In this paper, we present a method to automatically quantify the threedimensional (3D) volume of red blood cells (RBCs) using offaxis digital holographic microscopy. The RBCs digital holograms are recorded via a CCD camera using an offaxis interferometry setup. The RBCs’ phase image is reconstructed from the recorded offaxis digital hologram by a computational reconstruction algorithm. The watershed segmentation algorithm is applied to the reconstructed phase image to remove background parts and obtain clear targets in the phase image with many single RBCs. After segmenting the reconstructed RBCs’ phase image, all single RBCs are extracted, and the 3D volume of each single RBC is then measured with the surface area and the phase values of the corresponding RBC. In order to demonstrate the feasibility of the proposed method to automatically calculate the 3D volume of RBC, two typical shapes of RBCs, i.e.,
stomatocyte/discocyte
, are tested via experiments. Statistical distributions of 3D volume for each class of RBC are generated by using our algorithm. Statistical hypothesis testing is conducted to investigate the difference between the statistical distributions for the two typical shapes of RBCs. Our experimental results illustrate that our study opens the possibility of automated quantitative analysis of 3D volume in various types of RBCs.
I. INTRODUCTION
Threedimensional (3D) holographic imaging systems
[1

10]
have been investigated for achieving reliable and noninvasive recognition and for identifying biological micro /nano organisms
[11

14]
. The 3D holographic imaging system for biological microorganism analysis has a large number of biological applications such as medical diagnosis, medical therapeutics, biomedical imaging, defense, and security.
Red blood cells (RBCs) have been extensively studied for biomedical applications. Recent studies on RBCs have suggested that transfusions of blood having long storage periods or abnormal shapes increase risk after transfusion
[15
,
16]
since long storage and abnormal shapes progressively change the RBCs’ structure and function, which may reduce RBC function and viability
[15]
. Consequently, the development of free labelling techniques that allow rapid classification of older versus newer red blood cells is very useful. In general, most traditional methods of studying biological cells such as RBC require biochemical processing, which may be timeconsuming and labor intensive. Furthermore, the recognition techniques based on twodimensional imaging may not be effective because RBCs are usually semitransparent. Thus, the automatic 3D classification of RBCs can be very useful. Otherwise, the RBCs would be investigated in very low contrast or invisible under conventional microscopy.
Recently developed techniques in 3D holographic microscopy have provided a high resolution and noninvasive examination to identify biological cells
[11

14]
. In this paper, we propose an automated algorithm to quantitatively measure a 3D volume of RBC using digital holographic microscopy for the purpose of identifying various types of RBCs. The proposed technique for quantitative analysis of a 3D volume of RBCs includes 3D sensing, visualization, segmentation, and statistical methods. Offaxis digital holographic microscopy
[17

20]
is applied to threedimensionally sense the RBCs
[14]
. Their original phase images are restored by using special numerical algorithms. As part of preprocessing RBCs 3D volume analysis, it is important to remove the background parts and select regions of interest in the restored RBCs’ phase image
[14]
. Here, the watershed segmentation algorithm
[21]
is applied to the reconstructed phase image to obtain clear targets in the phase image with many single RBCs. After segmenting the reconstructed RBC’ s phase image, all single RBCs are extracted and the 3D volume of each single RBC is then measured via the surface area and the phase values of the corresponding RBC. In order to show the feasibility of the proposed method, two typical shapes of RBCs, i.e.,
stomatocyte/discocyte
, are tested via experiments. The statistical distributions of 3D volume for each class of RBC [
stomatocyte or discocyte
] are generated by using our algorithm. Statistical hypothesis testing
[22
,
23]
is conducted to investigate the differences between the statistical distributions of 3D volume for the two typical shapes of RBCs. The experiment results illustrate that the proposed method based on digital holographic microscopy can quantitatively analyze the 3D volumes of various types of RBCs.
II. AUTOMATIC PROCEDURES FOR RBCs 3D VOLUME MEASUREMENT
The offaxis digital holographic imaging system
[24
,
25]
is employed for 3D sensing of RBCs, which has been described in
[20]
. As shown in
Fig. 1
, the offaxis digital holograms of RBCs are acquired via a CCD camera. In this configuration, a laser beam with a wavelength of 682 nm is divided into reference and object beams. The object beam is diffracted by the RBC specimens. A 40× / 0.75NA microscope objective (MO) magnifies the diffracted object beam. The interference patterns between the diffracted object
Experimental setup for recording the offaxis digital hologram of red blood cells.
and reference beams in the offaxis geometry are recorded by the CCD camera. The numerical algorithms described in
[17
,
18]
reconstruct the RBCs’ wavefront or phase image from the offaxis digital hologram.
