Application of image flow cytometry for the ...€¦ · cell, RBC morphology, red cell concentrate, spheroechinocyte , storage lesio n 1.! INTRODUCTION With over 85 million red blood
Post on 19-Aug-2020
0 Views
Preview:
Transcript
Application of image flow cytometry for the characterization of red blood cell morphology
Ruben N. Pintoab, Joseph A. Sebastianac, Michael Parsonsd, Tim C. Change, Jason P. Ackerfg and Michael C. Koliosab
aInstitute of Biomedical Engineering, Science and Technology (iBEST), Toronto, Canada
bDepartment of Physics, Ryerson University, Toronto, Canada cDepartment of Electrical & Computer Engineering, Ryerson University, Toronto, Canada dLunenfeld-Tanenbaum Research Institute (LTRI), Sinai Health System, Toronto, Canada
eMilliporeSigma, a business of Merck KGaA, Seattle, WA fCentre for Innovation, Canadian Blood Services, Edmonton, Canada
gDepartment of Laboratory Medicine and Pathology, University of Alberta, Edmonton, Canada
ABSTRACT Red blood cells (RBCs) stored in hypothermic environments for the purpose of transfusion have been documented to undergo structural and functional changes over time. One sign of the so-called RBC storage lesion is irreversible damage to the cell membrane. Consequently, RBCs undergo a morphological transformation from regular, deformable biconcave discocytes to rigid spheroechinocytes. The spherically shaped RBCs lack the deformability to efficiently enter microvasculature, thereby reducing the capacity of RBCs to oxygenate tissue. Blood banks currently rely on microscope techniques that include fixing, staining and cell counting in order to morphologically characterize RBC samples; these methods are labor intensive and highly subjective. This study presents a novel, high-throughput RBC morphology characterization technique using image flow cytometry (IFC). An image segmentation template was developed to process 100,000 images acquired from the IFC system and output the relative spheroechinocyte percentage. The technique was applied on samples extracted from two blood bags to monitor the morphological changes of the RBCs during in vitro hypothermic storage. The study found that, for a given sample of RBCs, the IFC method was twice as fast in data acquisition, and analyzed 250-350 times more RBCs than the conventional method. Over the lifespan of the blood bags, the mean spheroechinocyte population increased by 37%. Future work will focus on expanding the template to segregate RBC images into more subpopulations for the validation of the IFC method against conventional techniques; the expanded template will aid in establishing quantitative links between spheroechinocyte increase and other RBC storage lesion characteristics. Keywords: Blood bag, discocyte, image flow cytometry, image segmentation, morphology characterization, red blood cell, RBC morphology, red cell concentrate, spheroechinocyte, storage lesion
1. INTRODUCTION With over 85 million red blood cell (RBC) units consumed annually, RBC transfusion is the most frequent procedure carried out within transfusion medicine1. The primary goal of RBC transfusion is to augment the oxygen carrying capability of a compromised cardiovascular system. Using specialized separation techniques, blood banks extract RBCs from donated whole blood and suspend them in a preservative solution2. This red cell concentrate (RCC) is stored in a sterile blood bag at hypothermic temperatures until used for transfusion. The in vitro shelf life has been set at a maximum of 42 days for RBCs suspended in a saline-adenine-glucose-mannitol (SAGM) preservative under storage temperatures of 1-6°C3. The storage duration was based on standard guidelines specifying that 1) a maximum of 1% in vitro percentage hemolysis (RBC rupturing) and 2) a minimum of 75% percent in vivo survivability (24 hours post transfusion) must be observed on the last storage day4.
