observing behaviour by computer

Share this on social media:

Tom Wilkie spoke to Lucas Noldus, whose company produces computerised systems for tracking and observing animal behaviour. The applications, he found, stretch from new drugs to functional genomics

According to evolutionary theory, human beings developed big brains so that they could observe, interpret, and respond appropriately to the behaviour of other humans and other animals. Our ability to understand the behaviour of others is what has made us such supremely successful social animals.

Paradoxically, however, when it comes to the scientific study of behaviour, humans are terrible observers. Precisely because our brains do so much interpreting and pre-processing, our observations are not to be trusted. This is obviously true of observations of other humans, but it holds good for laboratory studies of animal behaviour too. Humans are subjective; and they are not repeatable in their ways of observing. In addition, people get tired and they tend to want to work office hours, whereas many animals whose behaviour is the subject of scientific study are crepuscular or nocturnal and exhibit normal behaviour only at night.

The field of animal behaviour studies, therefore, has long needed techniques to remove the subjectivity of human observations and replace them with automated, reproducible technologies that reduce the chaotic outward appearance of behaviour to organised data for statistical analysis. The problem is two-fold: first one has to have some way of recording the animals' movements, then one has to have some sort of software to make sense out of them.

The technology for recording animal behaviour has evolved dramatically over the past decade. In the early days, systems could track only one animal at a time, and that in a highly artificial environment - the animal had to be in an enclosure that was empty of anything other than the animal itself. The sorts of detectors that were employed included a grid of infrared beams or strain gauge transducers under the floor to estimate the animal's position. To get a crude estimate of movement, one researcher even resorted to placing a lab rat on a bass loudspeaker and monitoring the loudspeaker's electrical output as its cone was deformed by the rat moving across it.

Other detection methods have included ultrasound and even radar, which can detect very small movements - particularly important in studies on insect behaviour - and which has the advantage that it will work in complete darkness.

Video tracking systems were introduced in the early 1990s, offering clear advantages of flexibility, spatial precision and accuracy over the other hardware devices. However, with early systems the experimenter had to follow the path of the animal with the computer mouse or a joystick, and thus did not really get much beyond manual data entry.

Another early method, still used in some commercially available systems, is to feed the analogue video signal to a dedicated video tracking unit, which detects peaks in the voltage of the video signal (indicating a region of high contrast between the tracked animal and background), and uses this to produce the x, y coordinates of the tracked animals, which is then fed to a computer. Such systems are relatively inflexible and can normally track only one animal.

Greater flexibility is achieved by the use of frame grabbers to digitise analogue video signals. This enables high-speed data acquisition, and therefore tracking of animals that are moving relatively fast. For the past decade, Noldus Information Technology, a Dutch company that specialises in this area, has been developing and marketing its system, called 'The Observer', which allows researchers to index the frames of a video recording and code the recorded behaviour. Users can filter and select video episodes for further analysis.

In parallel, the company has also developed EthoVision, a video tracking system that combines automated recording of activity, movement and interactions of animals with image processing, track analysis and behaviour recognition software. It has recently been completely redesigned, with a graphical user interface, and the staged release as a Windows-compatible product was completed earlier this year. The user can design the arena, programming regions of special interest into the software, and also setting triggers so that, eventually, the software will have the capability, if the animal spends more than a couple of seconds in a specified zone, of ensuring that the system automatically delivers a food reward. The system will track the animals automatically, measuring their behaviour and automatically creating computer files of data for further analysis.

According to Lucas Noldus, the founder of the company, the installed base now exceeds 400 organisations, of which more than 50 are pharmaceutical companies. He believes that pressure to introduce such automated systems will increase in future as a result of requirements by the regulatory authorities, such as the Food and Drug Administration, that everything has to be documented and has to be capable of being tracked back. There is a different sort of regulatory pressure which, he believes, will also add to the demand for such systems. Concerns for the welfare of laboratory animals mean that social animals are housed together, rather than each being isolated in an individual cage, and that they should be observed in their home cages rather than transferred into a laboratory environment. Such trends mean that there is a much more complex environment for the analysis system to deal with.

He said: 'With video-tracking systems, the limiting factor is the quality of the software algorithms that can extract meaningful information from the image. This led us from the Observer system to EthoVision, which can be used to automate social behavioural studies. There is pressure on software manufacturers to extract more information from the animal than just position in space, which used to be enough.'