In order to measure the RBC’ s 3D volume in the reconstructed phase image, the watershed segmentation algorithm
[21]
is applied to the phase image in order to remove background parts and obtain clear targets in the phase image with many single RBCs
[14]
. Every single RBC in the phase image is extracted individually. Then, the projected surface area and phase values are automatically calculated for every single RBC, which are utilized to calculate the 3D volumes of the corresponding RBCs. The 3D volume measurement for a single RBC, which is related with its 3D shape, can be given by:
where
N
is the total number of pixels within a single RBC,
p
denotes the pixel size,
φ
is the phase value of each pixel within single RBC, and
M
is the magnification of digital holographic microscopy. In our experiments, the RBC refractive index,
n_{rbc}
, is 1.396 and the index of refraction of the HEPA medium,
n_{m}
, is 1.3334, respectively. In order to investigate whether or not there is a statistically significant distinction between 3D volumes of different shapes of RBCs, the statistical distribution for the 3D volume of RBCs is generated by calculating Eq. (1) with every single RBC in the segmented phase image. Finally, parametric statistical hypothesis testing is conducted for a statistical decision.
III. EXPERIMENTAL RESULTS
In our experiments, the offaxis digital holograms of two different shapes of RBCs [class 1: RBCs with a
stomatocyte
shape and class 2: RBCs with a
discocyte
shape] were recorded via a CCD camera as shown in
Fig. 1
[26]
. The resolution of the CCD is 1024× 1024 and the pixel size is 10 μ m. The phase images of RBCs with different shapes were reconstructed from their own offaxis digital holograms using numerical algorithms.
Figure 2
shows the phase images of two typical shapes of RBCs obtained via offaxis digital holographic microscopy. These phase images of the two typical shapes of RBCs [
stomatocyte/ discocyte
] were used to exhibit the difference between the RBCs 3D volume and their shapes. For the purpose of RBCs 3D volume analysis, a watershed segmentation algorithm was applied to the RBCs phase images. All single RBCs in the phase image were obtained via the segmentation algorithm.
Figure 3
shows the segmented phase images of class 1 and class 2 RBCs, respectively.
Since most single RBCs appear to have two kinds of
The reconstructed RBCs phase images. (a) Class 1: RBCs having a stomatocyte shape (b) Class 2: RBCs having a discocyte shape.
The segmented RBCs phase images. (a) Class 1: RBCs having a stomatocyte shape (b) Class 2: RBCs having a discocyte shape.
A schematic diagram for three different regions in red blood cells (RBCs).
regions as shown in
Fig. 4
, it is better to respectively separate the different parts to achieve a comprehensive and detailed analysis of the RBCs 3D structure. Here, different regions of RBCs are defined in
Fig. 4
. The C region includes A and B parts, where the A part apparently has relative lower phase values than those of B part. Most of the A part is on the interior of the RBC, while the B part forms the boundary of the RBC.
In order to obtain the different regions in the RBCs, we first use the markercontrolled watershed algorithm to obtain the C part. Then, the C part is taken as a source
The segmented phase images of the A and B regions [see Fig. 4] in the RBCs. (a) Segmented phase image of the A part in the class 1 RBCs having a a stomatocyte shape. (b) Segmented phase image of the A part in the class 2 RBCs having a discocyte shape. (c) Segmented phase image of the B part in the class 1 RBCs having a stomatocyte shape. (d) Segmented phase image of the B part in the class 2 RBCs having a discocyte shape.
image, and we can obtain the A part by applying the markercontrolled watershed. Finally, the B part can be obtained by subtracting the A part from the C part.
Figure 5
shows the segmented phase images of the A and the B parts in the reference and input RBCs, respectively.
To conduct a quantitative investigation of the 3D volumes of RBCs with two different shapes, the 3D volumes in the A, B, and C parts [see
Fig. 4
] of single RBC are calculated using Eq. (1), respectively. To conduct a statistical analysis of RBCs volume, the mean and standard deviation of the volume of the A part in RBCs with a
stomatocyte
shape are computed to be 7.6 μ m
^{3}
and 7.2 μ m
^{3}
, while these values in RBCs with
discocyte
shape are 14.6 μ m
^{3}
and 8.0 μ m
^{3}
, respectively.
Figure 6
shows the statistical distributions of RBC volume of the A part in the two different shapes of RBCs.
