Raman microscopy is a powerful method combining non-invasiveness with no special sample preparation. Because of this remarkable simplicity, it has been widely exploited in many fields, ranging from life and materials sciences to engineering. Notoriously, due to the required imaging speeds for bio-imaging, it has remained a challenge how to use this technique for dynamic and large-scale imaging. Recently, a supervised compressive Raman framework has been put forward, allowing for fast imaging, therefore alleviating the issue of speed. Yet, due to the need for strong a priori information of the species forming the hyperspectrum, it has remained elusive how to apply this supervised method for microspectroscopy of (dynamic) biological tissues. Combining an original spectral under-sampling measurement technique with a matrix completion framework for reconstruction, we demonstrate fast and inexpensive label-free molecular imaging of biological specimens (brain tissues and single cells). Using the matrix completion outcome with the supervised method allows for large compressions () and bio-imaging speeds surpassing current technology in spontaneous Raman microspectroscopy. Therefore, our results open interesting perspectives for clinical and cell biology applications using the much faster compressive Raman framework.
© 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
Raman imaging is a simple label-free approach that exploits the intrinsic vibrational spectra of species as their fingerprint. It has been widely applied in various biological specimens , ranging from chemically selective imaging of cells [2–5] to spectroscopic detection of pathologies [6,7], bacteria [8,9], and algae , to cite a few examples. However, the spontaneous Raman process is a weak effect, therefore demanding costly and sensitive multi-pixel cameras with dispersive spectrometers. Such cameras limit dynamic applications where fast imaging is required, due to slow readout speed, and associated electronic noise. Apart from this technological bottleneck, the huge data sizes of hyperspectroscopy are an issue for real-life applications, due to data storage and display of large specimens’ hyperspectra (3D object as presented in Fig. 1(A)).
Recently, compressive Raman has been suggested to overcome data size and imaging speed limitations [11–16]. Compressive Raman is based on concepts of the emerging field of compressive sensing, which exploits new sampling paradigms based on experimental undersampling followed by computational reconstruction. In general, two strategies exist in compressive Raman: supervised [12,16–18] and unsupervised compression [13,15,19,20]. Both concepts are based on the fact that the hyperspectrum typically contains a small number of distinguishable chemical signatures, i.e., it is spectrally sparse and also extremely “chemically sparse” (Fig. 1(B)). Mathematically, chemically sparse is equivalent to saying that is a low-rank matrix; that is, it is generated by a finite number of eigenspectra. Hence, from a few data samples, the complete hyperspectrum can be reconstructed without loss of fidelity [12,17]. On one hand, the unsupervised approach is the most appealing, as it requires low a priori information for computational reconstruction. Current implementations of unsupervised compressive methods are based on a wide field [19,20], which precludes imaging opaque biological samples or confocal geometries . However, these methods are based on cameras that limit faster acquisition speed and have additive noise at high speed [16,21]. In addition, they use algorithms with unrealistic computational reconstruction times of the hyperspectrum , in effect, precluding large-scale spectroscopic imaging. On the other hand, various demonstrations of the supervised approach have shown a high level of data compression with imaging speeds much faster than allowed by conventional camera-based technologies. The supervised method exploits the eigenspectra of (Fig. 1(B), rightmost spectra), as a priori information, to develop optimized spectral filters for fast, accurate, and precise chemical abundances determination [17,21]. Nevertheless, this supervised method fails in chemically changing environments, as the “eigenspectra library” may evolve.
We present a new methodology based on the low-rank character of to allow for fast chemical imaging of biological specimens. The method is based on a fast random undersampling scheme, which is a prerequisite for using the framework of matrix completion  (Fig. 1(C)). The key computational concept is based on the factorization , where is a diagonal matrix related to the rank of , and the eigenvectors and represent the “eigenimages” and “eigenspectra,” respectively (Fig. 1(B)). Since is typically low rank in Raman bio-imaging , a completely sampled means that highly redundant information is acquired; in other words, each spatial point is a simple linear combination of the few eigenspectra. Therefore, one can undersample and use established and efficient algorithms of low-rank matrix completion to “fill-in” the missing samples. A powerful advantage of the computational framework for chemical analysis is that it outputs a reduced dimension representation, which is ultimately desired for real-life applications (for instance, this could avoid a specialist in vibrational spectroscopy for interpretation).
