Fluorescence imaging is a widely used in vivo optical imaging technique for preclinical studies, but strong tissue autofluorescence and external excitation light make it suffer from a low signal-to-noise ratio (SNR). Recently, a new optical imaging method using persistent luminescence has become of interest due to its advantage of circumvention of autofluorescence and bleed-through of excitation light during signal acquisition. In this work, we proposed a tomographic imaging method based on persistent luminescence named persistent luminescence tomography (PLT), which can obtain three dimensional distributions of persistent luminescence probes deep inside small animals. Persistent luminescence signals can last several hours after excitation, which makes it possible for PLT to collect signals without interference by autofluorescence and bleed-through of excitation light, and then to reconstruct tomographic images of high quality. Phantom and mouse experiments are implemented to verify the feasiblity of PLT.
© 2017 Optical Society of America
As a kind of macroscopic optical imaging, fluorescence imaging (FI) is widely used in biomedical research and drug development owing to its advantages of non-invasive, non-radioactive, high-throughput, easy-to-operate, and cost-effective detection [1–3]. Fluorescence acquisition needs external excitation both in a reflection or transillumination mode, which brings strong tissue autofluorescence and external excitation light into desired fluorescent signals, and suffers from a low signal-to-noise ratio (SNR) .
Unlike the short lifetime of fluorescence, persistent luminescence (PL) relates to an optical phenomenon in which a kind of particular material can glow for several hours after excitation with ultraviolet or visible light . These PL materials can store excitation energy and then slowly release it by a photonic emission [6, 7]. Using PL materials for imaging, persistent luminescence imaging (PLI) can separate excitation and emission processes, and completely circumvents the strong tissue autofluorescence and external excitation light in signal acquisition. As a result, PLI has the advantage of high SNR and sensitivity for in vivo imaging in small animal applications . Due to this excellent characteristic of PLI, PL materials in nanoscale or so-called PL nanoparticles (PLNPs) have also been functionalized to target specific cells and introduced into sensitive biological imaging of tumor-bearing mouse [9–11]. PLI has also been reported in cell tracking, tumor targeting, and vascularization imaging while exhibiting exceptional sensitivity and penetration that far exceed those afforded by FI . Recent developments have also led to the description of PL based Förster resonance energy transfer for imaging cancer cells .
Although PLI has a broad application prospect in biological research, this 2-D photographic imaging is incapable of providing depth information of the PL source inside biological tissues due to strong scattering characteristics. So, we propose tomographic PL imaging named persistent luminescence tomography (PLT) in this work. Combined with a diffuse optical model and image reconstruction algorithm, PLT has the ability to reveal 3-D distributions of the PL source deep inside small animal, which makes it possible to be used for detecting in situ tumors. Furthermore, PLT can provide higher quality tomographic images due to no interference of strong tissue autofluorescence and external excitation light compared with fluorescence tomography (FT), a type of tomographic imaging for FI. We investigated its feasibility by using physical phantom and mouse models based on homemade optical and computed tomography (CT) imaging systems.
2.1 Forward model
Although PLT contains excitation and emission prosesses, there is no excitation light during detection of PL signals, which is similar to bioluminescence tomography (BLT), another optical tomography employing luciferase enzymes for in vivo detection of tagged cells in living animals . Furthermore, the used area shape excitation source in this work can sufficiently excite the whole imaged object, and the flux density of the excitation light inside the imaged object is approximately uniform. Thus, only the emission process is considered for PLT, and the forward model of PLT can be described by a diffusion equation in view of accuracy and computation cost 
where Φm(r) is the flux density of the PL signal, and X denotes the unknown PL yield. µam is the absorption coefficient, is the diffusion coefficient, and is the reduced scattering coefficient. and represent the region of the object and its boundary respectively.is a boundary constant representing the refractive index mismatched between tissues and the surrounding medium .
