Computational NeuroscienceAn automated system for the recognition of various specific rat behaviours

- Automated recognition of the rat behaviours 'drink', 'eat', 'sniff', 'groom', 'jump', 'rear unsupported', 'rear wall', 'rest', 'twitch' and 'walk' from top-view video.
- Recognition of 71% on par with human recognition rates.
- Validation of automated recognition performed on videos recorded with different resolution, animal strain, illumination, background and cage layout.
- Validation by means of an experimental study with drug treatment and comparison of automated recognition with manual scoring by an expert.

The automated measurement of rodent behaviour is crucial to advance research in neuroscience and pharmacology. Rats and mice are used as models for human diseases; their behaviour is studied to discover and develop new drugs for psychiatric and neurological disorders and to establish the effect of genetic variation on behavioural changes. Such behaviour is primarily labelled by humans. Manual annotation is labour intensive, error-prone and subject to individual interpretation.We present a system for automated behaviour recognition (ABR) that recognises the rat behaviours 'drink', 'eat', 'sniff', 'groom', 'jump', 'rear unsupported', 'rear wall', 'rest', 'twitch' and 'walk'. The ABR system needs no on-site training; the only inputs needed are the sizes of the cage and the animal. This is a major advantage over other systems that need to be trained with hand-labelled data before they can be used in a new experimental setup. Furthermore, ABR uses an overhead camera view, which is more practical in lab situations and facilitates high-throughput testing more easily than a side-view setup.ABR has been validated by comparison with manual behavioural scoring by an expert. For this, animals were treated with two types of psychopharmaca: a stimulant drug (Amphetamine) and a sedative drug (Diazepam). The effects of drug treatment on certain behavioural categories were measured and compared for both analysis methods. Statistical analysis showed that ABR found similar behavioural effects as the human observer. We conclude that our ABR system represents a significant step forward in the automated observation of rodent behaviour.