Quantitative Imaging

EPFL–SV–PTBIOP
INTERNAL COURSE
2015
QUANTITATIVE IMAGING
AN EXPERIMENT INVOLVING
IMAGING
Biological
Question
Protocol
Development
Imaging
Image
Analysis
EPFL–SV–PTBIOP
Conclusion
2
IMAGE
PROCESSING/ANALYSIS
Image
Processing
Image in 
Image out
Image
Aquisition
Image
Analysis
Image in 
Numbers out
Image
understanding
spatial
domain
frequency
domain
IMAGE PROCESSING
Enhance
-ment
Filtering
Image
display
Image
restoration
Feature
extraction
complexity
Brightness/Contrast
Look up tables
Gamma function
Segmentation
Particle Analysis
3D rendering
Deconvolution
Structured Illumination
Fourier
transformation
QUANTIFICATION
Number
Size
Pixel intensity independent!
Shape
QUANTIFICATION
Transmission
Fluorescence
Concentration
Absorption
Concentration
Brightness
𝐼𝑇
𝑇=
𝐼0
𝐹𝐼 = 2.303 ∙ 𝑐 ∙ 𝜖(𝜆) ∙ 𝑙 ∙ 𝐼0 Φ𝑓
𝐸 = − lg 𝑇 = 𝑐 ∙ 𝜖(𝜆) ∙ 𝑙
Strictly
Valid only
for low
concentrations
IMAGES ARE ARTEFACTS
Two images of same object (sample)
imaged with the same microscope/objective!
Object
Image
of Object
STAINING & DETECTION
SCHEME
AD converter
Detector
𝑐𝐴 ~𝑝𝑖𝑥𝑒𝑙 𝑣𝑎𝑙𝑢𝑒
Iemm.
Iexc.
Fluorophores per Antibody (nF)
Secondary Antibody (AB2)
Primary Antibody (AB1)
Antigen (A)
DETECTION DEVICES
Array detector
Photons
Point detector
photoelectrons
Pixel value
CONFOCAL MICROSCOPE
Dwell time: 50 µs
Dwell time: 6 µs
Dwell time: 1.6 µs
Pixel intensity is
NOT
proportional to the number of collected photons !!!
SUMMARY DETECTOR
LINEARITY
CCD camera
• Linear over a large range
EM CCD
• EM-gain: exponential
PMT
• Gain: exponential
• Data normalization against pixel dwell
time
GaAsP/Avalanche photodiodes
• Photon counting mode
EPFL–SV–PTBIOP
SYSTEM CALIBRATION
• System calibration is necessary for comparison between different
images
•Image with the same microscope settings if possible
• Direct monitoring of excitation power is necessary for comparison of
different systems
•Check for absence of fluorophore saturation
• Measuring system PSF is a good way to monitor the system
performance
• Calibration should be repeated regularly to monitor stability of the
system
BACKGROUND
Condition 1
Condition 2
Difference
Ratio
Condition 1
Condition 2
Difference
Ratio
Signal Background Measured
Value
200
20
220
100
20
120
100
2
100.0
1.8
Signal Background Measured
Value
200
100
300
100
100
200
100
2
100.0
1.5
Sources of Background:
Dark Image, cross-talk unspecific staining, autofluorescence,…
IMAGE PROCESSING
• Background subtraction
• Flat field correction
• Photobleaching correction
• Correct sampling
• Filtering /deconvolution
BACKGROUND SUBTRACTION
original
background
Leica AF6000
63 x/1.3 Imm objective
background corrected
• Useful especially in transmission mode
• Removes of dust particles, spots from the image
FLAT FIELD CORRECTION
100 x/1.3 Oil objective
• Useful especially in fluorescence mode
• Flat field can be measured with calibration slide
• Subtract dark image before flat field correction
FLAT FIELD
CONFOCAL MICROSCOPE
Zoom: 1.0
pixel value
Zoom: 0.6
60000
60000
50000
50000
40000
40000
30000
30000
0
300
600
900
60000
50000
50000
40000
40000
30000
30000
0
300
600
900
0
300
600
900
Zoom: 4.0
pixel value
Zoom: 2.0
60000
0
300
position
600
900
position
PHOTOBLEACHING CORRECTION
• Max bleaching: fixed sample - 50%, live sample - 20%
• For strong bleaching SNR is different at the beginning and at
the end of the series
• Fluctuations of laser or lamp power are corrected similarly
IMAGE PROCESSING
SUMMARY
• Flatfield correction and background subtraction are
already necessary for analysis within single image
• For quantitation use images with at least 12 bit grey
values
• Use full dynamic range of detector with offset and
gain settings preventing clipping or saturation
• Use correct image sampling with pixel size 2 to 3
times smaller than resolution
• To remove noise use Gaussian or mean filter with
kernel size close to the resolution of the system
19
BIT DEPTH
Does the selected bit depth
correspond to the number
of measured
photons/photoelectrons?