High-Speed Biomedical Imaging and Spectroscopy: Toward Big Data Instrumentation and Management IIedited by Kevin K. Tsia, Keisuke Goda, Proc. of SPIE Vol. 10076, 100761F · © 2017 SPIE
CCC code: 1605-7422/17/$18 · doi: 10.1117/12.2253583
Proc. of SPIE Vol. 10076 100761F-1
Downloaded From: http://proceedings.spiedigitallibrary.org/pdfaccess.ashx?url=/data/conferences/spiep/91601/ on 05/27/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
While these criteria provide some information on the quality of RBCs that remain in circulation after transfusion, they lack the ability to assess the functional capacity of the transfused RBCs. It has been well documented that during in vitro storage, biochemical and biomechanical changes in RBCs (collectively termed as RBC storage lesions) lead to deleterious effects in cell function and viability5-7. One of the consequences of the RBC storage lesion is the loss of cell membrane integrity, which leads to irreversible damage and impaired in vivo functionality8. Due to a hypothermic storage environment, the glycolysis pathways slow down and cause a shortage of ATP for basic cellular function9. Lacking a constantly replenished supply of ATP in vitro, reductases lose the capacity to effectively combat the accumulation of oxidative products. The lengthened presence of anionic radicals leads to the damage of phospholipids and proteins that are responsible for membrane structure10. To counteract this increasing oxidative stress, RBCs begin the removal of damaged cell membrane via vesiculation11,12. Under in vitro storage conditions, however, these products remain within the same supernatant that suspends the RBCs. RBCs under this continuous biochemical stress gradually morph from regular biconcave discocytes into smaller, more rigid spherical objects with membranous projections8. Morphologically termed as spheroechinocytes, these RBCs lack the deformability to efficiently enter capillary networks, reducing their capacity to deliver oxygen to the microvasculature13. Hemolysis and cell survivability studies are unable to evaluate the morphological state of intact RBCs. In blood banks, a morphological assay based on light-microscopy determines the impact of the RBC storage lesion on cell shape14. Samples of RCC are diluted, smeared, fixed and stained on slides to be viewed under a light microscope; 100 cells are randomly chosen for morphological characterization by observation15. Spheroechinocytes are visually detected by markers that include a spherical shape, spiky surface projections, relatively small diameter and lack of central pallor16,17. The morphology assay measures the number of spheroechinocytes (as a percentage), providing an indication of the effect of the RBC storage lesion on the morphology. This measure of the in vitro quality of the RBC sample can infer its capacity to transport oxygen in vivo. The light-microscopy morphology assay provides an insight into the functional capacity of a sample of RCC. However, it is a laborious process that is prone to subjective bias. Flow cytometry possesses the potential to overcome some of these limitations. The technology rapidly assesses large cell populations for multiple parameters, using signals from the transmitted and scattered light from each cell18. It has been shown that the intensities of forward and side scatter can be correlated with cell size and internal heterogeneity, respectively19. Furthermore, fluorescent tagging can be used to acquire further details on contents of the cell membrane or cytosol. Flow cytometry is superior to light microscopy in the speed and automation available for measurement and statistical analysis, however, lacks in an ability to provide images for differentiating morphological changes. Image flow cytometry (IFC) combines the complementary techniques of light microscopy and conventional flow cytometry. IFC incorporates a bright field light source, microscope objectives and high-quality CCD cameras into a conventional flow cytometry platform to respectively produce the illumination, magnification and spatial sensitivity necessary for image capture20. Paired with an integrated image analysis package, IFC expands on the analytical capabilities available in conventional flow cytometry, while providing automation and speed that is impractical for conventional light microscopy. Specifically, IFC provides the image resolution and morphological content of light microscopy, along with large sample size, automation and population statistics provided by flow cytometry21. This paper presents a novel RBC morphology assay using the Amnis ImageStream®X Mark II IFC system (MilliporeSigma), and a customized image segmentation template developed within the Amnis IDEAS® software package (version 6.2.64). The morphology assay and template are shown to process 100,000 images to select only focused, single cell, front view RBC images (~25,000 – 35,000), from which the relative spheroechinocyte population percentage is calculated. This automated, high throughput technique is applied to samples of two RCC blood bags over their lifespan to monitor the morphological RBC changes as a function of storage time.
2. METHODS 2.1 RCC blood bags Two RCC blood bags were transported from Canadian Blood Services (CBS) blood-for-research center (netCAD, Vancouver, Canada) to the Institute of Biomedical Engineering, Science and Technology (iBEST, Toronto, Canada).
Proc. of SPIE Vol. 10076 100761F-2
Downloaded From: http://proceedings.spiedigitallibrary.org/pdfaccess.ashx?url=/data/conferences/spiep/91601/ on 05/27/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
Throughout the 15-20 hour transit period, the bags were insulated in standard blood shipment containers which are comprised of a styrofoam /cardboard box filled with ice packs to maintain hypothermic temperatures. The RCC provided in blood bags were separated from anticoagulated whole blood using a top/bottom manufacturing method, leukoreduced and stored in up to 110 ml of SAGM preservative solution. Upon reception, blood bags were promptly stored in a designated laboratory fridge until six weeks after the production date. Before the first sample extraction for IFC analysis, bags were fitted with a sampling site coupler that contained a natural rubber sleeve stopper for needle penetration. 2.2 IFC morphology assay In a biosafety cabinet, 1.5 ml aliquots of RCC were extracted with 20-gauge needles via the coupler. This method was repeated to collect three individual samples (per bag) into safePICO syringes, which contained a vented cap to expel air within the sample and seal the sample from the external environment (Radiometer, Copenhagen, Denmark). The syringes were placed in a plastic pouch, covered in bubble wrap and placed in a cooled styrofoam box for transportation to the ImageStream®X IFC system. Syringe samples were placed in an adjacent refrigerator until required for measurement. From each syringe, an RCC aliquot of 5µμl was diluted with 200µμl of phosphate-buffered saline in a vial, which was placed into the ImageStream®X sample holder for automated aspiration and image acquisition. 100,000 images were captured, per sample, at a low speed and high sensitivity fluidics mode and under 60× image magnification. Post acquisition, the syringes and excess RCC samples were discarded as per biosafety guidelines. The experiments were performed at approximately 10 day intervals until the expiration date of the blood bags. 2.3 Data analysis All raw data files were transferred via USB to a portable hard drive to be uploaded onto a computer with a Windows 7 64 bit OS, consisting of an Intel Core i7-4790 processor and a 16 GB memory. Each raw data file was opened on the IDEAS® software platform and was processed automatically by the developed RBC image segmentation template (Section 3.1). The processed file displayed the relative percent fraction of spheroechinocytes within the analyzed population.
