Fast image analysis in polarization SHG
microscopy.
Ivan Amat-Roldan,1 Sotiris Psilodimitrakopoulos,
1 Pablo Loza-Alvarez,
1,* and
David Artigas1,2
1ICFO-Institut de Ciències Fotòniques, Mediterranean Technology Park, 08860 Castelldefels (Barcelona), Spain
2Department of signal theory and communications, Universitat Politècnica de Catalunya, 08034,Spain
Abstract: Pixel resolution polarization-sensitive second harmonic
generation (PSHG) imaging has been recently shown as a promising
imaging modality, by largely enhancing the capabilities of conventional
intensity-based SHG microscopy. PSHG is able to obtain structural
information from the elementary SHG active structures, which play an
important role in many biological processes. Although the technique is of
major interest, acquiring such information requires long offline processing,
even with current computers. In this paper, we present an approach based on
Fourier analysis of the anisotropy signature that allows processing the
PSHG images in less than a second in standard single core computers. This
represents a temporal improvement of several orders of magnitude
compared to conventional fitting algorithms. This opens up the possibility
for fast PSHG information with the subsequent benefit of potential use in
medical applications.
©2010 Optical Society of America
OCIS codes: (180.4315) Nonlinear microscopy; (100.2960) Image Analysis; (170.3880)
Medical and biological imaging; (180.6900) Three-dimensional microscopy; (180.0180)
Microscopy; (190.4160) Multiharmonic generation; (170.0170) Medical optics and
biotechnology.
References and links
1. P. J. Campagnola, M. D. Wei, A. Lewis, and L. M. Loew, “High-resolution nonlinear optical imaging of live
cells by second harmonic generation,” Biophys. J. 77(6), 3341–3349 (1999).
2. M. E. Llewellyn, R. P. J. Barretto, S. L. Delp, and M. J. Schnitzer, “Minimally invasive high-speed imaging of
sarcomere contractile dynamics in mice and humans,” Nature 454(7205), 784–788 (2008).
3. L. Fu, and M. Gu, “Polarization anisotropy in fiber-optic second harmonic generation microscopy,” Opt. Express
16(7), 5000–5006 (2008).
4. H. Bao, A. Boussioutas, R. Jeremy, S. Russell, and M. Gu, “Second harmonic generation imaging via nonlinear
endomicroscopy,” Opt. Express 18(2), 1255–1260 (2010).
5. E. U. Rafailov, M. A. Cataluna, and W. Sibbett, “Mode-locked quantum-dot lasers,” Nat. Photonics 1(7), 395–
401 (2007).
6. P. J. Campagnola, A. C. Millard, M. Terasaki, P. E. Hoppe, C. J. Malone, and W. A. Mohler, “Three-
dimensional high-resolution second-harmonic generation imaging of endogenous structural proteins in biological
tissues,” Biophys. J. 82(1), 493–508 (2002).
7. R. M. Williams, W. R. Zipfel, and W. W. Webb, “Interpreting second-harmonic generation images of collagen I
fibrils,” Biophys. J. 88(2), 1377–1386 (2005).
8. S.-W. Chu, S.-P. Tai, M.-C. Chan, C.-K. Sun, I.-C. Hsiao, C.-H. Lin, Y.-C. Chen, and B.-L. Lin, “Thickness
dependence of optical second harmonic generation in collagen fibrils,” Opt. Express 15(19), 12005–12010
(2007).
9. G. Recher, D. Rouède, P. Richard, A. Simon, J.-J. Bellanger, and F. Tiaho, “Three distinct sarcomeric patterns of
skeletal muscle revealed by SHG and TPEF microscopy,” Opt. Express 17(22), 19763–19777 (2009).
10. S. V. Plotnikov, A. M. Kenny, S. J. Walsh, B. Zubrowski, C. Joseph, V. L. Scranton, G. A. Kuchel, D. Dauser,
M. Xu, C. C. Pilbeam, D. J. Adams, R. P. Dougherty, P. J. Campagnola, and W. A. Mohler, “Measurement of
muscle disease by quantitative second-harmonic generation imaging,” J. Biomed. Opt. 13(4), 044018 (2008).
11. O. Nadiarnykh, S. Plotnikov, W. A. Mohler, I. Kalajzic, D. Redford-Badwal, and P. J. Campagnola, “Second
harmonic generation imaging microscopy studies of osteogenesis imperfecta,” J. Biomed. Opt. 12(5), 051805
(2007).
#128412 - $15.00 USD Received 12 May 2010; revised 8 Jul 2010; accepted 15 Jul 2010; published 29 Jul 2010(C) 2010 OSA 2 August 2010 / Vol. 18, No. 16 / OPTICS EXPRESS 17209
12. S. Plotnikov, V. Juneja, A. B. Isaacson, W. A. Mohler, and P. J. Campagnola, “Optical clearing for improved
contrast in second harmonic generation imaging of skeletal muscle,” Biophys. J. 90(1), 328–339 (2006).
