Abstract - Behavior (movement patterns) of animals has been used for a long time in medical sciences as an indicator of the "system state" i.e. of the influence of a particular treatment upon the animal's condition. Until recently the evaluation of such behavior was done visually, by the experimenter, without any possibility of an objective measurement procedure. The introduction of the passive marker based motion analysis in controlled laboratory conditions led to the reproducible generation of observables, that can be quantified and further processed. The selection of an adequate model and the appropriate signal processing constitutes the major problem. Animal behavior should be described in two levels requiring two models. The quantitative (low level) model is used to describe the physical movement. In the paper we present the measurement setup, the low-level model and the conclusions drawn from several long-term measurements. The psychophysical state of an animal can be characterized with the help of the qualitative (high level) model that is to be established with the involvement of medical experts.

The human image processing ability is excellent - as far as static images are concerned. However, a great number of paintings and statues show animals (especially horses) in such positions that do not exist in reality. [1] gives such examples. Imagine a horse in slow motion lifting simultaneously both of its legs on the same side! This proves that human vision is imperfect when the details of a moving object should be observed.

Passive marker based motion analysis provides the experimenter with the possibility of capturing movements objectively without causing substantial discomfort to the animal.

The aim of our research work has been to characterize the psychophysical state of small animals (rats) based on their movement patterns. Such a complex indirect measurement can be partitioned into two parts [2]:

The (b) operation requires the definition of appropriate parameters that characterize the physical state of the animal.

We continuously tracked the movement of rats kept in special transparent (plastic) tube-like cages for a few days. There were two groups of rats, one kept in cages tilted by 45 degree and one, the control group, kept in horizontal cages. Based on the measurement data the aim has been to characterize the movement patterns of normal rats, i.e. animals without any medication and known illness.

In the Experimental Research Department - Second Institute of Physiology, Semmelweis University of Medicine, special tube-like cages were developed and constructed to study the chronic adaptation of the vascular system of rats to static load. The cages are tilted by 45° . The rats are in head-up position and they can move back and forth and turn around the longitudinal axis of their trunk freely but they cannot turn to head-down position. The femoral vein pressure of the animals in the tilted cages approximately doubles as a result of the gravitational load. Femoral arterial pressure is not altered by tilting. The increase in the femoral vein pressure results in physiological changes of the veins. The cages are 60 cm long, a ladder is incorporated to aid movement up and down. The animals can eat and drink at one end (the top one for the tilted cage) of the cage. Details are given in [3].

The movement pattern of rats in the tilted cages should be compared to that of a control group (rats kept in similar but horizontally placed cages). The movement of the rats were recorded for 8 days continuously in one experiment. Three rats were put into horizontal and three rats into tilted cages.

The animals rapidly adapted to the new environment. They played and groomed themselves during the daily periods of rest (approximately one hour) in free-roaming cages while the tube-like cages were cleaned. The animals did not show escape reactions when being placed back into the cages. The food and water intake can be considered normal, no substantial weight loss was observed during the 8 day periods. These parameters were measured daily.

The aim of the experiment was to elaborate the necessary feature extraction in order to characterize the movement patterns of normal rats with a few parameters only, based on the position-time functions of the passive reflecting markers attached to anatomical landmark points. These parameters are intended to be used for later classification of animal movement patterns deviating from normal as a result of illness or medication. During the feature extraction process we have been searching for parameters that are similar for both groups (rats in tilted and horizontal cages) and also for parameters that are different.

We used the PRIMAS precision motion analysis system developed at Delft University of Technology. The maximal sampling rate of the analyzer is high (100 frames per second) and the CCD sensors in its cameras have good resolution (604 x 288 pixels, non-interlaced). As a result of sub-pixel marker image center estimation, the resolution of the system is 1:6000 both horizontally and vertically when applying usual sized markers. Accuracy cannot be specified simply as it depends also on the camera arrangement. Passive marker based motion analysis is especially suitable for biomedical applications as the lightweight (<10 gram) markers cause negligible discomfort for the examined person or animal.

For further details, refer to [4]. Some biomedical applications of the PRIMAS system are described in [5].

We decided to use disk shaped markers which could be glued to the skin of the rats. The markers were placed to aid the acquisition of the following parameters:

Four measurement series were completed. The first experimental series with two animals lasted for two days. Based on this the camera placement was modified as well as the marker pattern, that was finally set to the one shown in Fig.1. The second and fourth measurement series with 6 animals lasted for 8 days. After the first 4 days the missing markers were restored. The third measurement series with 6 animals lasted for nearly 7 days. In this paper we report basically on the data gained during the second measurement series.

The aim of low-level data processing was to generate the following time functions:

A. Markers cannot always be seen.

The animals try to get rid of the reflective disks and they can scratch some off. The rats sometimes push their heads downward between their body and the wall of the cage or even under their trunk hiding the head markers. This position is quite common during sleep for rats being in tilted cages.

The animals sometimes tuck their heads into the dead-space caused by the feeder.