3. RESULTS/DISCUSSION 3.1 RBC image segmentation template Figure 1 outlines the gating strategy used to exclude unwanted subpopulations (represented by images A, B, C, F and G) from the analysis. The goal was to create an image segmentation template based on a series of mask/feature combinations to automatically differentiate front view discocyte (images D, E) from front view spheroechinocyte subpopulations (images H, I). In general, feature values for individual images were calculated from the intensity distribution of pixels enclosed by an underlying object mask. The segregation of RBC images into subpopulations was carried out in five steps (Figures 2-6). Since each dot on a plot was addressable to an image, subpopulation gates based on regions of interest (ROIs) could be visually verified. Furthermore, tagged RBCs that were representative of certain RBC subtypes were chosen in consultation with an RBC morphology expert (T. Turner, CBS, Edmonton, Canada). Once overlaid onto plots containing the entire population, these truth populations were used as a guide to refine subsequent gates and ROIs. Figure 2 shows a histogram used to discriminate between unfocused and focused images, as illustrated in section I (Figure 1). The feature gradient RMS calculates the root mean square (RMS) of the intensity gradient between pixels within the image mask22. Compared to focused images, unfocused and blurred images lacked sharp changes in intensity, thereby producing lower gradient RMS values. By visual verification, images with gradient RMS values below 54 were excluded from further analysis. Figure 3 displays the plot used to discriminate between single and multi-cell images, as illustrated in section II (Figure 1). Here the aspect ratio feature, calculated as the ratio of the object’s minor axis over its major axis, was plotted against the area feature based on the size of the object in square microns22. Both features used the default bright field object mask. A gate was created such that object aggregates within the same image field were excluded from further analysis.
Proc. of SPIE Vol. 10076 100761F-3
Downloaded From: http://proceedings.spiedigitallibrary.org/pdfaccess.ashx?url=/data/conferences/spiep/91601/ on 05/27/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
(%P)
M 0
(_41(=
)1,4 0
Cc=
)(14.)
(V
hr
nA8
oo ('444
tI
li.1
oa
T
UV
ou
OQ
355
Ch01
11,. .
889
Ch011026
Ch01
.1094
Ch01
.;.'1C8
Ch01
1
1115
Ch011169
Ch01
t'1197
Ch01
-1/
1444
Ch01
` >(/1 -.
1446
Ch01r 1481
Ch01
'1514
Ch01
l{1540
Ch01
../1624
Ch01
T,1653
r j,
8c;6
{i;1975 , 2025 {,/ ^117 ! 2187
îZj2381 2388
; ,1_
2582
,.: 2688
-..-.^98,
::%
3005
1..:s3147 3345
`,3354
,L:_,
3368 3434 3608 ' 3755 't. 3773 3796
O4062
I..4376 r' 4C97 9538
;l`9618 9634
,,9776 ¡
10384 e:10403
ff;- 10463 19775 ,' 19906 20373 21195.
c21240 21251
,; 21259 ,-..
21306
Q21456
C28319 28383 ..::
28416 1 28534
(Jl36853 J 36919
Q36941 36945
47314
67260
O.
s
37030 \/ 37082
f`37092
54075
O
Q
37189
57033
.i 41386
57096 0
41512
57189 e
45522
58707
O45579
58735
/
e53872
72822
0
: ;
53961
72938
(:_?