13. P. Matteini, F. Ratto, F. Rossi, R. Cicchi, C. Stringari, D. Kapsokalyvas, F. S. Pavone, and R. Pini,
“Photothermally-induced disordered patterns of corneal collagen revealed by SHG imaging,” Opt. Express 17(6),
4868–4878 (2009).
14. R. Cicchi, D. Kapsokalyvas, V. De Giorgi, V. Maio, A. Van Wiechen, D. Massi, T. Lotti, and F. S. Pavone,
“Scoring of collagen organization in healthy and diseased human dermis by multiphoton microscopy,” J
Biophoton. 3(1-2), 34–43 (2010).
15. R. A. Rao, M. R. Mehta, and K. C. Toussaint, Jr., “Fourier transform-second-harmonic generation imaging of
biological tissues,” Opt. Express 17(17), 14534–14542 (2009).
16. R. A. R. Rao, M. R. Mehta, S. Leithem, and K. C. Toussaint, Jr., “Quantitative analysis of forward and backward
second-harmonic images of collagen fibers using Fourier transform second-harmonic-generation microscopy,”
Opt. Lett. 34(24), 3779–3781 (2009).
17. K. M. Reiser, C. Bratton, D. R. Yankelevich, A. Knoesen, I. Rocha-Mendoza, and J. Lotz, “Quantitative analysis
of structural disorder in intervertebral disks using second harmonic generation imaging: comparison with
morphometric analysis,” J. Biomed. Opt. 12(6), 064019 (2007).
18. P. Stoller, K. M. Reiser, P. M. Celliers, and A. M. Rubenchik, “Polarization-modulated second harmonic
generation in collagen,” Biophys. J. 82(6), 3330–3342 (2002).
19. S. Plotnikov, V. Juneja, A. B. Isaacson, W. A. Mohler, and P. J. Campagnola, “Optical clearing for improved
contrast in second harmonic generation imaging of skeletal muscle,” Biophys. J. 90(1), 328–339 (2006).
20. F. Tiaho, G. Recher, and D. Rouède, “Estimation of helical angles of myosin and collagen by second harmonic
generation imaging microscopy,” Opt. Express 15(19), 12286–12295 (2007).
21. S. Psilodimitrakopoulos, S. I. Santos, I. Amat-Roldan, A. K. Thayil, D. Artigas, and P. Loza-Alvarez, “In vivo,
pixel-resolution mapping of thick filaments’ orientation in nonfibrilar muscle using polarization-sensitive second
harmonic generation microscopy,” J. Biomed. Opt. 14(1), 014001 (2009),
http://spiedl.aip.org/getpdf/servlet/GetPDFServlet?filetype=pdf&id=JBOPFO000014000001014001000001&idt
ype=cvips&prog=normal.
22. C. Odin, T. Guilbert, A. Alkilani, O. P. Boryskina, V. Fleury, and Y. Le Grand, “Collagen and myosin
characterization by orientation field second harmonic microscopy,” Opt. Express 16(20), 16151–16165 (2008).
23. S. Psilodimitrakopoulos, D. Artigas, G. Soria, I. Amat-Roldan, A. M. Planas, and P. Loza-Alvarez, “Quantitative
discrimination between endogenous SHG sources in mammalian tissue, based on their polarization response,”
Opt. Express 17(12), 10168–10176 (2009).
24. W. L. Chen, T. H. Li, P. J. Su, C. K. Chou, P. T. Fwu, S. J. Lin, D. Kim, P. T. C. So, and C. Y. Dong, “Second
harmonic generation chi tensor microscopy for tissue imaging,” Appl. Phys. Lett. 94, 3 (2009).
25. S. Psilodimitrakopoulos, V. Petegnief, G. Soria, I. Amat-Roldan, D. Artigas, A. M. Planas, and P. Loza-Alvarez,
“Estimation of the effective orientation of the SHG source in primary cortical neurons,” Opt. Express 17(16),
14418–14425 (2009).
26. S. Psilodimitrakopoulos, I. Amat-Roldan, P. Loza-Alvarez, and D. Artigas, “Estimating the helical pitch angle of
amylopectin in starch using polarization second harmonic generation microscopy,” J. Opt. 12(8), 084007 (2010).
27. S. Psilodimitrakopoulos, I. Amat-Roldan, S. Santos, M. Mathew, A.K.N. Thayil, D. Zalvidea, D. Artigas, and P.
Loza-Alvarez, “Starch granules as a probe for the polarization at the sample plane of a high resolution
multiphoton microscope,” SPIE, 68600E (2008).