A. Sample Preparation
Polymer beads were prepared by drop-casting colloidal suspensions, with low polydispersity (Polysciences Inc.), on a coverslip sealed with water. Mouse cerebellum brain slices (thickness 500 μm) were fixed in an agarose solution with phosphate-buffered saline (PBS) buffer and sodium azide. Cheek cells were extracted from a male donor, and further dispersed with physiological saline solution to reduce the local concentration of debris before imaging. No fixation method was used for the cell imaging.
B. Optical Setup
A thorough description of the Raman microscope and its high-throughput compressive spectrometer can be found elsewhere [24,25]. Briefly, a 532-nm-wavelength excitation laser (Oxxius LCX-532) is steered into an inverted microscope (Nikon Eclipse Ti-U) equipped with a high-NA oil-immersion objective (Nikon 60X/1.4NA). Samples are scanned with a nm-resolution piezoelectric translation stage (Physik Instrumente P-545.3R7). The inelastically scattered light is guided with a multimode fiber into the home-built compressive spectrometer , based on a traditional Czerny–Turner design, however, exchanging the usual high-sensitivity cameras for a programmable spectral filter using a digital micromirror device (DMD) (V-7001, ViALUX, 0.7” XGA resolution) that selects the various wavelengths to be detected by a photon-counter module (SPCM-AQRH-44, Excelitas Technologies). The master clock is provided by the scanner stage, which triggers the exposition of the spectral masks (DMD) and detector acquisition. Total excitation power measured before objective is 140 mW. Previous studies have shown that such power levels are safe under non-resonant conditions  and should be even less harmful upon fast pixel dwell times as used here.
C. Computational Techniques
Different algorithms were used for the analysis. They are based on either soft-threshold singular value decomposition (SVD)  or non-negative matrix factorization . In general, we chose reconstructions with ranks between 2 and 5, as previous results have suggested . In practice, we noticed that higher rank solutions only added noise to the reconstructed hyperspectrum. For the brain tissues results, the output of the matrix completion was passed to a standard SVD algorithm to generate the eigenimages, and in turn used for generating the spectra for input of the supervised approach [16,17]. For the spectral sampling domain, we have used two spectral basis sets for the spectral domain: canonical (polymer beads) and Hadamard bases (biological specimens).
We first describe the experimental methodology for the fast random sampling scheme (Fig. 1(D)). Basically, it consists of a standard confocal Raman microscope, which allows for opaque samples observations, coupled to a recently developed high-throughput programmable spectrometer , thus enabling high sensitivity bio-imaging. Briefly, the costly cameras of a conventional spectrometer are replaced by a DMD that can select wavelength bins to be detected with a highly sensitive single-pixel detector. As the focus moves, with a step size smaller than the point-spread function (PSF) of the microscope, we concomitantly sample spectral bases at random. Using this sampling strategy, we effectively sample several spectral components (alternatively, other bases can be used, e.g., Hadamard basis) for every spatial pixel in the hyperspectral image. We then obtain a (raw) matrix that is transformed in the uncompleted , i.e., in spatial coordinates (, ) versus spectral coordinates () (see the missing boxes in Fig. 1(C)). By using this random hyperspectrum fast sampling methodology, can be readily processed using off-the-shelf matrix completion algorithms [23,28]. The algorithm then outputs a SVD of , which we use for post-processing in a conventional manner used in Raman imaging. Alternatively, we also use non-negative matrix completion algorithms  based on the factorization , where and are non-negative matrices, motivated by the fact that Raman spectra are necessarily non-negative. Hence, this constraint may help to get a better reconstruction.
