High-sensitivity high-speed compressive ... - ACS Publications

Apr 25, 2019 - High-sensitivity high-speed compressive spectrometer for Raman ... bio-imaging exploiting compressive Raman: (i) the idle solvent acts ...
0 downloads 0 Views 8MB Size
Subscriber access provided by UNIV OF SOUTHERN INDIANA

Article

High-sensitivity high-speed compressive spectrometer for Raman imaging Benneth Sturm, Fernando Soldevila, Enrique Tajahuerce, Sylvain Gigan, Herve Rigneault, and Hilton B. de Aguiar ACS Photonics, Just Accepted Manuscript • DOI: 10.1021/acsphotonics.8b01643 • Publication Date (Web): 25 Apr 2019 Downloaded from http://pubs.acs.org on April 28, 2019

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Photonics

High-sensitivity high-speed compressive spectrometer for Raman imaging Benneth Sturm,† Fernando Soldevila,‡ Enrique Tajahuerce,‡ Sylvain Gigan,¶ Herve Rigneault,§ and Hilton B. de Aguiar∗,† †Département de Physique, Ecole Normale Supérieure / PSL Research University, CNRS, 24 rue Lhomond, 75005 Paris, France ‡GROC-UJI, Institute of New Imaging Technologies (INIT), Universitat Jaume I, E12071 Castellò, Spain ¶Laboratoire Kastler Brossel, CNRS UMR 8552, Ecole Normale Supérieure, PSL Research University Sorbonne Universités & Université Pierre et Marie Curie Paris 06, F-75005, Paris, France §Aix Marseille Univ, CNRS, Centrale Marseille, Institut Fresnel, F-13013 Marseille, France E-mail: [email protected]

Abstract Compressive Raman is a recent framework that allows for large data compression of microspectroscopy during its measurement. Because of its inherent multiplexing architecture, it has shown imaging speeds considerably higher than conventional Raman microspectroscopy. Nevertheless, the low signal-to-noise ratio (SNR) of Raman scattering still poses challenges for high-sensitivity bio-imaging exploiting compressive Raman: (i) the idle solvent acts as a background noise upon imaging small biological organelles, (ii) current compressive spectrometers are lossy precluding high-sensitivity imaging.

1

ACS Paragon Plus Environment

ACS Photonics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

We present inexpensive high-throughput spectrometer layouts for high-sensitivity compressive hyperspectroscopy. We exploit various modalities of compressive Raman allowing for up to 80X reduction of data storage and 2X microspectroscopy speed up at a 230-nm spatial resolution. Such achievements allowed us to chemically image subdiffraction-limited biological specimens (lipid bilayers) in few seconds, paving the way for applications of compressive Raman framework to cell biology and nanophotonics applications.

Keywords Raman scattering; microspectroscopy; compressive sensing; label-free; vesicles; membrane. Raman microspectroscopy allows probing a chemically heterogeneous system with high spatial resolution and superb molecular selectivity. In particular, Raman-based imaging holds promising applications for biological specimens studies, 1 such as for cancer 2,3 and bacteria detection, 4,5 molecular biology imaging 6–9 , to cite a few examples. Nevertheless, hyperspectroscopy (i.e. spatially-resolved spectroscopy), in general, generates large data sizes which will eventually become a hurdle when bringing Raman spectroscopic imaging as a valuable tool for clinical tests, for example, for issues related to data storage and access. As Raman microspectroscopy acquisition speeds continuously advance, 10 it will become increasingly important to achieve data compression, in particular for dynamic and/or large-scale imaging at high-resolution. Single-pixel-based compressive sensing technology has attracted a large interest recently. 11 The idea behind compressive sensing is to obtain the same information as in multi-pixel technologies approaches, however through a designed optical experiment that will compress the data at the acquisition stage, by undersampling. In a post-processing step, efficient computational methods retrieve the desired information accurately. Interestingly, singlepixel approaches are particularly powerful where multi-pixel technology is still lacking or challenging, such as in terahertz and pump-probe spectroscopy methods. 12,13 Single-pixel 2