28. O. Nadiarnykh, and P. J. Campagnola, “Retention of polarization signatures in SHG microscopy of scattering
tissues through optical clearing,” Opt. Express 17(7), 5794–5806 (2009).
1. Introduction
Second harmonic generation (SHG) laser scanning microscopy is considered nowadays one of
the most promising minimally invasive, high-resolution optical techniques for clinical
applications [1]. Because of recent technological advantages in micro-endoscopes/fibers [2–4]
and laser sources [5], SHG imaging shows a great potential for clinical usage as an optical
biopsy tool [6].
Earlier, numerous studies on the SHG contrast have demonstrated that collagen, myosin,
and microtubules are effective SHG converters in tissues [1]. Consequently, biological
structures that are consisting of the above endogenous SHG sources can be imaged using
intensity-based SHG microscopy. Despite the fact that the SHG intensity is the only contrast
mechanism for generating the images, several morphological parameters can be obtained [7].
For example, by comparing the forward and epi-detected SHG signals, conclusions on the
dimensions of collagen fibrils can be obtained [8]. Other methodologies take advantage of the
characteristic striation pattern for quantitative interpretation and extraction of information in
muscles [9]. In such cases, the analysis of the sarcomere pattern was used for the study of rare
diseases, including muscular dystrophy [10] and osteogenesis imperfecta [11]. More recently,
#128412 - $15.00 USD Received 12 May 2010; revised 8 Jul 2010; accepted 15 Jul 2010; published 29 Jul 2010(C) 2010 OSA 2 August 2010 / Vol. 18, No. 16 / OPTICS EXPRESS 17210
the use of image processing based on spatial Fourier analysis has been used to infer properties
of tissue and molecules of the intensity-based SHG imaging [12]. Specifically, bidimensional
(2D) Fourier transform (FT) was used to spatially quantify the disorganization of collagen
fibres due to photo-thermal damage in porcine corneas [13]. It was found that the regularities
in fibres organization leads to an elliptical distribution in the bidimensional (2D-FT)
transformed space, whereas randomness leads to a more circular distribution [14]. Also very
recently, it was presented that additional information on collagen fibre orientation and
maximum spatial frequency can be obtained using 2D-FT in an SHG image [15]. Likewise, it
was shown that the 2D-FT can also be performed in the epi-detection [16].
In addition to the above approaches, the polarization dependency of the SHG signal
(PSHG) can also be used to generate contrast [17, 18]. In particular, due to the geometric
characteristics of local hyperpolarizability tensor β(2)
arrangement, by rotating the incoming
linear polarization (or the sample), the detected SHG signal intensity from every point
provides a characteristic modulation, often called the anisotropy or signature/fingerprint curve
[19, 20]. Such PSHG modulation can be exploited to obtain information unreachable by
intensity-based only SHG imaging [21]. For example it provides intrinsic tissue
characterization without any labeling and without using an analyzer in the detection [22–24].
In particular, in collagen, PSHG has allowed obtaining the orientation of the β(2)
dominant
axis with respect the long axis of the supramolecule, which is related to the helical pitch angle
of the polypeptide chain of the collagen triple-helix [19]. In muscle such orientation is
attributed to the α-helix of the myosin’s coiled coil (myosin tail) of thick filaments [19]. In
cultured cortical neurons this orientation coincides with the architecture of the tubulin
heterodimmers forming the axons’ microtubules [25] and in starch granules it indicates
amylopectin as the SHG active molecule [26]. To obtain such information an iteration code is
usually used [19–25]. This iteration code can be based in a fitting algorithm able to fit the
PSHG images to a biophysical model in a pixel by pixel fashion [21]. However, because of
the pixel resolution examination, a typical fitting procedure takes many hours to process one
image (~6 hrs for a 512x512 pixels image). This slow processing time drastically hampers the
use of the of PSHG methodology for clinical applications. Having a fast routine able to obtain
such parameters in a robust way might open up many new and exciting applications into life
sciences.
In this work, we present a polarization 1-D FT analysis of the anisotropy curve to retrieve
the biophysical parameters of the proposed model, referred to us as Fast Fourier Polarization
SHG (FF-PSHG) analysis, to obtain the same information as with the iteration fitting-based
procedure, but instead of hours, in a few hundreds of milliseconds using a regular processor.
Considering a PSHG image of a three dimensional data set, I(x,y,α), where x-y refer to the
spatial axis of the image and α refers to the polarization dependency of the SHG image
(anisotropy curve), this FF-PSHG analysis is performed only on the polarization axis, α, of
every pixel. This is in contrast to the spatial image processing methods using 2D-FT in the
(x,y) axis discussed above [15, 16].