This methodology allows for fast imaging, with quasi-instantaneous reconstruction speeds in low signal-level scenarios, typical of biological specimens. We first benchmark the approach with standard polymer beads (see Methods). Figure 2(A) shows a representative undersampled hyperspectrum () together with its recovered completed hyperspectrum (). The averaged spectra (Fig. 2(B)) of the merged images (Fig. 2(C)) show the characteristic peaks of polystyrene (green), and the background (red) as a superposition of the water and glass coverslip spectra. The combined pseudo-color image (Fig. 2(C)) reveals that both species are anti-correlated as expected (the beads are larger than the microscope PSF). Figure 2(C) also shows the effect of compression ratio. High-level compression is obtained at the expense of increased noise in the reconstruction [13,15,29,30]. Careful analysis of the reconstruction fidelity reveals that the loss of fidelity upon compression is different for spatial and spectral domains, and also chemical species (Fig. 2(D), a complete analysis of the effects of compression on the spatial and spectral domains is presented in Supplement 1). Nevertheless, high chemically selective images are achieved at moderate compressions ( compression with fidelity). Finally, the key advantage of the matrix completion algorithm is its reconstruction time, which eventually allows for dynamic specimens imaging. Using standard laptop computers, we reconstruct the complete hyperspectrum at a rate of 8 ms/pixel (spatial).
The confocal geometry used here allows high-sensitivity imaging of biological specimens with -sectioning. We imaged cheek cells as a demonstration for cell sensitivity microspectroscopy (Fig. 3(A)). For that, we spectrally scanned only the C-H stretch region, as previous studies have shown this spectral region to be sensitive for cell compartments analysis [2,31]. The images generated from integrated C-H stretch peaks reveal multiple morphological features, such as the nucleus, membranes, and small organelles. Closer inspection of average spectra of selected locations (Fig. 3(A), lower panel) shows the expected trend in the C-H stretch intensity ratios , which has been previously shown to report on the protein (high ) and lipid (low ) content. We observe that the organelle could be potentially assigned to lipid droplets and, interestingly, the cell membrane contains an intermediate ratio, suggesting a mixture of proteins and lipid membranes. We also imaged opaque brain slices, recovering the expected sample morphology of tubular structures surrounded by continuous regions (Fig. 3(B)). The eigenimages from a SVD analysis (Fig. 3(B), lower left panels) show that the most significant species are separated in tubular and non-tubular morphologies. Remarkably, the averaged spectra (Fig. 3(B), top right panel) based on these images reveal that these two structures have high lipid content (red, low ) and high protein content (blue, high ), in agreement with the chemical morphology of brain tissues: tubular structures are myelins made of lipids surrounding the axons rich in protein. However, care has to be taken for a quantitative analysis of these images, as they are estimated eigenvectors (which can easily represent a combination of various chemical species), and do not correspond to 100% lipid or protein content.
We presented an unsupervised compressive Raman approach that enabled compressed imaging of biological specimens. The success relies on the combination of a new scanning methodology with the matrix completion framework. We note that documented scanning methods of supervised approaches [16,18,21] were based on a single spectral realization per image. Hence, they are not compatible with the matrix completion framework. Since the method presented here is based on a confocal geometry, it could in principle be used for imaging opaque samples at depth. It does not require any a priori knowledge of the hyperspectrum, apart from being low rank which, in fact, is fulfilled even under a chemically complex scenario of biological specimens. Such low-rank assumption seems to be the rule in Raman imaging of biological specimens as documented elsewhere [3,5,26,32]. Furthermore, the computational framework provides fast reconstruction, suitable for imaging dynamic and large-scale specimens, two aspects that are often faced when imaging biological tissues. Such high-speed reconstruction could not be achieved in previous algorithms of unsupervised compressive spectroscopy, as they require storage and multiplication of full-rank matrices (i.e., leading to slow reconstruction and large memory consumption) . Finally, similar to previous compressive sensing algorithms, we observed that high compression lowers reconstruction fidelity [13,15,29,30]; however, a high chemically contrasted image could still be obtained. This compromise of compression and reconstruction fidelity is well known in the compressive sensing framework, with a similar scaling in the matrix completion framework. We note that very high compression () could still be achieved with these algorithms not adapted for hyperspectroscopy; however, it became increasingly difficult to discern the two chemical species’ spectra (see spectra presented in Supplement 1). Note that spatial averaging allows to obtain cleaner spectra that can be used with the supervised approaches (see below).