45608
./
45636
59305
¡i,v
t
45659
67019
:'.....e
0
47146 47275
67233
v /
67074 j59272
q.)73065
1 :73097
`*78098
078126
O84224 84238
Ch01 Ch01 Ch01 Ch01 Ch01 Ch01 Ch01 Ch01 Ch01 Ch01 Ch01 Ch01 Ch01 Ch01285
0504 696
o1042
o1128
01198
01282 0 1401 0 1424 , 1497
c1592
(0)1634 1859
o1905 0
1923 p 2010 2022
O2172 0 2435
02445 e 2514
O2634 2666 3157
O3265 p 3357
Z3460 3757
e3801
©3815
6-1
3908
09458
rÒ9546 0 9690
V9774
o9911 r
_,,
10007
O10124
o10234 , 19722
O19733
O19734
O19858 0 20126 0 20205 21201 21208 28263 r 28315 . 28385
g. )28409
o28460
45439
0e,
28475
45565
28488 0 28535 0 36904 036963 37135 p 37227 p 41399
p41498 41522
O41639
U41653
O41740
O45590
O45664
p45705
047117
O47130
O53732
58719
O
0
53779
58720
O
0
53783
58922 0
53788
59027
00
53808
59118
G
0
53978
054090
59264
o57157
O57196 n 57210
O57239
O57297
O57383 58692
459150
U 066955
067C26 67134
Q.)67167
(à...)
67220 0
53795
Ch01
053826
Ch01
054007
Ch01
54020
Ch01
o
54022
Ch0154071
Ch01
o57085
Ch0157112
Ch01
a
57167
Ch0157221
Ch01
0 57250
Ch01
a57267
Ch0157305
Ch01
o57320
Ch01
4
57361 57364
o
57379
&58660
o58736
058740
p58769 o 58923
g66983
o67059
%
67062 o 67087
%67090 o 67245
o67301
0
72866
U
72876
o72880
0
72932
6)73020
073083
0
73094
°73099
o73107
czp77995
078006
478020
078064
%
78115
078119
078176
0
78186
078195
078217
0
78235
8
78241
©84154
084303
°84347
084349
084382
o84422
084465
&84479
0
90314
p90341
090352 90359
o90402
090459
o90467
O90600
p90608
o90609
p90624 a 90632
o94476
O
94480
094484
0
94487 0 94508
D94561
0
94597 947?1 94726
`A94728
094732 94739
o94762 94781
cz)
431
Ch01439
Ch01
4i1
640
Ch01
,..
rti779
Ch01818
Ch01
,Y1069
Ch01
ç-
1429
Ch01
..
c,:.
1551
Ch01
R;
1610
Ch01
T1681
Ch01
i+ÿ
1711
Ch01
i' -'
1967
Ch01. 213C
Ch01
..
`"33
Ch01
r,3213
C.;
3722
`t9475 `;
.f,.-:
9602 9625 9658
!'10135 . 10227 10232
i,10261 10340
:;`:'9 36 'Q '97,6
..19:.77
1
20011
G21339 28493
41422
67067
14,
35127
41616
67174
c'i'
, i ,of
35468
41658
67197
..
1,13',J
Itt`-i
35723
47064
67217
.{.L
,,
ult.
35769
47167
72852
35793
47297
72854
.r
e21'
35847
47406
72941
ÿ<
.IV.
j`'
36139
53847
7/947
C.
e414,1
36155
57035
73089
. -
O`
; JcA,
36200
51254
78010
'I:
i'36211 1, 36840
37040
59095
.
41383
59329
~':r
57348
78192
61
58929
I ,
78226
I..,
84126 84220
t
Vt
84398
484432
ç95320
*t3
95351
=!:95377
98530
:
95732
98552
*,
^̀'
95818
98677
!_ld+Ti
96044
98726
L!7
i' `\_
96323
98729
t1
-'.
96639
99045
!it..
96790
,-y-
5{68/1
~ ''-A -Y
1`s
97376 97385
`'.:.97707 98082
G.
98465
.!",99212
i .:.99243
.-.(
. -t
'556
44*
2552
383'
ir;
42'.
44386
r
4753
to,
`57'
$J
5545
)
475
L'251
fit
2:('3
2555
14
42'5
4529
4917
5228
57.'
._:4
729
1453
2017
as: ,5
3700
4240
4544
%.
4921
5279
5357
1070
1484
2304
2863
3411
3994
fitt
4337
47
4962
(I)5328
5896
._52
1133
1493
2502
2916
3515
4067
e4345
4745
4965
5435
5929
1.5
5074
,.1
Figure 1: (Top Row) Typical RBC images produced by the ImageStream®X IFC system, including unfocused (A), multi-cell (B), cells with calibration beads (C), front view discocytes (D, E), side view discocytes (F), side view spheroechinocytes (G) and front view spheroechinocytes (H, I). (Second Row and Below) Sequential gating based on various mask/feature combinations was used to exclude objects of type A (I.), B (II.), C (III.), F and G (IV.), and differentiate populations D and E from H and I to classify discocytes and spheroechinocytes, respectively.
