3. Hardware, Image Parameters and Common Issues
On principle, any USB camera is suitable for pupillometry using MEYE, even though not all the cameras can image the eye with the proper resolution and at the right distance. This depends on the objective mounted on the camera. We prefer to use M12 USB cameras because they are generally cost-effective, and you can use the same camera on different setups, just changing the objective. Another solution is to use a varifocal lens, so that focus, zoom, and aperture can be adjusted as required. For mice, we use a 22 mm lens. For humans, we use a varifocal lens. Cameras with higher resolutions require less strict requirements for the lens because the eye is composed of a higher number of pixels than cameras with lower resolutions. Our advice is to zoom until the eye is composed of a square with at least 128x128 pixels. We mount cameras on manual micrometric manipulators so that the camera position can be adjusted accurately from subject to subject. Many other possible solutions are also adequate.
The most important factor in obtaining the best performances is to use the right illumination. There are plenty of possible solutions to illuminate the eye properly in infrared light. For now, we employed only IR-LED because they provide a stable, invisible, and powerful light. Infrared light illuminators commonly mount two different kinds of LED:
- 850nm the most frequently used in CCTV applications. The majority of the cameras perform very well using this kind of illuminator. Even non-night vision cameras have a certain degree in detecting this light, particularly after the removal of IR-cut filters. One of the downsides of this kind of LED is that they produce a faint red glow in the visible spectrum that is detected by the subject, and it can interfere with some experiment. We use these LEDs in almost all kinds of experiments.
- 940nm These kinds of LED do not emit any detectable visible light, but not all cameras detect this wavelength. For this reason, before buying a 940nm illuminator, double-check if the camera is suitable for these wavelengths. We use these LEDs only in experiments in which the subjects wear VR headsets.
Independently of the wavelength used, a sufficient amount of light is critical to obtain stable and reliable performance. This is achieved by tuning both exposure time and LED power. Generally, we prefer to change image exposure instead of modulating light power.
Another possible factor in achieving better results is using LED matrices. This is because tiny light sources tend to produce powerful corneal glint on the eye, which sometimes can interfere with pupil detection. We trained CNN to ignore glints, even though occasionally you can experience artifacts given by this reflection. The advice is to keep the glint away from the pupil, if possible.
Our advice in humans is to use Chinrests. Even if, in principle is possible to perform pupillometry with non-head-fixed subjects, the best practice is to stabilize the head using Chinrests. There are a lot of cheap solutions, like this or this.
We tested USB M12 lens near-infrared (850nm and 940nm) webcams. These webcams are really cheap and can be found easily online. The setup used in humans is composed of:
- USB Camera (Remove or cover the light sensor on this camera to keep active the IR LEDs)
- Varifocal Objective
- IR Light Source (Remove or cover the light sensor on this illuminator to keep active the IR LEDs)
- IR Power Supply
- (OPTIONAL) camera tripod.
For stable recordings use a head fixation system like one of these:
- Chin rest on humans
- Head fixation system on mouse
This issue is given most probably by the non-sufficient illumination of the eye. One possible solution is to adjust the illumination of the eye or the exposure of the camera. If this is not possible we suggest using the image enhancement tools, modulating luminance, contrast, and gamma. In 2-photon experiments try to invert the image colours. If this is not possible or does not resolve the problem, you can try to change the prediction threshold. This tool can be found on the left side of the GUI, under the PREDICTION menu (see below).
This is a common issue that we resolved using mathematical morphology. If you experience this issue, try enabling the checkbox "morphology" and/or adjusting the threshold. These tools can be found on the left side of the GUI, under the PREDICTION menu.
For now, there is no way to perform offline pupillometry on .avi files using the web app. This is due to HTML5 programming policies and does not depends on us. There are two possible workarounds for this problem:
- Using the Python 3 Jupyter Notebook for offline pupillometry: pupillometryOfflineVideos.ipynb.
- Converting the .avi file to another HTML5 supported format.
It is necessary to ensure enough amount of RAM dedicated to MEYE (at least 8GB). The first piece of advice is to close all other tabs and applications before running any experiment. You have to be careful with the use of background processes such as antivirus etc, etc
Real-time Plots heavily influence frame rate and frame stability. The number of points in the plot will increase over time heavily affecting computer RAM & performances. This means a continuous frame drop over time. To overcome this issue you should enable the Show only option, which allows plotting only a stable number of points per frame. This mode will affect the maximum framerate, but the frame rate will be stable.
DeFocusing pupillometry.it page will stop MEYE for security reasons. The advice is to keep MEYE focused and run experiments on another computer. In case you need to sync stimulation with pupillometry, you can use Triggers.
If you experience other issues, please feel free to contact us. We would be happy to help you.