To show the feasibility of this method, in section 2 we describe the theory and physical
implications of the FF-PSHG analysis. This is followed in section 3 by the experimental
demonstration of the method in biological samples, showing the capability to retrieve the
orientation of the supramolecule assembly and the hyperpolarizability tensor β(2)
dominant
axis orientation (normally related to the helical pitch angle). In this section, the capability of
the method to perform discrimination between different tissues is also demonstrated using two
different strategies. Finally, these results are followed by the conclusions.
2. Theory
The biophysical model used here (see refs [20,21,23–25]) refers to an SHG active
supramolecular assembly with cylindrical symmetry. The SHG signal dependency of such
structures on the input polarization of the fundamental beam can be written as:
#128412 - $15.00 USD Received 12 May 2010; revised 8 Jul 2010; accepted 15 Jul 2010; published 29 Jul 2010(C) 2010 OSA 2 August 2010 / Vol. 18, No. 16 / OPTICS EXPRESS 17211
2 2 2 2 2
( ) sin [2( )] [ sin ( ) cos ( )] ,SHGI C A Bφ φ α φ α φ α= − + − + − (1)
where φ and α are the orientation of the long axis of the cylinder, which we assume coincides
with the supramolecular assembly alignment, and the angle of the fundamental beam
polarization, respectively, defined with respect the lab x-axis [21]. A = I0 d31, B = I0 d33 and C
= I0 d15, where I0 is proportional to the intensity of the excitation fundamental field, and d31,
d33 and d31 are the non-zero elements of the nonlinear susceptibility tensor characterizing the
tissue under cylindrical symmetry assumption [21].
The free parameters A, B, C and φ are usually obtained by fitting the experimental SHG
images for every polarization α to Eq. (1). To do that, an iterative nonlinear algorithm is
usually utilized. However, this is a lengthy task as, depending on the size of the image, this
fitting procedure may take several hours. To speed up such process, Eq. (1) can be rewritten
in a more convenient form as a sum of cosine frequency components as follows:
0 2 4
( ) cos 2( ) cos 4( ),SHGI φ α α φ α α φ α= + − + − (2)
where a0 =C2/2 +3/8 (A
2 +B
2) +AB/4, a2 = B
2/2-A
2/2 and a4 = (A-B)
2/8 – C
2/2. Note that the
parameters φ, a0, a2 and a4 contain now the whole information relative to our biophysical
model (tensor elements). In what follows, we show that these components can be readily
obtained by our FF-PSHG analysis in an efficient manner.
As commented in the introduction, in this work we are going to perform the 1D-FT only
on the polarization axis, α as i(x,y,Ω)=FαI(x,y,α). By doing so, the Fourier transform of
Eq. (2) in a pixel, determined by (x,y), with a polarization sampling between 0° and 180°,
results in
0 2 4
( ) (0) exp( 2 ) (1 ) exp( 4 ) (2 ) . .,i i i c cα δ α φ δ α φ δ− −Ω = + Ω + Ω + (3)
where c.c. indicates complex conjugated. From Eq. (3), we can now directly retrieve the
different cosine components and therefore, extract the elements’ ratio of the second order
susceptibility tensor of our model, in a pixel by pixel fashion. Note that since in an
experiment the polarization intensity period is 180°, a sampling has therefore be considered in
the 0° to 180° range. Polarization sampling performed between 0° and 360° is in fact
reproducing the measurement twice (α is equivalent to α+180°). In this case the FF-PSHG
analysis can be used by transforming the Dirac delta into δ(2−Ω) and δ(4−Ω) respectively,
with the advantage that an immediate averaging of the two set of results (from 0° to 180° and
from 180° to 360°) is obtained. In the rest of the document, for simplicity we assume the
sampling is in the range from 0° to 180°. Note that the quadratic nature of PSHG response
[see Eq. (1)] generates a symmetric polarization response in the polarization intervals α ∈ [0,
π], therefore φ has the same periodicity, which for convenience we chose the range φ ∈ [- π/2,
π/2].
Before go further, it is worth to note that Eq. (1) possesses a mathematical intrinsic
ambiguity that affects any PSHG experiment. This ambiguity is apparent when the same
result is obtained by exchanging A and B and adding π/2 phase to the orientation φ [21]. From
a physical point of view, this ambiguity appears because the model is build in a manner that
assumes a minimum SHG signal when the incident polarization is perpendicular to the
cylinder’s long axis. Experimentally, this has been reported to occur in several biosamples
such as microtubulin of axons [25], collagen [18] or starch [26], and results in B/C >A/C ≈1.
