Technical guide | Noise in Voltage-Sensitive Dye Imaging

Photon shot noise and Dark current noise

Video signals captured by electronic cameras such as CCDs and CMOS contain various noises. Of these, the one that has the greatest impact on imaging data is photon shot noise. Since the source of photon shot noise is light quantum, it is impossible to eliminate it.

The next contributor is dark current noise, which is caused by thermal fluctuations and electrical uncertainties. In optical mapping of membrane potential, dark current noise is often smaller than photon shot noise, so photon shot noise is the only thing you need to be careful about.

Only photon shot noise is important for membrane potential imaging

Photon shot noise occurs when light quanta are photoelectrically converted by an image sensor. For example, even if average number of light quanta that falls on one pixel in a given time is 10,000, there is a stochastic distribution, so light will fluctuate in proportion to square root of number of quanta.

In other words, square root of 10,000 is 100, so it generally fluctuates between 9,950 and 10,050. This is a physical phenomenon and does not change regardless of an image sensor. For this reason, when amount of change in light intensity is small, such as with membrane voltage-sensitive dyes, results are significantly affected. In order to accurately record a change of 0.1%, it can be said that at least 1 million photons are required per pixel.

Image sensor having large pixels and bright light improve S/N ratio

Therefore, an effective way to increase number of photons handled by one pixel is to increase amount of charge generated by photoelectric conversion that is stored in the pixel.

Since the capacitance that can be formed in a semiconductor is proportional to its area, it can be said that the larger the pixel, the more charge can be accumulated and the lower the noise.

In other words, if you use a sensor with large pixels and bright lighting, the noise performance will improve. Conversely, unless brightness is sufficient, no matter what kind of sensor you use, it is impossible to capture slight changes in light intensty with high resolution and at high speed. However, lighting has its limits. Particularly in case of high-speed fluorescence imaging, even if a considerable amount of power is used for light source, brightness is often insufficient, and it is usually not possible to irradiate large pixel sensor with sufficient light. Forcibly increasing brightness by using a huge lamp of 300W or more as a light source may damage filters and lenses in light path, so it is extremely difficult to increase fluorescence brightness by 10 times using ordinary optical equipment.

Bright light makes bleaching of voltage sensitive dye

Until now (2001), the mainstream approach to optical mapping has been a linear one: strengthen light, make an image sensor bigger, and reduce shot noise. In case of absorption dye, this is easily achieved, and this was achieved with an optical mapping device using a MOS sensor that we completed at the Electric Research Institute in 1991.

However, in case of fluorescence, in order to use this equipment efficiently, considerable consideration is required regarding optical system, staining, etc., which can be said to be difficult, although not impossible. The final issue is bleaching of fluorescent dyes. In the case of RH-795 dye, bleaching was noticeable even after recording for about 1 second, making it difficult to record systematically and stably, and it was necessary to collect data in one shot. In other words, the conclusion is that as long as stable measurements are desired, the brightness cannot be made that high. Therefore, the advantages of a large sensor cannot be fully utilized.

Averaging and signal processing

So, is it possible to perform fluorescence imaging that have both high resolution and high speed? In conclusion, the methods to solve this problem are averaging and signal processing.

Phenomena that cannot be seen in one measurement because they are buried in noise can be seen by averaging 16 or 64 times. Averaging is disliked because it cannot be applied to phenomena that cannot be repeated. Of course, if you need to measure only in one trial, averaging is not possible. However, if it is possible to do averaging measurement for things that "happen" even if they "can't happen," the applicability of averaging will expand.

For example, let's consider measuring state of spontaneous spiking activity. The challenge is to measure the activity of other neurons related to spontaneous firing when a cell spontaneously fires.

At first glance, this appears to be a one-off phenomenon and cannot be averaged. However, if cell activity can be observed in units using electrodes, averaging becomes possible. In other words, the recording of the unit is amplified and input to the trigger of the video recording device.If the images before and after this trigger are recorded in memory and averaged, asynchronous noise will be reduced and a synchronous signal will be sent to this cell. It will be emphasized. What is important here is the timing before and after the trigger. Before the trigger may be related to the activity that causes this cell to fire, and after the trigger may be related to the effect of this cell's firing. This is called a play-post trigger function because it records before and after the trigger, but the difference is whether or not it is equipped with this function.

Signal processing is also important, and it is possible to slightly improve shot noise by applying a noise removal algorithm that utilizes the properties of shot noise.

Optical system and light source

MiCAM camera we developed (in 2001) accumulates approximately 100,000 charges per pixel. Therefore, shot noise is approximately 0.3% when the saturation illuminance is 100%.

It has been experimentally confirmed that it is possible to observe changes in membrane potential using fluorescent dyes for many samples by averaging around 16 times and signal processing. What is important is that optical system and light source for measurement are extremely simple. As for optical system, if you want to use a magnification of 5x or more, you can use an ordinary fluorescence microscope as is. A 150W halogen lamp is sufficient for light source. Since brightness is normal, continuous irradiation for long periods of time is possible, even without an electromagnetic shutter (although we recommend using an electromagnetic shutter...) At low magnification, a bright macrofluorescence optics is required, but it can be easily made by combining commercially available lenses. MiCAM's standard sensor may be too small when matched with low magnification optics. Therefore, although the speed will be limited to about 3msec, we will gradually provide cameras using 1/2 and 2/3 inch sensors.

Conclusion

MiCAM's simple electrical performance is not necessarily the best in the world (for example, I admit that Pixel Vision's ultra-high-speed camera is superior in many respects).

However, MiCAM can definitely be used in experiments, and I am proud that it is a well-balanced device overall. Because it integrates the functions necessary for physiological experiments, we believe that we can provide many researchers with the closest solution to their desired experiments.

P.S. If you use MiCAM and feel that you need 10x S/N ratio, please consider that you also need 100 times the brightness. And consider that light source requires 10 times more stability and that dye bleaching occurs 10 times* to 100 times faster.(* If NA of objective lens can be increased by 3.2 times with the same magnification, irradiation light can be increased by 10 times and fluorescence can be increased by 100 times.)