Here a tangential planar approach named Kirchhoff approximation can be used to find an analytical formulation for Green’s functions of the diffusion equation . After discretizing the medium into three dimensional grids, Eqs. (1) can be transformed into the following matrix-form equations:
2.2 Reconstruction method
Reconstruction of PLT is to solve Eq. (2), which is an ill-posed problem, due to insufficient measurement compared with the whole solving domain. Considering the sparse characteristic of unknown source X, orthogonal matching pursuits (OMP) can be used to solve Eq. (2), which is a widely used iterative method for sparse recovery based on the Compressed Sensing theory . OMP uses the observation vector as a good local approximation to the unknown fluorescence yield X, and the largest coordinate of the observation vector in magnitude is chosen as a nonzero coordinate of the unknown source X in each iteration . Based on this similar idea, if the sparsity level of X is N, we can make use of N biggest coordinates of the observation vector , rather than the largest one, as N nonzero coordinates of unknown source X. Then, we can use the corresponding N columns of to form a new weight matrix mapping the nonzero solving domain into measurement Φ.
Finally, traditional Tikhonov regularization can be used to solve this sub-equation Eq. (3) with a reduced solving domain. The details of this reconstruction method can be described as the following algorithm:
- Input: weight matrix , measurement Φ, sparse level N and regularization parameter .
- Output: unknown fluorescence yield X.
- 1. Compute observation vector to local approximate X.
- 2. Choose a set formed by the coordinates of the biggest observations from .
- 3. Build the reduced solving domain problem using the columns of and rows of Φ defined in.
- 4. Solve the reduced solving domain problem using Tikhonov regularization .
- 5. Build the final unknown fluorescence yield X by with coordinates in and 0 with coordinates out of .
2.3 The imaging flow
The process of PLT is summarized as Fig. 1. PLNPs are injected into imaged mouse via caudal vein. After arrival at targeted lesions of PLNPs, the imaged mouse is excitated by ultraviolet or visible light for seveal minutes. Then, the PL image can be captured from the top view by an EMCCD camera (iXon DU-897, Andor, U.K.) coupled with a Nikkor 50-mm f/2.8D lens (Nikon, Melville, NY). To acquire the tomographic optical image, the geometry of the imaged mouse is necessary for forward model computation. So, CT scans of the imaged mouse are performed using a homemade CT imaging system, which includes an X-ray tube (Apogee, Oxford Instruments, UK) and a CMOS X-ray detector (C7921, Hamamatsu, Japan). The anatomical structure of the imaged mouse is obtained by using a cone-beam Feldkamp-Davis-Kress (FDK) reconstruction algorithm on the graphics processing unit using an acceleration scheme . PL image is mapped onto the 3-D surface of the imaged mouse, which is extracted from the segmented CT image. Finally, optical forward model computation and image reconstruction are carried out to obtain 3-D distributions of PLNPs inside the imaged mouse.
3.1 Comparison between persistent luminescence and fluorescence
To illustrate the advantages that PLI is free from autofluorescence and bleed-through of excitation, a glass cup of 40 mm in height and 20 mm in diameter filled with 1% intralipid solution was employed as a turbid medium. Absorption and reduced scattering coefficients for the medium are 0.002 mm−1 of 1 mm−1, respectively, which mimick biological tissues with high absorption and low scattering of light. For fluorescent imaging, a small glass tube filled with 100 uL Cy5.5 NIR dye with a concentration of 0.01 uM was used as the fluorescent target. For PL imaging, a small glass tube filled with 10 mg Y2O2S:Eu3+,Ti2+, and Mg4+, a red long-lasting PLNPs, was used as the PL target which has characteristic absorption peak in ultraviolet spectrum . Fluorescent and PL targets were fastened to the bottom of the glass cup, and experiments with different depths were implemented by adding different volumes of intralipid solution
To capture fluorescent images, an area light source was provided by a halogen lamp coupled with an optical fiber and a 640 ± 10 nm bandpass filter to excite the fluorescent target. A 690 ± 20 nm bandpass filter was used to select the fluorescent signals of Cy5.5. The emission spectruman of the PLNPs ranges from 500nm to 720nm with emission peak at 630nm when excited by a 430nm ultraviolet lamp as shown in Fig. 2. Obviously, there is no need to use any filter to capture PLI signals. But to make a fair comparision with FI, the same 690 ± 20 nm bandpass filter was also used to capture PL images after excitation by the ultraviolet lamp about 10 seconds.