Figure 7
shows the statistical distributions of 3D volume of the B part in the two different shapes of RBCs. The mean and standard deviation of the volume of the B part in RBCs with a
stomatocyte
shape are computed to be 41.5 μ m
^{3}
and 14.7 μ m
^{3}
, while these values in RBCs with a
discocyte
shape are 32.7 μ m
^{3}
and 7.3 μ m
^{3}
, respectively.
Figure 8
shows the statistical distributions of 3D volume of the C part in the two different shapes of RBCs. The mean and standard deviation of the volume of the C part in RBCs with a
stomatocyte
shape are computed to be 78.7 μ m
^{3}
and 17.9 μ m
^{3}
, respectively, while these values in RBCs with
discocyte
shape are 90.0 μ m
^{3}
and 25.3 μ m
^{3}
, respectively. It is noted that the average volume in an RBC
Statistical distributions of the 3D volume in A part for two typical shapes of RBCs. The solid line is RBCs having a stomatocyte shape. The dotted line is RBCs having a discocyte shape.
Statistical distributions of the 3D volume in B part for two typical shapes of RBCs. The solid line is RBCs having a stomatocyte shape. The dotted line is RBCs having a discocyte shape.
Statistical distributions of the 3D volume in C part for two typical shapes of RBCs. The solid line is RBCs having a stomatocyte shape. The dotted line is RBCs having a discocyte shape.
with
discocyte
shape is larger than that in
stomatocyte
shape. This may be due to the differences of projected surface area and 3D shape of RBC.
In order to statistically evaluate there is a significant 3D volume distinction in RBC populations that differ in shape, we define the 3D volumes in the A, B, and C parts of each single RBC as random variables
X
_{1}
,
X
_{2}
, and
X
_{3}
. The statistical distribution of the difference between two location parameters from the two different classes [c1 and c2] of RBC populations [c1: RBCs with a
stomatocyte
shape, c2: RBCs with a
discocyte
shape] is calculated. It is assumed as shown in
Fig 6
,
7
, and
8
that the two RBC populations of each random variable [A, B or C part] follow normal distributions
N_{c}
_{1}
(
μ
_{c}
_{1}
,
σ
^{2}
_{c}
_{1}
) and
N_{c}
_{2}
(
μ
_{c}
_{2}
,
σ
^{2}
_{c}
_{2}
), respectively.
In order to compare the dispersion parameters between the two statistical normal distributions [c1: stomatocyte shape, c2: discocyte shape], it is assumed that all four statistical parameters are unknown and
θ
= (
μ _{c}
_{1}
,
μ _{c}
_{2}
,
σ
^{2}
_{c1}
,
σ
^{2}
_{c2}
). The ratio of the dispersions of two independent normal populations can be represented with a null hypothesis [H0:
σ
^{2}
_{c1}
=
σ
^{2}
_{c2}
=
σ
^{2}
] as follows
[22]
:
where the
F
distribution has
n_{c}
_{1}
 1 and
n_{c}
_{2}
1 degrees of freedom,
n_{c}
_{1}
and
n_{c}
_{2}
are the sample sizes of the c1 class and c2 one, respectively.
V
[•] is a sample variance of random variable
X
_{i}
. The subscript
i
can be 1, 2, or 3.
To answer and discuss this important question on whether or not there is a distinction in 3D volume of A, B, or C part between the two different shapes of RBCs [c1:
stomatocyte
shape, c2:
discocyte
shape], the Ftest [null hypothesis:
H
_{0}
(
σ
^{2}
_{c1}
=
σ
^{2}
_{c2}
=
σ
^{2}
)] for each random variable
X
_{1}
,
X
_{2}
, or
X
_{3}
was first conducted on the basis of a twotailed test with a specific level of significance. Statistical pvalue was calculated by using the table of
Fisher F
distribution in
[22]
for the statistical decision. It is noted that the computed statistical pvalues for the random variables
X
_{1}
,
X
_{2}
, and
X
_{3}
are 0.0608, 2.66× 10
^{10}
, and 4.34× 10
^{4}
, respectively, which are the probability that the observed test statistic of Eq. (2) would occur in the same population [3D volume] in parts A, B, and C of RBC. From these experimental results, it would be concluded that the variability of the 3D volume in the A part of class 1 is not different from that of B since we can accept the null hypothesis,
H
_{0}
(
σ
^{2}
_{c1}
=
σ
^{2}
_{c2}
=
σ
^{2}
), at the 0.05 level of significance [statistical pvalue of the
Ftest
over the random variables
X
1 is 0.0608]. On the other hand, we may conclude that there is a significant difference in the variability of the 3D volumes in the B and C parts since both pvalues of the
Ftest
over the random variables
X
_{2}
,
X
_{3}
are much lower than 0.05.
