*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****


 Hi Christophe,

there was an illuminating post in April 2010 here on the confocal listserver by Karl Garsha which i copied in below, i think the trick is to calibrate the camera without using the EM gain, and then calibrate the EM seperately. 

best wishes

Andreas

 




-----Original Message-----
From: Karl Garsha <[log in to unmask]>
To: [log in to unmask]
Sent: Wed, 28 Apr 2010 2:19
Subject: EMCCD gain, photon scissors, bogeyman avoidance

Hi All, 

If I am to recant my worldly ways and seek enlightenment in the manner of this 'one true religion' of EMCCD calibration, I'd like to know who we can credit. Can someone please help me with a citation or reference for this straight-forward <<1 photon approach?

Perhaps it is my over-simple yet pragmatic approach to problems that limits me, but I'm having trouble getting my head around the practical application of the simple darks and flats translating into subphoton imaging that we are being prescribed here. Exposure to Hulkamania, Bay Watch and monster truck rallys here in the US have likely so rotted my intellect that I fear I may have trouble grasping the subtle nuances of The Way. In all the excitement I must have misplaced my photon scissors and I don't know where to begin. What happens if the light source fluctuates while I'm madly clipping away at the feeble illumination with my photon scissors? Does this impact my calibration?

I can appreciate we should isolate the camera at least, so that we can measure within the constraints of camera noise and not total imaging system noise. We can begin with this. We should have our favorite bonafide even-field illumination methods ready. 

I would like to point out that we should keep in mind that we shouldn't just calibrate our EMCCD once, this should be done at regular intervals. EM gain registers change they're properties as the camera is used, in other words, EMCCD cameras 'age'. Some cameras have built in calibration features that offset this change in the gain mapping through camera electronics, but that is technology and a primary tenent of our new lack of faith is that leveraging technology will make us dumber. So we'll try to do this the smart way. Depending on how heavily we use our cameras we may want to re-map gain on a weekly basis. After a while it should become routine, removing the camera, mounting on the even field illumination station, making sure the light source output is stable and calibrated. After a while it may seem tedious. To be completely throrough, I suggest playing 'Thriller' on an audio loop and working on a bogeyman dance while you do the weekly calibration. After just 52 weeks you will have a well characterized camera as well as a very impressive dance. It won't hurt anything unless you drop the camera moonwalking back to the microscope, and this measure will protect calibrations from any residual influence of evil spirits or bad juju. 

To simplify and look at the big picture, let us contemplate, for the sake of discussion. What if, we first calculate our gain under non-EM conditions, using the non-EM read port. (by the way, the read noise here on a good EMCCD should be around 5-6 e- which is well in the range of a respectable fluorescence grade sony 285 interline). We can figure out the gain (full well capacity/bit depth) using the well-established mean variance method. We can do this gain calculation at a short exposure time, and over again at incrementally increased the exposure times. Calculate the gain at each exposure time, it should be consistent. But as the exposure time gets longer, both the mean pixel value and variance should increase in a proportional manner. This demonstrates linearity and implies a Poisson distribution of intensity values in accordance with what we would expect from quantum uncertainty. If I'm wrong, please help me gain a better understanding, I'm here to learn too, and I'm not infallible. The slope of this plot should reflect the constant 'gain' in electrons per grey level. In simple terms, as the amount of light captured increases, so does the pixel value. This happens in a predictable manner. If you look at a brightness level, you have some idea of the charge collected in that pixel, but only at a single gain level. This is a simple experiment, we hold the gain constant and the light constant and changed the exposure time, and saw how that translates into pixel values. 

Now, what if, we start using the gain register. We can hold the light constant, hold the exposure constant, and take an image with zero gain applied. Subtract the zero-exposure 'bias' image. See what the modal brightness value is. Does it correspond to a place on the linear plot we created earlier? Raise the gain some convenient increment, does the (bias-subtracted modal) pixel value fall on our non-em gain plot of pixel values somewhere? How much exposure time difference corresponds to the two points at different EM gain settings? What does this relationship imply? What if you compare the mean value instead of the modal value?  You can start to look at other things like the distribution of pixel values and coefficient of variation at various EM gain settings as well. Are we measuring anything here or is this another grand illusion perpetrated by the Dark Prince of sub-photon imaging? Did the number of photons coming out of the light and hitting the chip change to half as many per unit time because we're using EM gain? 

A natural extension of this is to now do this simple heretical experiment from zero gain to 1000x EM gain at regular increases in gain. At a constant light level, does the increase in modal pixel intensities follow a linear trend with linear incremental increase on the gain scale? Because this is easy, check this many times for each gain level...how precise are the values? If you do this many times at the same gain setting, does the value keep changing? This could be temporal gain drift, a source of imprecision. Does your camera have a specification for gain drift? Or is it the light source? Let us hope not...you better have a stable light to measure the camera, in accordance to the isolation principle.

Check the bias...collect a series of images at zero exposure time. Plot the modal values as a function of time in the series...did the bias drift? Does the gain drift and the bias drift? Now what. How can you subtract bias from a series of images if the bias is fluctuating, and what if the gain fluctuates as a function of time too? High quality EMCCD cameras have provisions to minimize this and provide a specification for both bias stability and gain stability, but it's technology. You can choose to leverage it, or not. Linearizing that gain curve isn't trivial either, so if you have a situation with different people using 'almost' the same gain they may be working with samples that are very different in quantum yeild, especially towards the higher side of gain amplification.  

What this all boils down to is ability to report differences within and between samples in some form of responsible unit. Arbitrary units are subjective, mabye even passe'. Yes you can have an internal control and use ratios to normalize, but not if you change that gain between channels. Reporting in units that actually mean something allows results acquired at different gain states to be compared in a meaningful manner, it's more than just veneer. It's a means of ensuring consistent results and comparing data, and the point of this level of technology is to transcend what can simply be percieved through the eyepieces. This equipment isn't for making pretty pictures, we may as well go back to drawing our results with pen and ink for that. To extract information we would not otherwise be able to get...that is the key...but to get there we need some sort of tool. Our species is widely renown for the development and clever use of tools. We accept some risk of being wrong any time we perform a measurement. We all have our limitations.  

Hopefully this will stimulate some thought, if I'm wrong we'll at least be able to measure how wrong at some point. 

Best Regards,

Karl