Figure 3 shows PL and the FI images for different depths. The first row is the setups for different depths, and the corresponding PLI and FI images are listed below. Generally, the intensities of PLI and FI images are decreased with depth of intralipid. By adjusting the intensity of exciting light and exposure time for FI acquisition, the maximum of original FI image is almost the same as that of PLI image for each depth. PLI image contains only PLI signals, but original FI image contains not only fluorescent signals but also noises attributed to autofluorescence and bleed-through of excitation light. So the noise of FI image was also acquired by removing fluorescent target with the same acquisition parameters. The intensity for noise of FI image increases slightly with depth of intralipid, that is because a deep one has a high intralipid level and receives strong excitation light when a divergent excitation light source is fixed on the top. The subtracted FI image was obtained by subtracting noise of FI image form original FI image, which contains only FI signals. There is no observable visual difference between original FI image and subtracted FI image, when FI signals are strong enough for 5mm depth case. But original FI images emerge significant visual differences with subtracted FI images when depths are deeper than 5 mm. Besides that, PLI images present similar distributions of optical signals with those of subtracted FI images for all depth cases.
Profiles along vertical lines defined on the left of Fig. 4 for PLI, original and subtracted FI images of Fig. 3 are shown on the right of Fig. 4. For PLI, resolution declines only caused by increase of depth, but autofluorescence and bleed-through of excitation light also degrades FI resolution besides depth. After subtracting of noise of FI image, subtracted FI image has the similar profile with PLI image for each depth case.
3.2 Phantom experiments
In the phantom experiment, two polyoxymethylene made semi-cylinders with 20 mm in diameter and 40 mm in height were employed as physical phantoms to mimic biological tissues. Their absorption and scattering coefficients were 0.002 mm−1 and 1 mm−1 respectively, which was similar to those of biological tissues. Each phantom had a hole drilled along the axial direction with the depth of 3mm. A small glass tube filled with about 5 mg Y2O2S:Eu3+,Ti2+, Mg4+ was used as the PL target, which was inserted into the hole of the phantom to mimic the PLT phantom experiment. After excitation by an ultraviolet lamp for about 10 seconds, we turned off the lamp and captured the PL image with an exposure time of 1s for the CCD camera coupled with a 690 ± 20 nm bandpass filter. Subsequently, micro-CT scans of the phantom were performed (voltage of 50 kVp and current of 1.0 mA for the X-ray source, 360 views with 1° interval). Faltered back-projection (FBP) method was used to reconstruct micro-CT data of the phantom, which were used to supply the geometry of the phantom for forward model calculation of PLT. For comparison, a small glass tube filled with 50 uL Cy5.5 dye with a concentration of 0.01 uM was used as the fluorescent target and inserted into the hole of the phantom to mimic the FT phantom experiment. Original FI image was collected when the phantom was excited by an area light source with a spectrum of 640nm. Then the fluorescent target was removed and the noise of FI image was acquired using the same acquisition parameter. Finnally, a CT image for the FT experiment was obtained with the same setup as PLT experiment.
Without disturbs of autofluorescence, the PL image (Fig. 5(a)) contains only PL signals undergoing scattering and absorbing of the diffused phantom. The original FI image has a slight different distributions of optical signals with those of the PL image, although they contain the same emission spectrum and maximum optical intensity. After removing the noise of FI image from the original FI image, the obtained subtracted FI image has more similar distributions of optical signals with those of PL image, which demenstracts that noises caused by autofluorescence and bleed-through of excitation light can distort true distributions of optical signals.
Figure 6 shows the reconstructed PLT and FT images, which includes the transverse, sagittal and coronal 2D sections and 3D volume renderings at different viewing angles for PLT and FT reconstructions. The reconstructed PL and fluorescent targets are visually in good agreement with CT images, which can provide their accurate positions. The size of the reconstructed fluorescent target is slightly larger than that of the true one in the CT image. Dice coefficient (DIC) and position (center of gravity) were computed to quantify the reconstructed results as shown in Table 1, and DIC is defined as
where RS represents the regions of the true target segmented from the CT image, and RR represents the regions of the reconstructed target segmented from a reconstructed image. A large DIC means a high agreement of the reconstructed and true targets. For PLT, DIC is 0.66 and position error is 0.39 mm, which is better than FT with DIC 0.52 and position error 0.78 mm.