The animals can twist their trunk causing skin wrinkles. If the marker is on a wrinkle its projection towards the camera can be too small to be detected.

B. Ghost markers appear.

They might be the result of a reflecting object other than a marker, e.g. metal surfaces of the cages. This is eliminated at the beginning of the experiment, however the daily cleaning process may result in the appearance of ghost markers.

C. Missing intervals.

The daily cleaning interval must be identified and cut off from the recording.

D. Insufficient modeling.

The rigid body assumption of the trunk cannot be justified. The upper part of the trunk can be twisted along its longitudinal axis by more than 90° compared to the lower part.

To cope with these problems, the following rules have been applied in low-level data processing.

If two marker images are too close to each other, they are united into one image, taking the average of the midpoint co-ordinates. This happens when the animal reclines its head on one side so that two head markers are seen at the same time. Another possibility is that the image of one marker is split into two. This might be caused by the uneven light reflection from the surface of the marker.

Ghost marker images are static images causing typical distortion of the position of a rat. Filtering ghost images thus was relatively easy.

Estimation of the rotation was done after the head markers had been eliminated from the set of identified markers. The remaining trunk markers were processed by using template matching technique.

Position (in the tube) of a rat is unambiguous once the rotation is determined. In case the rotation cannot be determined the average of the co-ordinates of the trunk markers are taken as the estimated position.

In the forthcoming we narrow down our investigation to the generation and evaluation of the position-time functions. The following further filtering rules have been applied to the output data of low-level processing.

The rats can remove their head markers the least. Thus, as the experiment progresses it is not uncommon that only the head markers can be detected. The distance between the trunk and head markers must be taken into account. Some further processing algorithms, especially the one that calculates cumulated movement (total distance covered by an animal), are insensitive to the offset when a head marker is mistakenly handled as if it were a trunk marker.

The rats usually rest close to the feeder. If the calculated position happens to be too close to the feeder this indicates that the position was determined wrongly: taken the head marker as if it had been on the trunk.

A typical, well identifiable situation is the head-trunk oscillation. The head and trunk markers appear alternatively on consecutive frames. We suspect that this is caused by the trembling of the rats. Taking the average of a few marker positions originating from consecutive frames we get a virtual position thus eliminating the error that would be caused by the oscillation during the calculation of total distance covered by the animal.

Based on the evaluation of the measurement data taken during the first experiment, we determined a limit for the maximal speed of the rats. This helps to indicate the too fast changes in the position that result from the erroneous position determination. When markers are completely missing on some consecutive frames then the position last detected is held.

Data processing was accomplished by C language programs in LabWindows environment and by using MATLAB. The results are partly the reconstructed position-time functions as primary data and partly some derived parameters that reflect the similarities and the differences in the movement patterns of the two groups, being in tilted and in horizontal cages. Fig.2. characterizes the movement of rat 1 being in a horizontal cage during an 8 day experiment. The top curve is the position-time function. The middle curve demonstrates the total distance covered by the animal. Each bar represents the distance covered within an hour. The position-time functions are typical for all the rats (20) in our experiments except one, rat 5, being in a tilted cage, that behaved differently from the others.

The bottom curve shows the averaged light intensity, one bar stands for one hour.

The following conclusions have been drawn.

The position-time function of the rat 5 is dissimilar from those of all the other rats. On the contrary, based on the hourly cumulated movement, this animal cannot be distinguished from the others.

The daily periodicity in rat movement can be proved by correlation analysis. Fig.3. shows the cross correlation of the light intensity-time function with the cumulated movement-time functions of all the six rats for the 192 hours of the experiment. The movement activity of rats 1, 3, 4 and 5 correlated well with the light intensity. Rat 2 died on the forth day of the experiment. Rat 6 wounded its nose on the first day and this slowed down its movement activity. As it is clear from the figure, the animal recovered by the third day.

The bottom curve is the autocorrelation function of the light-time function. In the figure biased correlation functions are given.

The position-time functions were analyzed with the MTRCHAOS and MTRLYAP free-ware programs, released by Michael T. Rosenstein in 1993. No chaotic features have been found.

With the help of a passive marker based motion analyzer long-term rat movement could be investigated as never before. Filtering the raw data is essential, pattern matching and rule-based processing is applied. The evaluation of the position-time functions is completed, our conclusions are based on it. Methods to determine the other two parameters, trunk rotation and head-trunk distance are being developed.

We found the rat movement under the given circumstances being a stochastic phenomenon. The integral features characterize the behavior of rats. There is a daily periodicity in distance covered cumulated for longer time intervals (1 hour ... 3 hours). The position distribution histogram is similar for an average normal rat both in the tilted and in the horizontal cages. The total distance covered by the rats in horizontal cages is significantly greater than of those in the tilted cages. The work of the animals in the tilted cages devoted to change their positional energy is negligible compared to the work devoted to change their movement energy.

This work was partly funded by the COPERNICUS project CIPA 351OCT937845 and the OMFB project 94-04057.