Bit Depth
Grayscale
Levels
Dynamic
Range
(Decibels)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
16
18
2
4
8
16
32
64
128
256
512
1,024
2,048
4,096
8,192
16,384
65,536
262,144
6 dB
12 dB
18 dB
24 dB
30 dB
36 dB
42 dB
48 dB
54 dB
60 dB
66 dB
72 dB
78 dB
84 dB
96 dB
108 dB
20
1,048,576
120 dB
QUANTITATIVE IMAGING
• Getting quantitative information about the
specimen from the image
• It is necessary to avoid subjective bias and to
present the overall pattern of the data
• Relative quantification – measure relative intensity
• Absolute quantification – measure the number of
photons
• Thick specimens are difficult to quantify due to
scattering and absorption
• Important for the most of light microscopy
methods which are intensity based (e.g.
Colocalisation, FRET, linear unmixing)
PROTEIN LOCALIZATION
EPFL–SV–PTBIOP
nucleus
nucleolus
nuclear envelope
cytoplasm
nucleus + cytoplasm
mitochondria
peroxisomes
microtubules
focal adhesions
endoplasmic reticulum
Golgi
plasma membrane
10mm
http://gfp-cdna.embl.de/index.html
COLOCALIZATION
TYPICAL EXAMPLE
Vinculin
Alexa568
EPFL–SV–PTBIOP
Actin
Alexa488
http://www.olympusconfocal.com/applications/colocalization.html
Colocalization: The presence of two or more fluorophores on
the same physical structure (in a cell).
COLOCALIZATION INDICATORS
Pearson coefficient
𝑷𝑪𝑪 =
(𝑹𝒊 − 𝑹) × (𝑮𝒊 − 𝑮)
(𝑹𝒊 − 𝑹)𝟐 × (𝑮𝒊 − 𝑮)𝟐
Overlap coefficient
𝒐𝒗𝒆𝒓𝒍𝒂𝒑 =
𝑹𝒊 × 𝑮𝒊
𝑹𝒊 𝟐 ×
𝒌𝟏 =
𝑹𝒊 × 𝑮𝒊
𝑹𝒊 𝟐
; 𝒌𝟐 =
𝑮𝒊 𝟐
𝑹𝒊 × 𝑮𝒊
𝑮𝒊 𝟐
𝑹𝒊 , 𝑮𝒊 : 𝒊𝒏𝒕𝒆𝒏𝒔𝒊𝒕𝒚 𝒐𝒇 𝒕𝒉𝒆 𝒑𝒊𝒙𝒆𝒍 𝒊 𝒊𝒏 𝒊𝒎𝒂𝒈𝒆 𝑹 𝒂𝒏𝒅 𝑮 𝒓𝒆𝒔𝒑𝒆𝒄𝒕𝒊𝒗𝒆𝒍𝒚
𝑹, 𝑮: 𝒂𝒗𝒆𝒓𝒂𝒈𝒆 𝒊𝒏𝒕𝒆𝒏𝒔𝒊𝒕𝒚 𝒐𝒇 𝒕𝒉𝒆 𝒊𝒎𝒂𝒈𝒆 𝑹 𝒂𝒏𝒅 𝑮 𝒓𝒆𝒔𝒑𝒆𝒄𝒕𝒊𝒗𝒆𝒍𝒚
CORRELATION
Pearson’s Correlation Coefficient: 0.94
EPFL–SV–PTBIOP
ANTICORRELATION
Pearson’s Correlation Coefficient: -0.94
EPFL–SV–PTBIOP
EXCLUSION
Pearson’s Correlation Coefficient: -0.29
EPFL–SV–PTBIOP
PARTIAL OVERLAP
Pearson’s Correlation Coefficient: -0.016
EPFL–SV–PTBIOP
BIG/SMALL EXCLUSION
Pearson’s Correlation Coefficient: 0.19
EPFL–SV–PTBIOP
BIG/SMALL COLOCALIZATION
Pearson’s Correlation Coefficient: -0.047
EPFL–SV–PTBIOP
PEARSON COEFFICIENT
EPFL–SV–PTBIOP
COLOCALIZATION
EPFL–SV–PTBIOP
• Two different molecules can never be at the same
physical spot at the same time.
• Colocalization seen in images is coming from the
low-pass filtering of the image formation in light
microscopy.
• Colocalization is an artificial phenomenon and
therefore always relative.