3.3 Mouse experiments
For mouse experiment, the animal was cared for in accordance with a protocol approved by the Fourth Military Medical University Animal Care and Use Committee. Two nude mice were used as imaged animals. Glass tube containing FI or PL target as the same with the phantom experiment was implanted into rectum by surgical procedure under general anesthesia by intraperitoneal injection of 10% sodium pentobarbital. To avoid traumas of nude mice, the glass tube lubricated with man-rated lubricating oil was gently implanted into rectum with a small entering distance through natural openings of the nude mice. Optical images were captured from a supine position using the similar acquisition processes but with different exposure time for appropriate optical intensities. And then, CT images for PLT and FT were also obtained with the same setup as phantom experiment.
Comparing with phantom experiment, the imaged object used in mouse experiment was changed into a mouse but with the similar target implant approach. So it was just an emulation to realistic biological application using realistic geometry and optical properties of a mouse in a real biological implementation. Besides that, noise of FI image for phantom experiment mainly contains bleed-through of excitation light, while mouse experiment mainly contains autofluorescence. PLI can completely circumvent the strong autofluorescence as shown in Fig. 7(a), but autofluorescence is distributed all over the body of the nude mouse in the noise of FI image (Fig. 7(d)), wihich has a significant interference to original FI image (Fig. 7(b)).
Results of the mouse experiment are summarized in Fig. 8. Both reconstructed targets in 2D sections and 3D volume renderings for PLT and FT reconstructions are well conformed with those in CT images. But the size of the reconstructed FI target is slightly larger than that of the true one in the CT image. Quantitative analysis is listed in Table 2, and DIC of 0.66 and position error of 0.88 mm for PLT.
In this work, a tomographic imaging based on PL phenomenon termed as PLT was proposed. Although PLI has been used for in vivo applications in previous research [5, 6], it only focused on 2D imaging without depth information about PL probes. This work is the first time we know the feasibility of PLT to resolve the distributions of PL probes deep in turbid media and biological tissues.
After simplification, the forward model and reconstruction method for PLT are the same as BLT. But it is important to note that PLT has several advantages in biological applications. First, there are many usable PL probes for PLT to study different biological processes, but only the luciferase gene report system is usable for BLT. Second, BLT needs a complex transgene technique to construct animal models with cells marked with the luciferase gene. While PLT is a much easier tool, which just needs suitable PL probes injected into the tail vein like FI. Third, PLT is more flexible than BLT. The signals of BLT last just several minutes, which need to finish imaging in a short time window before the bioluminescence signals disappear. But signals of PLT can be produced using external excitation at any time before PL probes been metabolized, which may undergo several hours determined by the characteristics of the special PL probes.
Without interfering from tissue autofluorescence and bleed-through of excitation light, in-vitro results demonstrated that PLI had higher resolution than that of FI in signal acquisition, which implies PLT can provide a high sensitive and deep imaging in practical applications. But from the reconstructed results of the phantom and mouse experiments, it seems a slight improvement in PLT. That is because the PL and FI targets are artificial, which have high concentrations. In practical applications, PL and FI targets are formed by convergence in the lesion areas based on the specificities of themselves. Generally, convergence rates for PL and FI probes are not high enough to provide excellent SNR. So FI signals may be flooded in tissue autofluorescence and external excitation light, but PL signals do not have this problem.
Finally, it is believed that PLT will provide more potential in preclinical applications due to many PL probes produced by researchers. To improve the image quality of PLT, multi-projection acquisition is useful such as FMT.
This work was supported by the Program of National Natural Science Foundation of China under Grant Nos. 81227901, 61405149, 81230033, 61471279, 81530058, the Program of the National key Research and Development Program of China under Grant No. 2016YFC1300302, the Beijing Municipal Natural Science Foundation under Grant No. 7142012, the Natural Science Basic Research Plan in Shaanxi Province of China under Grant No. 2015JZ019, and the Fundamental Research Funds for the Central Universities NSIZ021402.