Future modifications of the spatial scanning methodology can provide considerably higher speeds surpassing current technologies used in standard Raman microscopes. In the current implementation, the effective pixel dwell time was limited by the scanning stage, rather than photon budget. Therefore, higher frame rates can be achieved by exploiting galvanometric scanners. For example, based on commercially available DMDs with 100 kHz refresh rates and 20 spectral basis sampled, one can achieve 200 μs effective pixel dwell times. While further work is necessary to assess the limits of the detection of our implementation, we can provide estimates based on the photon budgets detected here (, Fig. 3(C)). To reach 2 MHz count rates without increasing power levels, one may simply lower the spatial resolution two-fold to increase signal levels by four-fold , a resolution that still permits chemically selective imaging. Such high spectral sampling speed would allow to image diverse cell dynamics (apoptosis , bio-mineralization , and mitosis ) that requires moderate field of view (). Nevertheless, a limitation of the current method is the need to scan several spectral bases, in particular, for samples that yield a low photon budget. In this case, it may be faster to use the costly CCD-based systems, as the whole spectrum is acquired in “one shot”.
We anticipate that combining the unsupervised approach (matrix completion) with the supervised one can lead to high-speed dynamic imaging. Despite the fact that supervised methods have shown 10 μs pixel dwell times , the major drawback is to learn the eigenspectra. To demonstrate the considerable speed-up of such a blend, we present a proof-of-principle experiment in Fig. 3(C). Based on the eigenspectra learned from one tissue (Fig. 3(B)), we could image a second tissue at faster speeds (Fig. 3(C)) exploiting optimized spectral filters (supervised compressive Raman [12,16]). Note that this demonstration was limited by the highest speed we could achieve with our scanner, i.e., higher speeds can be obtained using galvanometric scanners. Summarizing, this combination of methods resulted in 64 times data compression, and imaging speeds surpassing current camera-based technology .
In conclusion, we have presented a new methodology for enabling compressive Raman bio-imaging. Apart from the inherent data size compression, the method is fast, with reconstruction time negligible compared to image acquisition time, and inexpensive compared with alternative methods, with potential for much faster imaging speeds. We further showed that it is compatible with opaque samples imaging. Therefore, the methodology presented here paves the way for fast clinical imaging using the inexpensive spontaneous Raman effect.
Agence Nationale de la Recherche (ANR) (ANR-10-IDEX-0001-02 PSL*, ANR-10-LABX-0010); H2020 European Research Council (ERC) (724473); Universitat Jaume I (UJI) (PREDOC/2013/32); Generalitat Valenciana (PROMETEO/2016/079); Ministerio de Economía y Competitividad (MINECO), Gobierno de España (FIS2016-75618-R).
We thank Marie-Staphane Aigrot for kindly providing the brain slice samples. H. B. A. was supported by ANR. S. G. is a member of the Institut Universitaire de France.
See Supplement 1 for supporting content.
1. D. W. Shipp, F. Sinjab, and I. Notingher, “Raman spectroscopy: techniques and applications in the life sciences,” Adv. Opt. Photon. 9, 315–428 (2017). [CrossRef]