However, muscle [20] shows the minimum SHG signal when the incident polarization is
parallel to the thick filaments orientation (assumed to posses the cylindrical symmetry), with
B/C <A/C ≈1. Since the ambiguity cannot be solved using mathematical criteria, a priori
knowledge on the different sample PSHG response was needed. When using the fitting
algorithm, the ambiguity is solved in every pixel by assigning the value closer to the unity to
#128412 - $15.00 USD Received 12 May 2010; revised 8 Jul 2010; accepted 15 Jul 2010; published 29 Jul 2010(C) 2010 OSA 2 August 2010 / Vol. 18, No. 16 / OPTICS EXPRESS 17212
A/C. Then if this value is associated to the sinus in Eq. (1) the orientation is directly the
retrieved angle φ. On the contrary, if A/C is associated to the cosine in Eq. (1), the actual
orientation isφ+π/2.
2.1. Determining the orientation of the supramolecular assembly φ
The above ambiguity also affects our FF-PSHG analysis, particularly in the orientation φ.
Direct observation of Eq. (2) shows that φ only affects the cosine argument. Therefore, when
performing Fourier transform to Eq. (3), it will appear as a phase in the first and second
(second and four) coefficients when performing the sampling between 0° and 180° (0° and
360°). Then, the extraction of the orientation φ consists in computing the complex argument
of the second coefficient as
2' arg[ exp( 2 )] / 2iφ α φ= (4)
In Eq. (4), the ambiguity is apparent in the fact that the obtained angle φ' and the
orientation φ can be different, since φ' also include information on the sign of a2. This is
because a2 = B2/2-A
2/2 can either be positive (|B|>|A|) or negative (|B|<|A|). This unknown
sign is transformed in the intrinsic ambiguity of π/2 in calculating the angle orientation φ. This
is totally equivalent to the ambiguity in Eq. (1) by exchanging A and B and adding π/2 to the
orientation φ [21]. Similarly, the condition used with iterative fitting algorithms, |A/C|<|B/C|,
results in a2>a4, characteristic of collagen and starch, while |A/C|>|B/C| results in a2<a4,
which is typical in myosin. Therefore, by comparing a2 and a4 it is possible to solve the
indetermination as follows:
2 2 4
2 2 4
' for
' 2 for
a a
a a
φφ
φ π
≥=
+ <
(5)
For other tissues it will be possible to design different strategies and define specific
criteria. Also note that since the extraction of φ.is based on the phase of the polarization-
spectral components, it is, in principle, independent of possible errors affecting the amplitudes
a0, a2 and a4, adding robustness to the method.
2.2. Extraction of the biophysical parameters
The rest of the parameters, A, B and C can be then extracted analytically by combining Eqs.
(1) and (2) as:
2
0 2 4
2
0 2 4
2
0 4
2/ 2 / 2)( ,
A
B
C A B
α α α
α α α
α α
= +
= ± +
= +− −
∓
(6)
Where the change of sign in Eq. (6) is related to the discussed ambiguity, and must be chosen
accordingly to the criteria used in selecting the cylinder’s orientation, φ. Once these three
parameters have been computed, the tensor element ratios can be calculated as d31/ d15=A/C
and d33/d15=B/C. This can be performed in an almost instantaneous way (considering the
current speed of modern computers) with no constrains.
Once this is done, it is possible to go a step forward and calculate the angle θe between the
hyperpolarizability tensor β(2)
dominant axis (the nonlinear SHG-dipole, normally related to
the helical pitch angle) and the long axis of the supramolecular assemble . To do that, a series
of restrictions, related with or in addition to the “single-axis molecule” approach, are normally
imposed: (1) there is only one major orientation φ in each pixel, (2) the long axis of the
supramolecular assemble is parallel to the imaging plane (2D), (3) both Kleinman and
cylindrical symmetries hold. All these conditions imply that A=C, or equivalently, d31/ d15=1.
#128412 - $15.00 USD Received 12 May 2010; revised 8 Jul 2010; accepted 15 Jul 2010; published 29 Jul 2010(C) 2010 OSA 2 August 2010 / Vol. 18, No. 16 / OPTICS EXPRESS 17213
However, any deviation from the above restriction, experimental errors and detection noise
level, resulted in a certain distribution around d31/ d15=1 [21, 23, 25, 26]. In this situation, the
algorithm can be forced to fulfill A=C by considering a0 =A/2 +3/8 (A2 +B
2) +AB/4, a2 =
B2/2-A
2/2 and a4 = (A + B)
2/8 – A/, and use only one of the two equations in (6) to determine
A and B. Then, the results of d33/ d15, or alternatively A and B can be used to obtain θe as [21–
23,25,26]:
2
cos / (2 ).e
B A Bθ = + (7)
2.3. Pixels with erroneous results
In previous works, we have shown the capability of the iterative algorithms to remove pixels
with erroneous results by filtering them out by setting a threshold on the fitting coefficient of
determination, r2 (usually keeping pixels presented r
2 > 90%) [21,23,25,26]. Here we propose
a different strategy that is based on the analysis of the spectral components for the PSHG
modulation response.