Automated observation using video tracking is particularly suitable for measuring locomotive behaviour, expressed as spatial measurements (distance, speed, turning, etc.) that human observers are often unable to estimate accurately. Automated systems also allow the study of behaviours that occur briefly and are then interspersed with long periods of inaction.

Video tracking, for example, carries out pattern analysis on a video image of the observed animals to extract quantitative measurements of the animals' behaviour.

EthoVision offers a large number of parameters to analyse behaviour. To quantify activity or exploratory behaviour, total distance travelled, speed and time spent in zones, can be used. In studies of spatial orientation and task-related paradigms, track heading, meander, turn angle and other shape parameters can be measured. With more than one animal in an arena, inter-object distance and relative movement are used to quantify social interactions. Furthermore, bouts of movement or rearing can be calculated. EthoVision takes the analogue video signal from a CCD video camera recording the animals, digitises the signal using a frame grabber and passes it on to the computer's memory. The software then analyses each frame in order to distinguish the tracked objects from the background, on the basis of either their brightness (grey scale) or colour (hue and saturation values). Having detected the objects, the software extracts the required features - including the position of the mathematical centre of each object and its surface area. These values are written to a track file on disk. Calculations produce quantified measurements of the animals' behaviour. For instance, if the position of an animal is known for each video frame, and the whole series of frames is analysed, the average speed of an animal during an experiment can be calculated, or the distance between several individually identified animals. In addition, if certain regions are identified as being of interest (the centre and edges of an open field experiment, for example), the proportion of time spent by the animals in those regions can also be calculated.

The analysis can be fine-tuned with advanced data selection, filtering techniques and parameter settings. For each parameter, distribution measures can be calculated per animal or summarised for groups. For further analysis, the output can be exported to spreadsheets and statistical packages.

One of the most important areas in which automated behavioural observation is being employed is in toxicity testing of new drugs. For drugs intended to cure diseases of the central nervous system, there are no in vitro models on which to try and uncover adverse effects. But often other drugs - for example those dealing with the cardiovascular system - need to be tested for potential side effects on behaviour; toxicology is often neurotoxicology. In such cases, use of whole animals cannot be eliminated, but of the three 'R's in animal research - reduction, refinement, and replacement - 'we contribute to refinement and reduction by getting more information about one animal, using better measurement techniques,' Dr Noldus said.

As genomics moves on from concentration on single genes to a focus on 'functional genomics', studies of animal behaviour will grow in importance in genomics research. An animal's behaviour is, after all, the functional expression of DNA at the highest and most integrated level. Historically, the premier 'guinea pig' for genetics studies is the fruit fly, Drosophila. For the best part of a century, researchers have scanned millions of these flies looking for mutations. Sometimes these affect merely the colour of the eye; sometimes they can be bizarre, with a leg growing where there should be an antenna. At Cambridge University in England, however, they are now moving beyond the study of such 'structural' or morphological mutations to see if they can find mutations that affect the flies' behaviour. In particular, the researchers are looking for mutations that cause learning difficulties in the fruit fly. It involves scanning more than 2,000 strains of flies to try to see which one is a learning mutant. The task would be impossible were it not for automated tracking and analysis techniques. It is a clear example of how scientific progress depends on technological improvements and, in this case, an unexpected application of computers to research.

One of the important motivations for genetics research has been to try to understand human disease and the technique has also opened up ways of genetically modifying laboratory mice so that they become better 'models' of the human disease. Sometimes this is done by adding a human gene; sometimes merely knocking out the operation of an existing mouse gene will serve the purpose.

According to Dr Noldus, people are realising that 'for the increasing numbers of transgenic mice - knock-outs, for example - the endpoint of the modification is the behaviour of the animal and it leads to all sorts of unexpected and unknown behavioural patterns.

People have not been able to quantify these behaviours because they have not met them before. There is a big need for more refined testing tools to help understand the subtle behavioural effects of these modifications.'

In many cases, the behaviour of laboratory rats and mice can be reduced to the movement of the trunk and the head. However, more subtle behaviours, such as aggressive interactions, still defeat all automated observing systems and, according to Dr Noldus, researchers must turn to the videologging system, The Observer, to record 'who's biting whom, where'. Automated systems do not work on primates either - they are too complicated to automate because of their facial expressions and movement of the hands and so on. But for the mainstay of laboratory work, automated behaviour recording is a huge advance on manual systems. As Dr Noldus pointed out: 'Facial expression is vital in primates but no one has read facial expression in rats.'