ACS Paragon Plus Environment

Page 2 of 21

Page 3 of 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Photonics

compressive sensing is a very appealing approach for Raman imaging as it is inexpensive and may have very fast computational reconstruction times, allowing for real-time applications. Compressive microspectroscopy is based on spectral, or spatio-spectral multiplexing, 14 which is achieved by fast programmable spatial light modulators, such as digital micromirror devices (DMD), operating at 10’s kHz. While compressive microspectroscopy 14–26 is a promising technology, it still lacks sensitivity for bio-imaging when compared to sensitive CCD cameras. High-sensitivity imaging is of particular relevance in the sub-micron spatial scale, as it is the size range of many organelle in a cell, 7–9,27 including lipid droplets, cell membranes, and membrane-less organelles. The challenges for high-sensitivity compressive microspectroscopy are two-fold. First, apart from Raman scattering being a weak effect, the Raman signal (IR ) scales unfavourably with spatial resolution: IR ∝ ω02 for objects that fill the focus (ω0 being the focus size of the excitation laser). Therefore, upon low-resolution imaging, the overall signal increases but the sensitivity decreases for detecting objects that have a size smaller than the focus, as the surrounding solvent adds as a strong background. Second, and related to compressive spectrometer layouts, there is a strong size mismatch (100X) between sensitive detector sizes and the size of the spectrum spread in the Fourier plane of a spectrometer (i.e. on the DMD). This mismatch is the main cause of sensitivity loss in compressive spectrometers. Here, we exploit novel high-throughput and inexpensive spectrometer designs for highresolution and high-sensitivity compressive Raman imaging of biological specimens. This is achieved by dispersion-controlled layouts to focus a large spectral bandwidth on a small single-pixel detector. After discussion of the methods we propose, we showcase two compressive Raman modalities exploiting different levels of a priori knowledge from the system being imaged. We then demonstrate chemical imaging of sub-wavelength Raman sources using the compressive spectrometer.

3

ACS Paragon Plus Environment

ACS Photonics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Methods Confocal microscope The compressive Raman microscope is based on a standard layout of an epi-detection Raman microscope (Fig. 1). The excitation laser (λ = 532 nm, 150 mW, Oxxius LCX-532) is cleaned up by a bandpass filter (FLH05532-4, Thorlabs), and expanded to about 7.2 mm, before entering an inverted microscope (Nikon Eclipse Ti-U). An additional half-waveplate and quarter-waveplate (not shown) pre-compensate and generate circularly polarized light 28 after the beam being reflected off from the dichroic filter (F38-532-T1, AHF) into a high-numerical aperture (NA), oil-immersion objective (Nikon 60X, 1.4 NA). Specimens are mounted on a translation-stage (Physik Instrumente P-545.3R7) for raster scanning. The inelastically backscattered Raman light (epi configuration) is focussed by the microscope tube lens on a multimode fiber (M50L02S-A, Thorlabs, 50 µm core-size), with a notch filter blocking residual pump light (NF533-17, Thorlabs), thus guiding the signal to the home-built compressive spectrometer. Excitation power measured before objective was 140 mW, power levels at which we did not observe any damage to the samples upon the fast exposure times used here.

Single-pixel compressive spectrometer The home-built spectrometer is based on a traditional Czerny-Turner design. The signal coming out of the fiber is collimated by an achromatic lens L1 (f=19 mm, AC127-019A-ML, Thorlabs), which then impinges on a high-efficiency transmission grating (FSTGVIS1379-911, Ibsen Photonics, 1379 l/mm) providing about 85 % diffraction efficiency around λ=600 nm. After the grating, the beam is spectrally dispersed and different colors are focused with an off-axis parabolic mirror PM (f=101.6 mm, 90◦ off-axis angle, MPD149P01, Thorlabs). The DMD module (V-7001, ViALUX, 0.7" XGA resolution, with V4395 controller board) selects the various wavelengths to be detected. Typically, we bin about 4

ACS Paragon Plus Environment

Page 4 of 21

Page 5 of 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Photonics

8 pixels, thus having a macropixel size that is slightly below the spectral resolution of the spectrometer. The signal from the DMD is steered into an avalanche photodiode operating in photon counting mode (SPCM-AQRH-44, Excelitas Technologies). This module combines high quantum efficiency from Si avalanche photodiodes (70 %), large dynamic range (low dark and high maximum count rates), with an active area diameter of 180 µm. The small area, while keeping high quantum efficiency, is of particular importance in the design because focusing the complete spectrum to the size of the detector-chip efficiently is not straightforward. We note other single-photon counters exist with mm’s active area size, however, with lower quantum efficiency at the spectral range detected here.

