Scribd
Upload
19 views

Basics of Imaging 2020

Uploaded by

None Important
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
19 views

Basics of Imaging 2020

Uploaded by

None Important
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 36

Bioimage Analysis

Basics of Imaging
The signal
So, now that we have some understanding on what an image actually
is, we can start thinking about how images come to be.

To put precisely, we will discuss what the signal is and how we measure
it. Remember, most of the time we are not taking pictures, we are
doing measurements.

The signal of course depends on what we want to measure, but as we


are discussing light microscopy our raw signal will inevitably be
composed of photons.
Fluoresence
Works roughly like this. But that is simplified.
This is more realistic.
And that is just the excitation point
Additionally, few other things may happen to the photon
before it reaches the fluorophore or the detector.

Refraction
Light can be refracted, i.e. scatter because of collision with
something substantial such as electrons. Scattered light might
end up where want, or it might not.
Absorption
Light can be absorbed by the sample. So the photons are lost
on the way to the sample, or on the way from the sample.

Both these effects happen more often the longer the photon
has to travel through the sample. This has implications.
Modalities
Capturing the light
We’ll focus on the two modalities bioscientists
are most likely to encounter in the laboratory,
though let it be noted that this is not the Full
Story ™.

First we’ll walk through how widefield


microscopes work and then we’ll walk through
how confocal microscopes work.
Widefield Microscopes
Widefield microscopes
Widefield microscope acquires an image of the entire field at once.
This is often simple and robust*.
Image aqcuisition time is often called
’exposure time’ during which sample is
exposed to the light and the emission signal
is collected.

The longer the exposure time, the more


signal we get.

*Terms and conditions apply.


Widefield microscopes
Widefield microscope bathes the sample in light. Everything in the
field of view will be excited. Everything in focus, everything out of
focus.*

You really don’t want to imagine more than a single layer of cells.
* Of course we get bit less signal from the out of focus areas, but still if you sample is thick, your image
is going to be messy.
Widefield microscopes
Can be pretty darn fast. With good dye/stain, exposure times of 100-
300 ms are often enough, allowing you to get 3-10 images per second.

Relatively easy to use – just point at what you want to see, set
exposure time´, chose filters and you are ready to go.

Depth resolution is poor / nonexistent. You can take stack and


deconvolve but confocal is almost always better.
Widefield microscopes
Detection
Microscopy camera chips consist of millions of photodiodes. Incoming
photon energy is used to excite electrons of the silicon which are in
turn collected in a storage well and afterwards transferred to an
amplifier. This process is almost linear.
Imaging parameters (WF)
Exposure time
The more time you use to expose the image, the better signal you will
get. The more time you use, the more you will burn your sample.

Optical magnification
Objective as well as various lenses in the light path determines the
pixel size, effectively determining the resolution.

Binning
Can be used to gain more signal at the cost of resolution.
Binning
Binning means combining pixels so that intensity values become higher,
but we get fewer pixels.

This has fringe uses when the sample is dim and/or we want to acquire
images quickly.

0 10 20 20
20 100
0 10 20 40
2x2 binning
20 30 40 60
110 240
20 40 60 80
Binning
Let’s have a look how it works. So let’s bin poor Fabio and some cells.

With Fabio, there was no need for this operation. This image is shown
to explain how binning affects the resolution.

However with the cells, the signal is greatly magnified.


Binning
That being said, you only really need binning if
your signal is really poor, or if you are working
with a stain that disappears when you look at it.

If you find yourself in such a situation, the first


thing you should do is optimize your protocols
and get a better label.
In cases where you need as many pictures as
possible quickly as possible (short exposure
times), binning might actually help.
Confocal Microscopes
Scanning confocal microscopes
Magic word here is ’scanning’. Laser is used to excite the fluorophores
at each pixel individually. Normal scanning speeds being around 400
Hz, or 400 lines / second, this means that laser spends less than 5
microseconds in one pixel.

Yes, the sample is literally scanned.


Scanning confocal microscopes
Pinhole us used to remove out of excitation light meaning that while
the laser illuminates ‘everything’ we only measure signal from the
focal point.
Confocal v.s. widefield
Scanning confocal microscopes
Detection
The first part of a photomultiplier tube is
a photocatode that emits electrons when
hit by a photon.

