenging for many biology and behavioral research labs to establish, which probably explains why they have not yet been widely implemented by the behavioral research community.…”

Track-Control, an automatic video-based real-time closed-loop behavioral control toolbox

“…As discussed in a previous study (de Chaumont et al, 2019), long-term recording can be a problem for most available tracking programs, especially those requiring manual interference.…”