2. C. Matthäus, T. Chernenko, J. A. Newmark, C. M. Warner, and M. Diem, “Label-free detection of mitochondrial distribution in cells by nonresonant Raman microspectroscopy,” Biophys. J. 93, 668–673 (2007). [CrossRef]
3. K. Klein, A. M. Gigler, T. Aschenbrenner, R. Monetti, W. Bunk, F. Jamitzky, G. Morfill, R. W. Stark, and J. Schlegel, “Label-free live-cell imaging with confocal Raman microscopy,” Biophys. J. 102, 360–368 (2012). [CrossRef]
4. M. Okada, N. I. Smith, A. F. Palonpon, H. Endo, S. Kawata, M. Sodeoka, and K. Fujita, “Label-free Raman observation of cytochrome c dynamics during apoptosis,” Proc. Natl. Acad. Sci. USA 109, 28–32 (2012). [CrossRef]
5. W. Müller, M. Kielhorn, M. Schmitt, J. Popp, and R. Heintzmann, “Light sheet Raman micro-spectroscopy,” Optica 3, 452–457 (2016). [CrossRef]
6. S. Harmsen, R. Huang, M. A. Wall, H. Karabeber, J. M. Samii, M. Spaliviero, J. R. White, S. Monette, R. O’Connor, K. L. Pitter, S. A. Sastra, M. Saborowski, E. C. Holland, S. Singer, K. P. Olive, S. W. Lowe, R. G. Blasberg, and M. F. Kircher, “Surface-enhanced resonance Raman scattering nanostars for high-precision cancer imaging,” Sci. Transl. Med. 7, 271ra7 (2015). [CrossRef]
7. M. Jermyn, K. Mok, J. Mercier, J. Desroches, J. Pichette, K. Saint-Arnaud, L. Bernstein, M.-C. Guiot, K. Petrecca, and F. Leblond, “Intraoperative brain cancer detection with Raman spectroscopy in humans,” Sci. Transl. Med. 7, 274ra19 (2015). [CrossRef]
8. G. Bodelón, V. Montes-García, V. López-Puente, E. H. Hill, C. Hamon, M. N. Sanz-Ortiz, S. Rodal-Cedeira, C. Costas, S. Celiksoy, I. Pérez-Juste, L. Scarabelli, A. La Porta, J. Pérez-Juste, I. Pastoriza-Santos, and L. M. Liz-Marzán, “Detection and imaging of quorum sensing in pseudomonas aeruginosa biofilm communities by surface-enhanced resonance Raman scattering,” Nat. Mater. 15, 1203–1211 (2016). [CrossRef]
9. B. Lorenz, C. Wichmann, S. Stöckel, P. Rösch, and J. Popp, “Cultivation-free Raman spectroscopic investigations of bacteria,” Trends Microbiol. 25, 413–424 (2017). [CrossRef]
10. Q. Wu, W. Nelson, P. Hargraves, J. Zhang, C. Brown, and J. Seelenbinder, “Differentiation of algae clones on the basis of resonance Raman spectra excited by visible light,” Anal. Chem. 70, 1782–1787 (1998). [CrossRef]
11. B. M. Davis, A. J. Hemphill, D. Cebeci Maltas, M. A. Zipper, P. Wang, and D. Ben-Amotz, “Multivariate hyperspectral Raman imaging using compressive detection,” Anal. Chem. 83, 5086–5092 (2011). [CrossRef]
12. D. S. Wilcox, G. T. Buzzard, B. J. Lucier, P. Wang, and D. Ben-Amotz, “Photon level chemical classification using digital compressive detection,” Anal. Chim. Acta 755, 17–27 (2012). [CrossRef]
13. D. F. Galvis-Carreño, Y. H. Mejía-Melgarejo, and H. Arguello-Fuentes, “Efficient reconstruction of Raman spectroscopy imaging based on compressive sensing,” Dyna 81, 116–124 (2014). [CrossRef]
14. D. Wei, S. Chen, Y. H. Ong, C. Perlaki, and Q. Liu, “Fast wide-field Raman spectroscopic imaging based on simultaneous multi-channel image acquisition and wiener estimation,” Opt. Lett. 41, 2783–2786 (2016). [CrossRef]
15. N. Pavillon and N. Smith, “Compressed sensing laser scanning microscopy,” Opt. Express 24, 30038–30052 (2016). [CrossRef]
16. C. Scotté, H. B. de Aguiar, D. Marguet, E. Green, P. Bouzy, S. Vergnole, P. Winlove, N. Stone, and H. Rigneault, “Assessment of compressive Raman versus hyperspectral Raman for microcalcification chemical imaging,” Anal. Chem. 90, 7197–7203 (2018). [CrossRef]
17. P. Réfrégier, C. Scotté, H. B. de Aguiar, H. Rigneault, and F. Galland, “Precision of proportion estimation with binary compressed Raman spectrum,” J. Opt. Soc. Am. A 35, 125–134 (2018). [CrossRef]