References and links
3. V. Ntziachristos, J. Ripoll, L. V. Wang, and R. Weissleder, “Looking and listening to light: the evolution of whole-body photonic imaging,” Nat. Biotechnol. 23(3), 313–320 (2005). [CrossRef] [PubMed]
5. T. Matsuzawa, Y. Aoki, N. Takeuchi, and Y. Murayama, “A new long phosphorescent phosphor with high brightness, SrAl2O4:Eu2+,Dy3+,” J. Electrochem. Soc. 143(8), 2670–2673 (1996). [CrossRef]
6. T. Aitasalo, P. Deren, J. Hölsä, H. Jungner, J. C. Krupa, M. Lastusaari, J. Legendziewicz, J. Niittykoski, and W. Strek, “Persistent luminescence phenomena in materials doped with rare earth ions,” J. Solid State Chem. 171(1–2), 114–122 (2003). [CrossRef]
8. Q. le Masne de Chermont, C. Chanéac, J. Seguin, F. Pellé, S. Maîtrejean, J. P. Jolivet, D. Gourier, M. Bessodes, and D. Scherman, “Nanoprobes with near-infrared persistent luminescence for in vivo imaging,” Proc. Natl. Acad. Sci. U.S.A. 104(22), 9266–9271 (2007). [CrossRef] [PubMed]
9. T. Maldiney, C. Richard, J. Seguin, N. Wattier, M. Bessodes, and D. Scherman, “Effect of core diameter, surface coating, and PEG chain length on the biodistribution of persistent luminescence nanoparticles in mice,” ACS Nano 5(2), 854–862 (2011). [CrossRef] [PubMed]
10. A. Abdukayum, J. T. Chen, Q. Zhao, and X. P. Yan, “Functional near infrared-emitting Cr3+/Pr3+ co-doped zinc gallogermanate persistent luminescent nanoparticles with superlong afterglow for in vivo targeted bioimaging,” J. Am. Chem. Soc. 135(38), 14125–14133 (2013). [CrossRef] [PubMed]
11. T. Maldiney, M. U. Kaikkonen, J. Seguin, Q. le Masne de Chermont, M. Bessodes, K. J. Airenne, S. Ylä-Herttuala, D. Scherman, and C. Richard, “In vitro targeting of avidin-expressing glioma cells with biotinylated persistent luminescence nanoparticles,” Bioconjug. Chem. 23(3), 472–478 (2012). [CrossRef] [PubMed]
12. T. Maldiney, A. Bessière, J. Seguin, E. Teston, S. K. Sharma, B. Viana, A. J. Bos, P. Dorenbos, M. Bessodes, D. Gourier, D. Scherman, and C. Richard, “The in vivo activation of persistent nanophosphors for optical imaging of vascularization, tumours and grafted cells,” Nat. Mater. 13(4), 418–426 (2014). [CrossRef] [PubMed]
13. B. Y. Wu, H. F. Wang, J. T. Chen, and X. P. Yan, “Fluorescence resonance energy transfer inhibition assay for α-fetoprotein excreted during cancer cell growth using functionalized persistent luminescence nanoparticles,” J. Am. Chem. Soc. 133(4), 686–688 (2011). [CrossRef] [PubMed]
14. W. Cong, G. Wang, D. Kumar, Y. Liu, M. Jiang, L. Wang, E. Hoffman, G. McLennan, P. McCray, J. Zabner, and A. Cong, “Practical reconstruction method for bioluminescence tomography,” Opt. Express 13(18), 6756–6771 (2005). [CrossRef] [PubMed]
16. A. R. Arridge, “Optical tomography in medical imaging,” Inverse Probl. 15(2), 41–93 (1999). [CrossRef]
17. J. Ripoll, V. Ntziachristos, R. Carminati, and M. Nieto-Vesperinas, “Kirchhoff approximation for diffusive waves,” Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 64(5), 051917 (2001). [CrossRef] [PubMed]
18. J. A. Tropp and A. C. Gilbert, “Signal recovery from random measurements via orthogonal matching pursuit,” IEEE Trans. Inf. Theory 53(12), 4655–4666 (2007). [CrossRef]
19. V. Temlyakov, “Nonlinear methods of approximation,” Found. Comput. Math. 3(1), 33–107 (2003). [CrossRef]
20. G. Yan, J. Tian, S. Zhu, Y. Dai, and C. Qin, “Fast cone-beam CT image reconstruction using GPU hardware,” J. XRay Sci. Technol. 16(4), 225–234 (2008).
21. H. Chen, Y. C. Zhang, C. Chen, Z. L. Wang, and N. Yao, “The Research of Solvothermal Method Preparation of Y2O2S:Eu3+0.06, Mg2+0.04,Ti4+0.01 Water-Soluble Red Afterglow Materials,” Applied Mechanics and Materials 513–517, 138–142 (2014). [CrossRef]