The signal is then amplified by a cascade


of dynodes and the signal is measured at
the anode.
Voltage dropping resistors determine how
much amplification is applied – this is the
gain you can adjust on the panel.
Imaging parameters (confocal)
Scanning speed
The faster you scan, the faster you get an image and the less you
bleach your sample. However, the faster you scan, the worse
signal you get.

Zoom/format (or pixel size)


Boils down to the question, how large part of the sample lays
under one pixel. Larger the pixel size, the better the signal.
Smaller the pixel, the better the resolution… to a limit.

Gain/Offset
Increased gain increases the signal and also the noise. Offset can
be used to reduce noise and signal.
Imaging parameters (confocal)
Scanning speed combined with the image format give a parameter
called ‘pixel dwell time’.

As you might guess this tells you how long the laser stays on a single
pixel before moving onward. Pixel dwell times are usually in the range
of microseconds.

The longer the laser stays in one place, the less noise we get. On the
other hand, the longer the laser stays in one place, the slower the
whole thing is.
Gain and offset
Offset tells us how low signal we regard as zero. This can be used to
remove the background noise when it is homogenous.

Gain tells us how much the signal is amplified. This parameter is not
linear even if Leica tells you it is.

Gain, scanning speed, laser power and resolution should be adjusted


to avoid overexposure and optimized according to you needs.
Offset
Offset is user to correct for the background. Poorly adjusted offset
introduces additional signal to the image.
If you wish to quantify
structures, you can set
offset pretty haphazardly.
‘Good enough’ will do.

If you want to quantify


intensity with a confocal,
you need to evaluate
your life choices.
Gain
I took a picture of the same location using same laser power but
different gains. The fraction is stable only really in narrow band of gain.

600 700 800 900

600 700 800 900 This has several implications on measuring


ROI1 20,39 51,4 111,01 182,44 intensity. You must use the same settings for
ROI2 9,73 23,26 50,36 85,25 all samples, but not all samples are visible
Fraction 2,095581 2,209802 2,204329 2,140059 with the same settings.
Noise
Noise, or unwanted signal, is present in every image. This is mostly
‘physics happens’ kind of a problem.

You can reduce the noise almost always, but this improvement comes
at a price. The price you pay is time, either to prepare a better sample
or spend more time measuring your data.

Post-processing only ever alleviates the problem. The problem can be


only solved by preparing a better sample or recording better images.
Averaging
Averaging images is a commonly used procedure with noisy images
used to reduce the noise and so increase the signal to noise ration.

Let’s say there is some ’truth’, a cell where we have stained vesicles of
some sort. Then let’s say we acquire three pictures of it. All of them are
bit grainy because of measurement errors.
Truth Measurement 1 Measurement 2 Measurement 3
Averaging
Basically works basically like this.
Averaging
You can always increase the time used to record the data.

This is an easy solution, but unfortunately it increases the time needed


and sometimes you don’t have the time, or your sample cannot take it.
1 4 8 16
Same data recorded with
average of 1,4,8 and 16.

As we increase the averaging,


the data becomes less noisy.

Noise is reduced by
1
the factor 𝑁
where N = average!
Mind your tools
Care must be taken when viewing
images. Many of the tools meant
for photography and general image
manipulation automatically smooth
the images for the user.

This is less of a problem when you


are looking at cat pictures, but
when dealing with data caution
must be taken.
Resizing
If and if you need to resize your
images, that is the last thing you do to
them before attaching the to a
document.

Image on the right has been shrank by


factor of four and then expanded
again to the same size as before.
To wrap things up
Quick note on post-processing
You often hear claims that image quality can be improved after
acquisition. This depends a lot on what you mean by quality.

Any and all attempts to reduce noise in post-prosessing will also reduce
resolution. Deconvoluton is an exception to this.
Any and all attempts to increase resolution will also increase noise.
Noise and signal are both defined as ‘sudden change of intensity’,
which means that coming up with mathematical tools that affect one
without affecting the other are hard to come by.
On optimization
When you start a new imaging project where you need to
measure something from the pictures, it is often good to spend
one session taking pictures with various settings.

This allows you to figure out how quickly you can record usable
data. This is especially crucial when you are working with 3D
data (so confocals) as shortening imaging time even few minutes
per stack could mean hours of saved time.
Concept of worst usable image
Ideally you do not want to spend any more time at the
microscope than absolutely necessary.

If you take better images than you need, you are wasting time
and money. If you need lots of data, this cost might be
considerable.

You should establish ‘worst usable image’ that gives you the
information you need while using as little time as possible. This
might require some optimization.
Questions

You might also like