The difference of location parameters of the two independent normal populations for the null hypothesis [
H
_{0}
:
μ _{c}
_{1}
=
μ
_{c2}
=
μ
)] can be represented as follows
[22]
:
where the variable
T
has
student t
distribution with
n_{c}
_{1}
+
n_{c}
_{2}
 2 degrees of freedom.
n_{c}
_{1}
and
n_{c}
_{2}
are the sample sizes of the c1 class and c2 one, respectively.
E
[?] is a sample mean of random variable
X
_{i}
. The subscript
i
can be 1, 2, or 3.
[
X_{i}
(
c
1)] and
[
X_{i}
(
c
2)] are the unbiased estimates of
σ
^{2}
_{c1}
and
σ
^{2}
_{c2}
, respectively.
The statistical decision analysis on the equality of the location parameter between the statistical distributions of the two different shapes of RBCs for each random variable
X
_{1}
,
X
_{2}
, or
X
_{3}
, was performed using the statistical hypothesis testing [
ttest
]. Finally, the statistical pvalue was calculated for the statistical decision on whether or not there is significant difference between the location parameters of the two statistical distributions. The statistical pvalue was obtained by using the table of
Student’s t
distribution in reference
[22]
. It is noted that the computed statistical pvalues for the random variables
X
_{1}
,
X
_{2}
, and
X
_{3}
are 3.36× 10
^{8}
, 2.22× 10
^{6}
, and 7.29× 10
^{5}
, respectively, which are the probability that the observed test statistic of Eq. (3) would occur in the same population [3D volume] in parts A, B, and C of RBC. Therefore, we may conclude from these experimental results that the distinction in the 3D volumes of the A, B, or C parts in RBC [see
Fig. 4
] is probably significant because we can reject the null hypothesis,
H
_{0}
:
μ
_{c1}
=
μ
_{c2}
=
μ
at the 0.05 level of significance.
These statistical experiments demonstrate that there is a considerable distinction between the RBCs 3D volumes and their shapes. Accordingly, we believe that the proposed method has potential for achieving statistical quantifications of the 3D volume in various types of RBCs. In addition, it can be a promising tool for investigating any modifications or dynamics of 3D volume in RBC by monitoring RBCs.
IV. CONCLUSION
In this paper, we have proposed an automated method for 3D volume measurement of red blood cells (RBCs) using offaxis digital holographic microscopy. The offaxis digital holographic microscopy has recorded hologram patterns of RBCs. The RBCs phase images have been reconstructed from their holograms by using special numerical algorithms. Segmentation algorithms have been applied to the RBCs phase image in order to remove the unnecessary background and obtain clear targets in the reconstructed phase image. The 3D volumes of every single RBC in the segmented phase image have been calculated. Statistical hypothesis testing has been conducted in order to investigate whether or not there is a significant difference between the statistical distributions of the 3D volume of two different shapes of RBCs, i.e.,
stomatocyte
and
discocyte
. The experimental results exhibit that there is a considerable distinction between the RBCs 3D volumes and their shapes. Therefore, we believe that the proposed method provides valuable insight into the 3D volume in RBCs that differ in shape and can be extended to the classification of various types of RBCs.
Acknowledgements
This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (20090088195). We thank Daniel Boss, Pierre Marquet from Ecole Polytechnique Fédérale de Lausanne (EPFL), Switzerland, for their help with experiments.
Yamaguchi I.
,
Zhang T.
(1997)
“Phaseshifting digital holography”
Opt. Lett.
22
1268 
1270
Nomura T.
,
Murata S.
,
Nitanai E.
,
Numata T.
(2006)
“Phaseshifting digital holography with a phase difference between orthogonal polarizations”
Appl. Opt.
45
4873 
4877
Frauel Y.
,
Naughton T.
,
Matoba O.
,
Tahajuerce E.
,
Javidi B.
(2006)
“Three dimensional imaging and display using computational holographic imaging”
Proc. IEEE
94
636 
654
Zhang Y.
,
Pedrini G.
,
Osten W.
,
Tiziani H. J.
(2004)
“Reconstruction of inline digital holograms from two intensity measurements”
Opt. Lett.
29
1787 
1789
Ferraro P.
,
De Nicola S.
,
Coppola G.
,
Finizio A.