While multi-cell images were excluded from further analysis (Figure 3), a small but significant fraction of object images containing calibration beads (speed beads) were not successfully removed (image C, section II, Figure 1). The presence of these beads impaired mask application and subsequent feature analysis and therefore needed to be removed. Figure 4 displays the plot used to discriminate between images containing RBCs with calibration beads and single RBC images, as illustrated in section III (Figure 1). Here the area feature was plotted against the circularity feature to show the relative size of the object in relation to its deviation from a circle22. A gate was created to exclude images containing RBCs with calibration beads from subsequent analysis. Figure 5 displays the plot used to discriminate between front view and side view RBC images, as illustrated in section IV (Figure 1). The IDEAS® feature finder tool was used to statistically determine the best features to discriminate front view and side view RBC images. To do this, approximately 50-100 truth populations were chosen for both front and side view RBC images. The feature finder calculated the RD mean (Fisher’s discriminant ratio) – the difference in the means divided by the sum of the standard deviations for the two populations - for each feature22. Features with higher RD values equated
I.
II.
III.
IV.
V.
B A C D E F G H I
Proc. of SPIE Vol. 10076 100761F-4
Downloaded From: http://proceedings.spiedigitallibrary.org/pdfaccess.ashx?url=/data/conferences/spiep/91601/ on 05/27/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
2 7
jL
;3 5
J o
2'
J23 4 26
? h27
f)
23
j2
J33
a
42
" 1
46 2
11V
54 57
o60
63 65 68
o
70
75
cÿ
83 87
t ,
95 ca
92® ?94
102
J112 115
17 t81
t22
'25
;41
o.
!
;28
C1
!44
\J
t38
/-'
145
140
o
o154
1. 171
All
5-`1ck`tió ni
..
4-coNg-3-wu_-0°' 2-N
EOz 1-
0, f
0 20 40 60 80 100
Gradient RMS M01 BF
7109
o37234 37237
37421
o
o
38010
058742
o
59157
eVENE
60425
<r
l
37428
ll
ÇLJ
40
53760
(Z!,1
595.
59811
C:-
5992:
i
58789
Q
53845
I.
Gradient RMS
Normalized
Relative Freq
uency
Figure 2: (see section I, Figure 1); histogram of gradient RMS (counts of normalized relative frequency) within the default bright field mask. Inset images provide examples of unfocused (left) and focused (right) RBC images. Value of gradient RMS of images are indicated by red arrows. Images containing gradient RMS values above 54 were selected for further analysis (horizontal light blue line).
Figure 3: (see section II, Figure 1); plot of aspect ratio versus area using the default bright field mask. Inset images provide examples of front view (top right), multi-cell (middle right) and side view (bottom left) RBC images. Object masks are highlighted in light blue. Red arrows point to the location of images within the scatter plot. Color scale consists of, in order of increasing relative population density: blue, green, yellow, orange, red. A gate based on images within the ROI (light green boundary) was selected for further analysis.
II. focus
single cellssingle cellssingle cellssingle cellssingle cellssingle cells
90 18060 150120Area_M01
0.6
Aspect Ratio_M01
Aspect Ratio
Area [µμm%]
Proc. of SPIE Vol. 10076 100761F-5
Downloaded From: http://proceedings.spiedigitallibrary.org/pdfaccess.ashx?url=/data/conferences/spiep/91601/ on 05/27/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
Ch0116104
Filih41/11b
Object no bead
not side discocytesnot side discocytesnot side discocytes
Side DiscocytesSide DiscocytesSide Discocytes
not side discocytesnot side discocytesnot side discocytes
Side DiscocytesSide DiscocytesSide Discocytes
0.6Aspect Ratio_M01
9
6
3
Minor Axis Intensity_M01_Ch01
Aspect Ratio_M01, Minor Axis Intensity_M01_Ch01Population Count %GatedObject no bead & single cells & focus 22101 100 not side discocytes & Object no bead & single cells & focus 13978 63.2 Side Discocytes & Object no bead & single cells & focus 8024 36.3
Minor Axis
Aspect Ratio
Figure 4: (see section III, Figure 1); plot of area versus circularity using the default bright field mask. Inset images provide examples of cells with calibration beads (top left), front view (bottom right) and side view (bottom left) RBC images. Object masks are highlighted in light blue. Red arrows point to the location of images within the scatter plot. The color scale consists of, in order of increasing relative population density: blue, green, yellow, orange, red. A gate based on images within the right-side ROI was selected for further analysis.