The source of noise in the PSHG images includes experimental errors like anisotropy of
the sample, depolarization introduced by the optical components, optical misalignment, non-
exact determination of the polarization, saturation and poor signal to noise ratio (SNR). This
means that the noise within a pixel can be considered being equally distributed among all the
spectral components obtained after performing the 1D-FF in the polarization axis. Among all
these components, only those with Ω ≤ 2 have biophysical meaning according to current
model [see. Eq. (3)].. Therefore, the origin of any signal appearing in polarization frequency
components with Ω > 2 can be associated to noise. As result, since FF-PSHG analysis only
uses the spectral components Ω ≤ 2, noise associated to the components with Ω > 2 is
intrinsically filtered in determining all the previous parameters.
In addition to this filtering, the noise at components Ω > 2 can be used to estimate the total
amount of error in the coefficients a0, a2 and a4. To do that, we assume that the error in the
frequency components at Ω = 0, 1 and 2 (related with a0, a2 and a4) is affected in a similar
way as those components with Ω > 2. Therefore the experimental error in determining a0, a2
and a4 in a pixel can be estimated comparing the spectral components as
0 2 4
( , ) [ ( ( , , ), 2)] / [ ( , , )]e x y mean i x y mean α α α= Ω Ω > (8)
For example, the signal detected in pixels in areas outside any SHG active tissue is noise
in nature and therefore results in a value of e ≈1. However, pixels in areas with a good signal
to noise ratio will result in e ≈0. Then, since e quantifies the error in a pixel, it can be used to
filter out pixels with e above a certain threshold value, eth, which are considered erroneous
[i.e., do not match Eqs. (1)–(3)]. Typical values for the threshold are in the range eth ≈0.02-
0.1.
3 Results-discussion
In this section, we show the capability of determining the orientation of the supramolecular
assembly φ, also referred as fiber orientation, discussing the ambiguity described in
subsection 2.1, and the determination of θe. This is followed by two methods to perform
discrimination among tissues.
3.1 Single SHG-active structure images
To show the ability of the algorithm to locally determine the orientation of the supramolecular
assembly we analyze a representative case: a granule of starch. A granule of starch has been
previously reported to possess a radial molecular orientation [26]. This sample is ideal to
show the performance of the method since it allows obtaining data within the whole
orientation range, from 0° to 180°. The multiphoton microscope used to acquire the PSHG
#128412 - $15.00 USD Received 12 May 2010; revised 8 Jul 2010; accepted 15 Jul 2010; published 29 Jul 2010(C) 2010 OSA 2 August 2010 / Vol. 18, No. 16 / OPTICS EXPRESS 17214
images has been thoroughly described in Refs [21,23,25,26]. The linear polarization at the
sample plane exhibited an extinction ratio of 25:1. By comparing the summed reconstruction
of linear polarization images with the image created using circular polarization, we found this
ratio adequate. Figure 1 shows the results, measuring the angle φ using both the FF-PSHG
analysis, which is obtained in 100 ms, and a fitting algorithm, which lasted ~6 hours with 400
iterations per pixel. The results are very similar, clearly retrieving the radial-like structure.
The small deviation of the radial symmetry in Fig. 1 might be attributed to the imperfections
on the starch granule. We can also observe that a smooth change from pixel to pixel is
obtained with FF-PSHG analysis, without the need of pixel averaging, as is the case in the
image obtained with the iterative algorithm. We attribute this smooth variation to the filtering
process intrinsic in the FF-PSHG analysis. In addition to this filtering procedure, pixels with e
> 0.05 have been removed from the image (black color). Notice that only points near the
external surface of the starch granule disappears, denoting the quality of the measurement. In
the case of the fitting algorithm, pixels with coefficient of determination r2<90%, has been
filtered out. Finally, when using a fitting algorithm, final results slight change depending on
the initial conditions and number of iterations. This is not the case in our FF-PSHG analysis,
since its analytical nature always provides the same results. This adds robustness and
consistency to the analysis.
Fig. 1. Calculation of fiber orientationφ in starch: (a) Mean SHG intensity of the 9 PSHG
images. Scale bar shows 10µm. (b) FF-PSHG analysis using 9 polarizations and (c) iterative
fitting algorithm using 8 polarizations.