COLOCALIZATION ANALYSIS
Object based
EPFL–SV–PTBIOP
“Nearest-Neighbor distance”
“Overlap”
•
•
•
Binarization by: top hat (image) = image
– morphological opening (image)
Objects of channel2 reduced to their
center of mass
Count of “Yellow Dot => percentage
•
•
•
same as “Overlap” Analysis
objects of both channels reduced to their
center of mass
If 2 centers of mass in different channels are
less then a defined distance (depending on
the resolution limit of the objective). They
are colocalized => percentage
E. Lachmanovich, Journal of Microscopy, 2003
33
COLOCALIZATION ANALYSIS
Intensity based
EPFL–SV–PTBIOP
Scatterplots - Fluorograms
Each position on a fluorogram correspond to the pixels with the green and red
corresponding components
The color on the scatterplot corresponds to the frequency of this value in the image.
DISPLAYING COLOCALIZATION
EPFL–SV–PTBIOP
•
Consider color-blindness.
•
•
Use Yellow/Blue or Green/Magenta
Show the scatterplot at the same time!
Green Channel
Showing Overlaps
Red Channel
Color-blindness simulation
(Dichromacy)
Quiz: What’s the color of the colocalization?
35
DISPLAYING COLOCALIZATION
EPFL–SV–PTBIOP
The Scatterplot or 2D Histogram
Green Channel
The eyes (and brain) lie!
Complement:
Colocalized Pixel Map
Red Channel
Helps see correlation qualitatively, no
bias, unlike color merge.
Helps to see problems from imaging
ANALYSIS: FLUOROGRAMS
EPFL–SV–PTBIOP
Why do we like Fluorograms?
Saturated
Noisy
Saturated
No correlation?
Wrong
Offset
Wrong Offset
Bleed through
ANALYSIS
EPFL–SV–PTBIOP
“Intensity based” colocalization
Quiz: Which scatterplot belongs to which
image series?
1
2
3
4
Result: A3, B1, C4, D2
ANALYSIS: INTENSITY-BASED
“Intensity based” colocalization
THRESHOLDING
from
CONTROLS!!
EPFL–SV–PTBIOP
ANALYSIS: INTENSITY-BASED
EPFL–SV–PTBIOP
“Intensity based” colocalization
Li’s Intensity Correlation Quotient (-0.5 to 0.5)
Partial colocalization
Mito-DsRed; ER-GFP
The ICQ is based on the non-parametric sign-test analysis of the PDM values.
and is equal to the ratio of the number of positive PDM values to
the total number of pixel values.
The ICQ values are distributed between 0.5 and +0.5 by subtracting 0.5 from this ratio.
ICQ = 0: Random Staining
-0.5 < ICQ < 0: Seggregated Staining
PT-BIOP 2012: Colocalization
0 < ICQ < 0.5: Dependent Staining
40
ANALYSIS: AUTOTHRESHOLDING
EPFL–SV–PTBIOP
Subjective bias during threshold selection makes for
poor reproducibility
Suggested method: Iteratively find where the Pearson’s correlation below thresholds <= 0
PT-BIOP 2012: Colocalization
41
STATISTICAL TESTING
EPFL–SV–PTBIOP
Are your colocalization results better than random
chance?
Costes’s Method
10/35 Pixels Colocalize: Is that
real or due to random chance?
Statistical confidence P - Costes et al. 2004 Biophysical J. vol 86 p3993
PT-BIOP 2012: Colocalization
42
STATISTICAL TESTING
EPFL–SV–PTBIOP
Costes’ Method - Randomization
Our result: 10
 Not by chance
1.
Randomize one image in PSFsized chunks
2.
Measure correlation, overlap or
other metric
3.
Repeat 100 times or more.
4.
How many randomized images
have better or equal correlation
than the original?
•
9
10
Statistical confidence P - Costes et al. 2004 Biophysical J. vol 86 p3993
PT-BIOP 2012: Colocalization
If >95% are worse: we
can start believing our
numbers. The p-value
P = 0.50 = 50% (No)
P = 0.95 = 95% (yes..)
P = 1.00 = 100% (YES!)
43
STATISTICAL TESTING
EPFL–SV–PTBIOP
Costes’ Method – Example
Virus entry to Calveolae
T = 10 Min.
• 32% of virus colocalized
•
•
Costes P-value 0.00
0% chance it’s true
T = 20 Min.
• 39% of virus colocalized
•
•
Daniel J. White – CBJ @ Max Planck
PT-BIOP 2012: Colocalization
Costes P-value 1.00
«100%» chance it’s true
Without significance test, we wrongly assume virus is
colocalised with caveolae at 10 min P.I.
It is not! Only at 20 min is there signficant correlation.
44
THANK YOU!
EPFL–SV–PTBIOP
References
https://ifn.mpicbg.de/wiki/ifn/images/f/fa/QuantitativeColocAnalysis-10-2011.pdf
http://fiji.sc/wiki/index.php/Colocalization__hardware_setup_and_image_acquisition
And The Monty Python!
PT-BIOP 2012: Colocalization
45
COLOCALIZATION MUST READ
http://pacific.mpicbg.de/wiki/index.php/Colocalization_Analysis
Costes et al. Manders et al.
Bolte & Cordelières
EPFL–SV–PTBIOP