Figure 5: (see section IV, Figure 1); plot of minor axis vs. aspect ratio using the default bright field mask. Inset images provide examples of the front view (bottom right) and side view (top left) RBC images. Object masks are highlighted in light blue. Red arrows point to the location of images within the scatter plot. Sample images (light green) are overlaid with truth populations of the front view (purple) and side view (red) images. A gate based on images within the right-side ROI was selected for further analysis.
single cells
object with beadobject with beadobject with bead
Object no beadObject no beadObject no bead
object with beadobject with beadobject with bead
Object no beadObject no beadObject no bead
20 3010 400Circularity_M01
60
90
120
Area_M01
Circularity_M01Population Countsingle cells & focus 22568 object with bead & single cells & focus 452 Object no bead & single cells & focus 22101
Area
Circularity
III.
IV.
Proc. of SPIE Vol. 10076 100761F-6
Downloaded From: http://proceedings.spiedigitallibrary.org/pdfaccess.ashx?url=/data/conferences/spiep/91601/ on 05/27/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
60702t.. .
61074
77298
BF BF BF BF
431
897
439it
59075
59739
60021
60436attep
77081
N78226
%L.
79681,.80414
A-+.
1044
59119
58893
`.Y
58894e.t 589291 #
59156li
59817.r,
60081
60463
59243,
59617
59849a»,,j_.
59884 59983_(44(
60123 60301 60350
\?C.)
60500
78085 78094
60:
78245 79002
464,
79324 79612,t.
79753
80438ti
79838ter
80279.f..80283
80730
;It
80966
~T
81210("PA;
Figure 6: (see section V, Figure 1); plot of H contrast versus H variance using the default bright field mask. Inset images provide examples of front view discocyte (bottom right) and front view spheroechinocyte (top left) RBC images. Default bright field masks are highlighted in light blue. Red arrows point to the location of images within the scatter plot. Sample images (light green) are overlaid with truth populations of the front view discocytes (purple) and front view spheroechinocytes (yellow) images. A gate based on truth populations segregated the two subpopulations.
to higher discriminatory power and hence better subpopulation separation. In this way, it was determined that aspect ratio and minor axis features would provide good discrimination between front and side view RBCs. The IDEAS® default bright field masks enclosing side view images (Figure 1: C, D) could not provide sufficient detail to segregate discocytes and spheroechinocytes using the feature finder. Side view images were hence excluded from analysis. On average, approximately 40% of the single cell image population consisted of side view images. The general trends in discocyte/spheroechinocyte population changes over time are speculated to be unaffected by the exclusion of this subpopulation; however, this exclusion poses limitations when interpreting any quantitative analysis of discocyte/spheroechinocyte populations. Future work will investigate masks that contour the cell boundary more tightly without losing necessary image detail. Figure 6 displays the plot used to discriminate between front view discocyte and spheroechinocyte images, as illustrated in section V (Figure 1). A similar strategy employing the feature finder tool (Figure 5) was used to determine the best features to discriminate front view discocytes and front view spheroechinocytes. The strongest statistical discriminators were found to be Haralick (H) contrast and H variance22. These parameters were named after a set of texture features defined by Haralick et al. to normalize and characterize the local variation in intensity distribution within the masks23. A key contributor to normalized local variance in RBC images was the central pallor present in discocyte images. Hence, the resulting H contrast versus H variance plot placed the central pallor-lacking masks of spheroechinocyte images towards the lower left corner.
not side discocytes
echinocyte/sperocyteechinocyte/sperocyteechinocyte/sperocyte
full discocytefull discocytefull discocyte
echinocyte/sperocyteechinocyte/sperocyteechinocyte/sperocyte
full discocytefull discocytefull discocyte
12 153 6 9combined H varience mean and std
2e4
1e4
4e4
0
3e4H contrast mean and intensity thresh
combined H varience mean and std, H contrast mean and intensity threshPopulation Count %Gatednot side discocytes & Object no bead & single cells & focus 13978 100 echinocyte/sperocyte & not side discocytes & Object no bead & single cells & focus 10650 76.2 full discocyte & not side discocytes & Object no bead & single cells & focus 3269 23.4
H Con
trast
H Variance
V.