We next analyze the capability of the method to clearly determine changes in the φ
orientation by using a more sophisticated sample. Figure 2(a) shows a detail on the orientation
for a collagen fiber shown later in Fig. 4. In this figure we have manually delineated the
orientation of the fiber (white line), computing the angle at every point of the line. This result,
shown in Fig. 2(b), is then compared with the angle retrieved at selected pixels of the curve
using our FF-PSHG analysis. A nice agreement is observed, showing consistency of the
method that is able to calculate angular deformations. This result entails us to use this method
to analyze complex fiber situations as shown in Fig. 2(c). Again, an image showing a smooth
variation of the fiber orientations is obtained.
#128412 - $15.00 USD Received 12 May 2010; revised 8 Jul 2010; accepted 15 Jul 2010; published 29 Jul 2010(C) 2010 OSA 2 August 2010 / Vol. 18, No. 16 / OPTICS EXPRESS 17215
Fig. 2. (a) detail of fiber of collagen and manual delineation of fiber trajectory in white and (b)
comparison of local PSHG fiber orientation (solid) and angle of the manual delineation
(dashed); and retrieved fiber orientation using the FF-PSHG analysis is shown in (c). Pixels
with e > 0.1 have been filtered out and are represented in black pseudocolor.
With the above results, the reliability of the method to determine any fiber orientation
change is clearly demonstrated. This allows analyze complicated situations and go a step
forward to obtain the helical pitch angle in a sample, with pixel resolution, and compare the
results obtained with the fitting algorithm and the FF-PSHG analysis. The results,
corresponding to bundle of collagen fibers oriented in different directions, are showed in Fig.
3 (the corresponding fiber orientation is shown in Fig. 2c). Figure 3(a) shows the
superposition of the SHG intensity images for all the polarizations. This figure show the
difficulties to obtain SHG signal in some areas, specifically in most of the points in the top
part of the collagen bundle that will result in a poor noise to signal ratio. As a consequence,
the analysis performed using the iterative algorithm, shown in Fig. 3(b), lacks important parts
of the image, which has been filtered out due to the low quality of fitting in the top part of the
image (points with coefficient of determination, r2<85% where removed). In spite of the
decrease of useful pixel, the fitting algorithm is able to correctly retrieve the helical pitch
angle, whose distribution is shown in Fig. 3(c), with the maximum frequency at θe = 44.4° and
a distribution width of ∆θe = 5.4° (the helical pitch angle obtained with X-ray diffraction
measurements is ~45°). On the contrary, the FF-PSHG analysis shown in figure Fig. 3(d) is
able to map θe in the entire sample, even for those areas with low SHG signal quality (notice
the top part of the bundle of collagen fibers). This is possible to the noise filtering intrinsic to
the method. The image shows smooth changes that give the impression of volume, which
correlates well with the contour of Fig. 3(a). This makes us suspect that the variation in θe can
be attributed to be mainly produced by out of plane fiber axis orientations. The θe frequency
distribution obtained with the FF-PSHG analysis is shown in Fig. 3(e), showing a
displacement of the maximum, at θe = 42.3° and a distribution width of ∆θe = 4.9°. This
displacement of the maximum is attributed to major number of pixels with θe ≈40°, appearing
in the top part of the bundle of collagen fibers, which are filtered out by the fitting algorithm
in Fig. 3(c).
Regarding the time required to compute the above Figs. 2-3 (500 x 500 pixels), the FF-
PSHG analysis lasted around 100 ms to compute the fiber orientation, while the calculus of θe
#128412 - $15.00 USD Received 12 May 2010; revised 8 Jul 2010; accepted 15 Jul 2010; published 29 Jul 2010(C) 2010 OSA 2 August 2010 / Vol. 18, No. 16 / OPTICS EXPRESS 17216
required less than 300 ms. Figures obtained with the fitting algorithm, where obtained with
400 iterations per pixel and lasted ~6 hours.
Fig. 3. Lyophilized Achilles’ tendon collagen. (a) Superposition of the SHG intensity images
for eight polarizations. Scale bar shows 10µm. (b) Image showing the helical pitch angle in
every pixel obtained using an iterative fitting algorithm and its frequency distribution in (c).
Similarly, (d) shows the helical pitch angle in every pixel and its frequency distribution in (e),
this time using the FF-PSHG analysis.
3.2 Multiple SHG-active structure images
PSHG offers the unique characteristic of identifying and discriminating different SHG active
molecules, with pixel resolution, in the same image [23]. In this section we show that our FF-
PSHG analysis can also be used with discrimination purposes by computing B/A parameter
and θe in every pixel [using Eq. (7)]. The results for unstained temporalis muscle from rat are
shown in Fig. 4(a), where it is possible to observe a clear discrimination between two tissues,
orange corresponding to muscle and blue to collagen. In this case, the time required to
compute Fig. 4(a) was less than 300 ms.
