Point Spread Function, Spectral Calibration & Spectral Separation: Quality Assurance Testing

licoricehealthΤεχνίτη Νοημοσύνη και Ρομποτική

14 Νοε 2013 (πριν από 3 χρόνια και 11 μήνες)

60 εμφανίσεις

Point Spread Function, Spectral Calibration
& Spectral
Separation: Quality
Assurance Testing

Light Microscopy Research Group

Richard
W. Cole

Wadsworth Center / NYSDOH

Albany, New York










Why ?


until the last 5
-
10 yrs, simply observing a specimen was sufficient;


advances
in light microscopes
necessitates
traceable standards & procedures



Overall Goal


the creation of a range of imaging parameters traceable to standard
references




NIH



Realizes the need for and supports the “core” model




40% of S10 grants funded
were imaging
in general
;
13%
confocal




NIST

goal
of moving medical imaging &
lab
testing from an art to a science




FDA


ensure
manufacturers systems are
reliable, guaranteeing that the drugs will
be
safe & efficacious




Congress



provide the financial support
for comparable
standards
of research


Quality and standards: Making bioimaging ‘measure up’

Susan M. Reiss BioOptics World, Jan/Feb 2010, Vol.3 No.1, p.14
-
18

Access sparks action

Lila Guterman NCRR Reporter, Winter 2010, p.4
-
8


Phase
One
a worldwide
research study to ascertain the
current state

of light
microscope performance
using

simple, efficient & robust tests
for

LASER stability,
field illumination &
coregistration




define & improve
cross
-
platform

standards
--
assist
core
managers & users


maintaining microscopes
for optimal operation with the ultimate goal of


improving the validity of quantitative measurements in light microscopy




the results of this study were accepted for publication in late 2010 in


Microscopy and Microanalysis,
one of the highest rated imaging journals





throughout
2011,
the LMRG tested and defined additional areas of


instrument performance (
Phase Two
) and refined the methodology


for determining a system’s
Spectral calibration, Spectral separation


ability and finally the Point Spread Function of an imaging system



What we have done

Stack, R.,
Bayles
, C., Girard, A., Martin, K.,
Opansky
, C., Schulz, K., and Cole, R. (2011) Quality Assurance

Testing
for Modern Optical Imaging Systems. Microscopy & Microanalysis 17(4):598
-
606.



Cole, R.W.,
Jinadasa
, T., and Brown, C.M. (2011) Resolution and Quality Control of Confocal Microscopy
Optics
.
Nature Prot. 6 (12): 1929

1941.


Spectral
calibration




Purpose:



Measure spectral calibration of the detection system.



MIDL lamp / mirror slide protocol:



Use 10x lens or no lens (system dependent)


Set up the MIDL lamp as the illumination source
or

use laser(s) and mirror slide (remove blocking)


Set the PMT gains to be ‘equivalent’


Perform a lambda scan and measure the signal
-
to
-
noise


Compare acquired spectra with published spectra



Analysis:



1.
Determine if your PMT(s) show
significant

spectral variation (sliders) or signs of aging



reference:
http://www.lightforminc.com/MIDL/index.html





PARISS Spectral Calibration Lamp, Lightform,Inc. Asheville, NC

Overlay of 5 PMT responses and MIDL lamp calibrated output / before repair

PARISS Spectral Calibration Lamp, Lightform,Inc. Asheville, NC

Overlay of 5 PMT responses and MIDL lamp calibrated output / after repair

Quality of Spectral un
-
mixing
:



Purpose
:



Measure the spectral un
-
mixing capability of an imaging system.



Protocol:



Bead slide: 6.0 µm FocalCheck Double Orange fluorescent microspheres


(excitation/emission maxima:
c
ore
= 532/552 &
shell
= 545/565)



o

use same optical settings/components (i.e. laser line/excitation, dichroic filter)


to acquire reference
and

experimental spectra

o

set detection to maximize S/N
without

any pixel saturation

o

select a detection bandwidth wide enough to encompass full emission range


(e.g.
DoubleOrange

beads 520
-
575 nm)

o

if available, choose detection set
-
up (i.e. parallel vs. lambda)

o

split detection into
smallest

discreet ‘bins’ if using
lambda

scanning mode

o

select an area of the reference spectra (via ROI) with the highest S/N and store in database




Analysis:



Select the most appropriate unmixing data
-
processing algorithm available:



automatic mode (1
st

pass / not generally adequate)


parallel mode (
simultaneous
data acquisition across multiple PMTs)


lambda mode (lambda scanning utilizing one PMT)




Spectral separation

FocalCheck
™ fluorescence microscope test slide

core = 532/552 & shell = 545/565
)

Image of a bead where the core and ring have a small spectral separation

R
ing
and core are pseudo
-
colored for illustration purposes

Linear Unmixing Algorithms



The measured spectra of a ‘mixed pixel’ is broken down into a collection of


component spectra (
endmembers
) and a set of subsequent fractions


(
abundances
) that indicate the ratio of each
endmember

in the pixel




Three distinct stages of spectral unmixing :



-

dimension reduction (i.e. data reduction)



-

endmember

determination (i.e. # of distinct spectra)



-

inversion (i.e. abundance estimation)



Employs a
linear mixing model



A Survey of Spectral Unmixing Algorithms

Nirmal Keshava Lincoln Laboratory Journal, Vol.14 No.1, 2003, p.55
-
78

Brain tissue (5 different labels)


blue

= cell nuclei,
green

=
Nissl
-
specific for neurons
,
yellow

= reactive
astrocytes
,

red

= microglia
,
purple

= endothelial cells representing blood vessels.