Proc. of SPIE Vol. 10076 100761F-7
Downloaded From: http://proceedings.spiedigitallibrary.org/pdfaccess.ashx?url=/data/conferences/spiep/91601/ on 05/27/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
3.2 Image acquisition and processing speed Upon loading one RCC vial for automated aspiration into the ImageStream®X, a raw image data file was generated in approximately 5 minutes. Each saved raw data file contained 100,000 brightfield images and took up approximately 500MB of storage space. Uploading and automated processing of the raw data file using the morphology template (Section 3.1) took under 5 minutes. Assuming that the measuring apparatus and hardware are within one location, the relative fraction of spheroechinocytes within a sample population of approximately 25,000-35,000 cells could be acquired in less than 15 minutes after RCC extraction. In contrast, the light-microscopy morphology assay requires roughly half an hour after RCC extraction to acquire the relative fraction of spheroechinocytes within a sample of 100 cells. Evidence for the potential advantages of this automated RBC morphology characterization technique is provided through the enhancement in acquisition and processing speed. 3.3 Time-series study Figure 7 plots the IFC time series measurements of the spheroechinocyte population percentages acquired from two RCC blood bags (labeled 1 and 2, dashed lines). Also plotted on the figure are average RBC populations used for analysis, which ranged between 25,000 and 35,000 images (solid line). During the lifespan of bags 1 and 2, template processing of IFC images measured an average spheroechinocyte population increase of 38.7% and 35.5% respectively. Both bags were measured to have a starting spheroechinocyte percentage of 55.7% on day 2; however, the percentage values may differ if the side view RBC images are included in the analysis (Section 3.1). One hypothesis to explain the variation in the rate of spheroechinocyte population growth between the bags was that the rate might potentially be dependent on the quality of RBCs provided by the donor. It is suggested that by tracking the rate of change of spheroechinocyte population, the severity of the RBC storage lesion may be assessed. To test this hypothesis, future work will concentrate on establishing quantitative links between spheroechinocyte population increase and other characteristics of the RBC storage lesion.
Figure 7: Average analyzed RBC populations (solid) and spheroechinocyte population percentages of samples of RCC blood bags 1 (dashed) and 2 (dash-dotted) as a function of storage day. Error bars represent the standard deviation of 3 samples.
Validation against current lab techniques is necessary. Conventional morphology assessments segregate RBCs into six subcategories and use a morphology index to provide a measure of sample quality2,14. Future work will expand on the current template to segregate RBCs into these six subcategories, which will involve improving masks to contour cell
0 5 10 15 20 25 30 35 40 45STORAGE DAY
50
55
60
65
70
75
80
85
90
95
100
SPH
ERO
ECH
INO
CYT
E PO
PULA
TIO
N [%
]
0
1
2
3
4
5
6
7
8
9
10
AVER
AGE
ANAL
YZED
RBC
PO
PULA
TIO
N
104
RCC Blood Bag 1 Spheroechinocyte %RCC Blood Bag 2 Spheroechinocyte %Average Analyzed RBC population
Proc. of SPIE Vol. 10076 100761F-8
Downloaded From: http://proceedings.spiedigitallibrary.org/pdfaccess.ashx?url=/data/conferences/spiep/91601/ on 05/27/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
membranes more closely (Section 3.1). Using this improved image segmentation template, the IFC method can be compared to the conventional techniques to determine the level of agreement. Furthermore, the inclusion of side view images for analysis (Section 3.1) can allow for a quantitative investigation of the influence of donor characteristics on RBC quality.
4. CONCLUSIONS This paper presents the potential of using IFC as a highly automated, statistically robust and objective technique for morphologically characterizing RBCs, in contrast to current methods that are labor intensive and are highly subjective. An image segmentation template was developed on the IDEAS® software platform to determine the relative spheroechinocyte population percentage. Measurements showed a spheroechinocyte population increase of 38.7% and 35.5% for two RCC blood bags over the 42-day preservation period. Variation in the rate of change of spheroechinocyte population suggests that this parameter may potentially be donor-dependent; however, such inferences are limited by the fact that side view images of RBCs could not be included in the analysis. Compared to the conventional method, this novel technique was found to have a significantly faster image acquisition and processing speed. More significantly, the 250-350 times increase in the analyzed sample size allows for an improvement in the statistical robustness of morphological characterization. Future work will concentrate on improving the masks used to select RBCs for analysis. The current template will be expanded to segregate RBCs into six subpopulations; this new template will be used to validate the IFC method against conventional techniques. Quantitative links between results obtained from the IFC method and current techniques in the characterization of RBC storage lesions will be investigated. It is thought that the monitoring of RBC morphology distribution change as a function of time can provide an indication of the RBC storage lesion severity in RCC blood bags during storage.
ACKNOWLEDGMENTS Funding for this project was provided through a Collaborative Health Research Projects (CHRP) grant (application # 315271), a joint initiative between the Natural Sciences and Engineering Research Council of Canada (NSERC) and the Canadian Institutes of Health Research (CIHR) awarded to the principle investigators M. C. Kolios and J. P. Acker. T.R. Turner is gratefully acknowledged for consultations associated with the development of the RBC image segmentation template. The Lunenfeld Tanenbaum Research Institute (LTRI) flow cytometry facility is acknowledged for providing access for image flow cytometry experiments.