18. T. C. Corcoran, “Compressive detection of highly overlapped spectra using Walsh-Hadamard-based filter functions,” Appl. Spectrosc. 72, 392–403 (2018). [CrossRef]
19. J. V. Thompson, J. N. Bixler, B. H. Hokr, G. D. Noojin, M. O. Scully, and V. V. Yakovlev, “Single-shot chemical detection and identification with compressed hyperspectral Raman imaging,” Opt. Lett. 42, 2169–2172 (2017). [CrossRef]
20. D. N. Wadduwage, V. R. Singh, H. Choi, Z. Yaqoob, H. Heemskerk, P. Matsudaira, and P. T. So, “Near-common-path interferometer for imaging Fourier-transform spectroscopy in wide-field microscopy,” Optica 4, 546–556 (2017). [CrossRef]
21. D. S. Wilcox, G. T. Buzzard, B. J. Lucier, O. G. Rehrauer, P. Wang, and D. Ben-Amotz, “Digital compressive chemical quantitation and hyperspectral imaging,” Analyst 138, 4982–4990 (2013). [CrossRef]
22. Y. August, C. Vachman, Y. Rivenson, and A. Stern, “Compressive hyperspectral imaging by random separable projections in both the spatial and the spectral domains,” Appl. Opt. 52, D46–D54 (2013). [CrossRef]
23. E. J. Candès and B. Recht, “Exact matrix completion via convex optimization,” Found. Comput. Math. 9, 717–772 (2009). [CrossRef]
24. S. H. Donaldson Jr. and H. B. de Aguiar, “Molecular imaging of cholesterol and lipid distributions in model membranes,” J. Phys. Chem. Lett. 9, 1528–1533 (2018). [CrossRef]
25. B. Sturm, F. Soldevila, E. Tajahuerce, S. Gigan, H. Rigneault, and H. B. de Aguiar, “High-sensitivity high-speed compressive spectrometer for Raman imaging,” arXiv:1811.06954 (2018).
26. N. Uzunbajakava, A. Lenferink, Y. Kraan, E. Volokhina, G. Vrensen, J. Greve, and C. Otto, “Nonresonant confocal Raman imaging of DNA and protein distribution in apoptotic cells,” Biophys. J. 84, 3968–3981 (2003). [CrossRef]
27. J.-F. Cai, E. J. Candès, and Z. Shen, “A singular value thresholding algorithm for matrix completion,” SIAM J. Optim. 20, 1956–1982 (2010). [CrossRef]
28. Y. Xu, W. Yin, Z. Wen, and Y. Zhang, “An alternating direction algorithm for matrix completion with nonnegative factors,” Front. Math. China 7, 365–384 (2012). [CrossRef]
29. P. Berto, C. Scotté, F. Galland, H. Rigneault, and H. B. de Aguiar, “Programmable single-pixel-based broadband stimulated Raman scattering,” Opt. Lett. 42, 1696–1699 (2017). [CrossRef]
30. D. J. Starling and J. Ranalli, “Compressive sensing for spatial and spectral flame diagnostics,” Sci. Rep. 8, 2556 (2018). [CrossRef]
31. C. Krafft, T. Knetschke, R. H. Funk, and R. Salzer, “Identification of organelles and vesicles in single cells by Raman microspectroscopic mapping,” Vib. Spectrosc. 38, 85–93 (2005). [CrossRef]
32. B. Prats-Mateu, M. Felhofer, A. de Juan, and N. Gierlinger, “Multivariate unmixing approaches on Raman images of plant cell walls: new insights or overinterpretation of results?” Plant Methods 14, 52 (2018). [CrossRef]
33. A. Hashimoto, Y. Yamaguchi, L.-D. Chiu, C. Morimoto, K. Fujita, M. Takedachi, S. Kawata, S. Murakami, and E. Tamiya, “Time-lapse Raman imaging of osteoblast differentiation,” Sci. Rep. 5, 12529 (2015). [CrossRef]
34. F.-K. Lu, S. Basu, V. Igras, M. P. Hoang, M. Ji, D. Fu, G. R. Holtom, V. A. Neel, C. W. Freudiger, D. E. Fisher, and X. S. Xie, “Label-free DNA imaging in vivo with stimulated Raman scattering microscopy,” Proc. Natl. Acad. Sci. USA 112, 11624–11629 (2015). [CrossRef]
35. O. G. Rehrauer, V. C. Dinh, B. R. Mankani, G. T. Buzzard, B. J. Lucier, and D. Ben-Amotz, “Binary complementary filters for compressive Raman spectroscopy,” Appl. Spectrosc. 72, 69–78 (2018). [CrossRef]