,
Alfieri D.
,
Pierattini G.
(2004)
“Controlling image size as a function of distance and wavelength in Fresneltransform reconstruction of digital holograms”
Opt. Lett.
29
854 
856
Lee J.
,
Moon I.
,
Liu H.
,
Yi F.
(2011)
“3D holographic image recognition by using graphic processing unit”
J. Opt. Soc. Korea
15
264 
271
Tajahuerce E.
,
Matoba O.
,
Javidi B.
(2001)
“Shiftinvariant three dimensional object recognition by means of digital holography”
Appl. Opt.
40
3877 
3886
Martinez L.
,
Javidi B.
(2008)
“Synthetic aperture singleexposure onaxis digital holography”
Opt. Express
16
161 
169
Kim T.
,
Kim Y.
(2010)
“Extraction of a distance parameter in optical scanning holography using axis transformation”
J. Opt. Soc. Korea
14
104 
108
Abdelsalam D.
,
Baek B.
,
Kim D.
(2011)
“Precise test sieves calibration method based on offaxis digital holography”
J. Opt. Soc. Korea
15
146 
151
Javidi B.
,
Moon I.
,
Yeom S.
,
Carapezza E.
(2005)
“Threedimensional imaging and recognition of microorganism using singleexposure online (SEOL) digital holography”
Opt. Express
13
4492 
4506
Moon I.
,
Javidi B.
(2008)
“3D visualization and identification of biological microorganisms using partially temporal incoherent light inline computational holographic imaging”
IEEE Trans. Medical Imaging
27
1782 
1790
Moon I.
,
Daneshpanah M.
,
Javidi B.
,
Stern A.
(2009)
“Automated three dimensional identification and tracking of micro/ nano biological organisms by computational holographic microscopy”
Proc. IEEE
97
990 
1010
Moon I.
,
Javidi B.
,
Yi F.
,
Boss D.
,
Marquet P.
(2012)
“Automated statistical quantification of threedimensional morphology and mean corpuscular hemoglobin of multiple red blood cells”
Opt. Express
20
10295 
10309
Laurie J.
,
Wyncoll D.
,
Harrison C.
(2010)
“New versus old bloodthe debate continues”
Critical Care
14
130 
Koch C.
,
Li L.
,
Sessler D.
,
Figueroa P.
,
Hoeltge G.
,
Mihaljevic T.
,
Blackstone E.
(2008)
“Duration of redcell storage and complications after cardiac surgery”
New England Journal of Medicine
20
1229 
1239
Cuche E.
,
Marquet P.
,
Depeursinge C.
(1999)
“Simultaneous amplitude and quantitative phase contrast microscopy by numerical reconstruction of Fresnel offaxis holograms”
Appl. Opt.
38
6994 
7001
Colomb T.
,
Cuche E.
,
Charrière F
,
Kühn J.
,
Aspert N.
,
Marquet P.
,
Depeursinge C.
(2006)
“Automatic procedure for aberration compensation in digital holographic microscopy and application to specimen shape compensation”
Appl. Opt.
45
851 
863
Dubois F.
,
Joannes L.
,
Legros J.C.
(1999)
“Improved threedimensional imaging with digital holography microscope using a partial spatial coherent source”
Appl. Opt.
38
7085 
7094
Marquet P.
,
Rappaz B.
,
Cuche E.
,
Colomb T.
,
Emery Y.
,
Depeursinge C.
,
Magistretti, P.
(2005)
“Digital holography microscopy: a noninvasive quantitative contrast imaging technique allowing visualization of living cells”
Opt. Lett.
30
468 
470
Gonzalez R.
,
Woods R.
2002
Digital Imaging Processing
Prentice Hall
New York, USA
Mukhopadhyay N.
2000
Probability and Statistical Inference
Marcel Dekker
New York, USA
Rencher C.
1998
Multivariate Statistical Inference and Application
Wiley
New York, USA
Goodman J.
,
Lawrence R.
(1967)
“Digital image formation from electronically detected holograms”
Appl. Phys. Lett.
11
77 
79
Cuche E.
,
Marquet P.
,
Depeursinge C.
(2000)
“Spatial filtering for zeroorder and twinimage elimination in digital offaxis holography”
Appl. Opt.
39
4070 
4075
Liu R.
,
Dey D.
,
Boss D.
,
Marquet, P.
,
Javidi B.
(2011)
“Recognition and classification of red blood cells using digital holographic microscopy and data clustering with discriminant analysis”
J. Opt. Soc. Am. A
28
1204 
1204