REFERENCES
[1] Carson, J. L., et al., "Red blood cell transfusion: a clinical practice guideline from the AABB," Ann. Intern. Med. 157(1), 49-58 (2012).
[2] Acker, J. P., et al., "A quality monitoring program for red blood cell components: in vitro quality indicators before and after implementation of semiautomated processing," Transfusion 54(10), 2534-2543 (2014).
[3] Högman, C. F., et al., "Red cell Suspensions in SAGM Medium. Further experience of in vivo survival of red cells, clinical usefulness and plasma-‐saving effects," Vox Sang. 45(3), 217-223 (1983).
[4] Dumont, L. J., and AuBuchon, J. P., "Evaluation of proposed FDA criteria for the evaluation of radiolabeled red cell recovery trials," Transfusion 48(6), 1053-1060 (2008).
[5] Gabrio, B. W., et al., "Erythrocyte preservation. I. The relation of the storage lesion to in vivo erythrocyte senescence," J. Clin. Invest. 33(2), 242-246 (1954).
[6] Högman, C. F. and Meryman, H. T., "Storage parameters affecting red blood cell survival and function after transfusion," Transfus. Med. Rev. 13(4), 275-296 (1999).
[7] D'Alessandro, A., et al., "Red blood cell storage: the story so far," Blood Transfus. 8(2), 82-88 (2010). [8] Bessis, M., Weed, R.I. and Leblond, P. F., [Red cell shape: Physiology, Pathology, Ultrastructure], Springer Verlag,
New York, Heidelberg and Berlin, 35-37 (1973). [9] Hess, J. R., "Red cell changes during storage," Transfus. Apher. Sci. 43(1), 51-59 (2010). [10] Chaudhary, R. and Katharia, R., "Oxidative injury as contributory factor for red cells storage lesion during twenty-
eight days of storage," Blood Transfus. 10(1), 59-62 (2012). [11] Antonelou, M. H., et al., "Effects of pre-storage leukoreduction on stored red blood cells signaling: a time-course
evaluation from shape to proteome," Journal Proteomics 76, 220-238 (2012).
Proc. of SPIE Vol. 10076 100761F-9
Downloaded From: http://proceedings.spiedigitallibrary.org/pdfaccess.ashx?url=/data/conferences/spiep/91601/ on 05/27/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
[12] Salzer, U., et al., "Vesicles generated during storage of red cellsare rich in the lipid raft marker stomatin," Transfusion 48(3), 451-62 (2008).
[13] Frank, S.M., "Decreased erythrocyte deformability after transfusion and the effects of erythrocyte storage duration," Anesth. Analg. 116(5), 975–981 (2013).
[14] Usry, R. T., Moore, G. L., and Manalo, F. W., "Morphology of stored, rejuvenated human erythrocytes," Vox Sang. 28(3), 176-183 (1975).
[15] Tchir, J. D. R., Acker, J. P. and Holovati, J. L., "Rejuvenation of ATP during storage does not reverse effects of the hypothermic storage lesion," Transfusion 53(12), 3184-3191 (2013).
[16] Palmer, L., et al., "ICSH recommendations for the standardization of nomenclature and grading of peripheral blood cell morphological features," Int. J. Hematol. 37(3), 287-303 (2015).
[17] Longster, G. H., et al., "Scanning electron microscope studies of red cell morphology," Vox Sang. 22(2), 161-170 (1972).
[18] Watson, J. V., "The early fluidic and optical physics of cytometry," Cytometry Part A 38(1), 2-14 (1999). [19] Shapiro, H. M., [Practical flow cytometry, 4th Ed.], John Wiley & Sons, New Jersey, 2-12 (2003). [20] Sklar, L. A., [Flow Cytometry for Biotechnology], Oxford University Press, New York, 57-67 (2005). [21] McGrath, K. E., Bushnell, T. P., and Palis, J., "Multispectral imaging of hematopoietic cells: where flow meets
morphology," J. Immunol. Methods. 336(2), 91-97 (2008). [22] Amnis Corporation, [Image data exploration and analysis software user’s manual, Version 4.0], Seattle, 123-205
(2009). [23] Haralick, R.M., Shanmugan, K. and Dinstein, I., "Textural Features for Image Classification," IEEE Trans. Syst.,
Man, Cybern., Syst. 3(6), 610-621 (1973).
Proc. of SPIE Vol. 10076 100761F-10
Downloaded From: http://proceedings.spiedigitallibrary.org/pdfaccess.ashx?url=/data/conferences/spiep/91601/ on 05/27/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
top related