Resolution



point at which
two

objects are perceived as separate and distinct from one
another



Resolution
obtained from an imaging system is affected by:


the
specific

wavelength of light in use


the diffraction of light (Rayleigh, Abbe & Sparrow limits)


lens
aberrations


sample prep (coverslip thickness, mounting media, RI matching
)




Lens imperfections such as coma, astigmatism and spherical aberrations will result


in a loss of resolution




microscope
resolution and the extent of image ‘blur’ is typically described in terms


of its Point Spread Function (PSF
)




an ‘ideal’ PSF demonstrates symmetric balance and proportion

Limit of resolution:

d = 0.612 (
λ
) / N.A.

What is a Point Spread Function & why is it so important ?



a measure of the degree of blurring of an object
&
any potential
aberrations




speaks directly to the quality/resolution of an imaging
system



Image
= convolution of an object and the point spread
function



an
object plane light wave refocused by a lens produces a blurred


focal plane point commonly referred to as an ‘airy disc / airy pattern




sub
-
resolutional

beads are typically used


Point Spread Function
:





Purpose
:



Measure the point spread function of an imaging system.



Protocol:



Bead slide: 175 nm PS
-
Speck beads (mixture of blue, green, orange & deep red single
-
color beads)


o

test multiple lens: i.e. 20x, 40x , 63x & 100x (
all objectives routinely used for imaging in your lab)

o

collect a Z series
or

scan in XZY mode

o

if needed, suitably rotate image to obtain a ‘side view’

o

if your system is
filter based

(non
-
AOBS), check various dichroic filters




Analysis:




use the MetroloJ plug
-
in (Fiji / ImageJ) to determine the FWHM lateral & axial resolution


compare the experimental vs. theoretical resolution values


check the curve fits for all three



MetroloJ PSF report

http://pacific.mpi
-
cbg.de/wiki/index.php/Fiji

‘Idealized’ PSF images

courtesy of Zeiss

Theoretical PSF images / Confocal vs. Widefield

courtesy of Media Cybernetics

Widefield PSF of thick specimen

coverslip


increasing

depth &

worsening PSFs

|

|

V


‘3D’ Widefield PSF


20x / Refractive Index mismatch

collar incorrectly set to water // RI(water)=1.33, RI(Leica imm.oil)=1.518

40x oil NA 1.25 / pinhole = 0.5 &
5
airy units

63x N.A. 1.4 oil immersion lens / Brownian motion

Corrective Actions

Spectral Calibration


a.
Service call


Spectral Unmixing


a.
Try a different unmixing algorithm



-

avoid using ‘automatic’



-

try various
linear

algorithms



-

try
non
-
linear
(e.g. SWCCA) algorithms

b.
Try a different detector set
-
up



-

use (5) PMTs with simultaneous scanning OR



-

use (1) PMT with
lambda

scanning

c. Improve the signal
-
to
-
noise


Point Spread Function


a.
Clean the lens and optics / remove all
air bubbles

b.
Check for any possible refractive index mismatches

c.
Try a different lens

d.
Open pinhole aperture to mimic widefield conditions

e.
Check for optical misalignment



** It is important to note that the above suggestions DO NOT


encompass
all

possible solutions to these issues **

The test specimens proposed for both phases of this study were decided upon by

the members of the LMRG for their applicability, robustness, ease
-
of
-
use and relative

cost. While the phase I & II tests utilize materials from specific vendors who offer

excellent products for these purposes, neither the members of the LMRG nor the

ABRF
endorse

the use of these specific
vendors
, and fully acknowledge the use of

legitimate alternatives for the purposes of instrument performance testing.

Acknowledgements

Light Microscopy Research Group

Carol
Bayles


Cornell University

Claire
Brown (chair)

McGill University

Richard Cole


Wadsworth Center / NYSDOH

Brady Eason


McGill
University

Anne
-
Marie Girard


Oregon State University

Jay Jerome


Vanderbilt University

Tushare Jinadasa

McGill
University

Karen
Jonscher
(EB Liaison)

University of Colorado

Cynthia
Opansky


Blood Center of Wisconsin

George McNamara


University of Miami

Katherine
Schulz


Blood Center of
Wisconsin

Marc Thibault


Ecole

Polytechnique


* We would also like to thank the ABRF for their financial support and commitment to this project *