In addition to the discrimination method described above, the FF-PSHG analysis offers a
simple discrimination alternative based on directly mapping the cosine frequency components
a0, a2 and a4 into RGB images. Since the values of a0, a2 and a4 depend on the actual SHG
molecule, the weight for every RGB channel is different for different tissues. Therefore,
different tissues appear with different pseudocolor in the same image. The results are shown
in Fig. 4(b). We can observe that both tissues are clearly differentiated. In order to identify
what are the actual tissues displayed, the typical relation among values a0, a2 and a4 must be
characterized. In the case of Fig. 4(b), and by comparing with Fig. 4(a), in the RGB
representation yellow corresponds to collagen and purple to myosin. This provides a simple
method to discriminate among different tissues, getting an instantaneous perception of the
image, with the advantage that the time required to compute Fig. 4(b) was less than 100 ms, in
a single core computer, after acquiring the corresponding PSHG data.
Comparing Figs. 4(a) and 4(b), we see that both methods provide similar discrimination
capabilities, the main difference being the image appearance. This differences in the images
appears because the cosine frequency components a0, a2 and a4 contains information on the
ratio among a0, a2 and a4, which is always the same for a tissue (providing the discrimination
capability), and on the intensity, providing smoother changes in the image. This results in an
#128412 - $15.00 USD Received 12 May 2010; revised 8 Jul 2010; accepted 15 Jul 2010; published 29 Jul 2010(C) 2010 OSA 2 August 2010 / Vol. 18, No. 16 / OPTICS EXPRESS 17217
apparent better quality of Fig. 4(b) when compared with Fig. 4(a). However, although the
RGB representation provides the fastest method to discriminate between different SHG
sources, the standard method has the advantage that the tissue identification is based on a
known intrinsic characteristic of the SHG active molecule, as is the helical pitch angle,
providing a clear criterion in case of ambiguous situations. This is clearly apparent when
plotting the frequency distribution for the helical pitch angle as showed in Fig. 4(c). The two
well separated, not overlapping peaks centered at 43° and 64° for collagen and muscle,
respectively, show the ability of the method to unambiguously distinguish between tissues in
the same image.
Fig. 4. (a) Map of the computed helical pitch angle using FF-PSHG (b) Pseudocolored cosine
coefficients a0, a2 and a4 of Eq. (2), plotted into RGB images, red channel is a0, green channel
is a2 and blue channel is a4. Scale bar shows 10µm, and (c) the helical pitch angle frequency
distribution of Fig. 3(a), exhibiting a centre of the distribution at 43° for collagen and 64° for
myosin.
4. Conclusions
Polarization-sensitive Second Harmonic Generation is a promising imaging modality that
enables statistically studying the orientational distribution of the β(2)
dominant axis (related to
the helical pitch angle) of a number of molecules which play a role in many biological
processes: collagen, microtubulin and myosin and additional structures, like starch, which has
been also used to probe polarization state in a microscope [27]. Especially when optical
clearing is used, this information can be acquired several hundreds of microns deep in tissues
[28]. Therefore, many applications can be enhanced by the development of new and faster
algorithms than the current ones, which are executed “offline”, requiring from minutes to
hours to process an image of 500 by 500 pixels, even with multi-core computers.
In this paper, we have presented for the first time an approach that allows processing in
few milliseconds an image based on 1D Fourier analysis of the PSHG modulation response
obtaining a temporal improvement of near five orders of magnitude. This opens the possibility
for PSHG imaging to penetrate new fields in medicine at video rates, acting for example as an
instantaneous diagnostic supporting method in surgery. The results are in total agreement of
those obtained by conventional fitting algorithms, where the intrinsic noise filtering results in
a smother response and a better contrast, while its analytical nature provides robustness and
consistency to the analysis. In conclusion, we have presented a sub-second method to process
PSHG images to extract full biophysical meaning and straight visualization methods that can
be useful for many fields in microscopy and biomedicine that possess additional advantages
that do not possess its prior competitors.
Acknowledgments
This work is supported by the Generalitat de Catalunya grant 2009-SGR-159 and by the
Spanish government grant TEC2009-09698 Authors also acknowledge the Laserlab-Europe
#128412 - $15.00 USD Received 12 May 2010; revised 8 Jul 2010; accepted 15 Jul 2010; published 29 Jul 2010(C) 2010 OSA 2 August 2010 / Vol. 18, No. 16 / OPTICS EXPRESS 17218
Cont (JRA4: Optobio 212025) and the Photonics4Life networks of excellence. This research
has been partially supported by Fundació Cellex Barcelona.
#128412 - $15.00 USD Received 12 May 2010; revised 8 Jul 2010; accepted 15 Jul 2010; published 29 Jul 2010(C) 2010 OSA 2 August 2010 / Vol. 18, No. 16 / OPTICS EXPRESS 17219