Sample preparation Polystyrene particles (1 µm and 0.088 µm-diameter, Polysciences Inc.) were drop cast on a coverslip, then filled with water. For the three-species system, we have added a drop of a silicone oil-in-water emulsion on top of the polytyrene beads film. Multilamellar vesicles (MLV) were prepared by drop-casting a film of lipid solution (10 mg/ml dipalmitoylphosphatidylcholine in chloroform) on a vial, then adding water and heat up at 38 ◦ C for about 30 min. 29 After spontaneous formation of the vesicles, an aliquot of the dispersion is then sandwiched between two coverslips.

Results Dispersion management for single-pixel compressed spectroscopy The main challenge of a single-pixel compressive spectrometer is to reach high-sensitivity with a large spectral bandwidth. In a conventional spectrometer, a sensitive CCD camera is used to detect the dispersed input light, which size is typically few mm’s. In contrast, the compressive spectrometer consists of a DMD, in place of the CCD, which steers wavelength 5

ACS Paragon Plus Environment

ACS Photonics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Translation stage

Page 6 of 21

L2

Grating 2 L3

Objective Beam Bandpass expander filter Dichroic Filter

Fiber

Detection: grating-based

Laser Tube lens

Inverted Microscope

SPAD

L2

DMD

Notch filter

L3 SPAD

Detection: lens-based

L1 Grating 1

PM

Compressive spectrometer

Figure 1: Simplified layout of the compressive Raman microscope. A CW laser source excited Raman scattering process, which is detected in epi configuration. The sample is raster-scanned with a high-resolution piezo-scanner. The Raman inelastically scattered light is steered into a home-built spectrometer designed for compressive single-pixel experiments. The dashed boxes depicts two strategies (grating or lens telescope) for managing the dispersion introduced by the first grating, and allow to focus large bandwidths in the high sensitivity single-pixel detector. DMD=digital micromirror device, PM=off-axis parabolic mirror, L=lens, SPAD: single-photon avalanche diode. bins onto a small single-pixel detector (100’s µm) to achieve high-sensitivity. Therefore, one has to considerably demagnify the beam reflected off by the DMD. Hence, we studied two strategies that could keep the interesting aspects (costs and throughput) of single-pixel detection. We investigated two approaches that are presented in dashed box insets of Fig. 1. The first approach is based solely on lenses (lens-based), motivated by the fact that transmission of lenses are highest compared to other optical elements. To aid the discussion, we recall the Smith-Helmholtz invariant (yσ = y 0 σ 0 ) relating planes y (grating 1), ray angles σ (arising from grating dispersion before PM), to y 0 (L3) and σ 0 (arising from focusing with L2 with a focal length f2 ) of an optical system. As the grating plane y is imaged onto plane y 0 by a PM and L2, we have: y0 f2 σ = = = m, 0 σ y f1

(1)

where f1 is the focal length of lens L1. Eq. 1 shows that dispersions and magnifications can be interchanged. For instance, σ 0 can be reduced by simply increasing the magnification m.

6

ACS Paragon Plus Environment

Page 7 of 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Photonics

However, the maximum magnification is limited by the clear aperture of L3. The second approach is based on a second grating to cancel the dispersion of the first one (Fig. 1, gratingbased), as typically done in pulse shapers of ultrafast spectroscopy, if they have the same groove density. Note that if the second grating is of different dispersion, one can adjust f2 to match the dispersion given by σ 0 . We have compared the two methods. For the lens-based approach, we used an effective 4X demagnification (in the figure it is presented as a single lens L2). For the gratingbased method, we used a second grating with the same dispersion as the first grating. We have inspected the spectrum of the lens-based and grating-based approaches, achieving 17 and 77 nm-bandwidth, respectively. While both methods coupled large bandwidths (over 100’s cm−1 ), the lens-based had too much losses originating from clipping in the edge of lenses. Therefore, we used the grating-based design with which we could achieve a spectral resolution of 29 cm−1 . Nevertheless, we believe the lens-based throughput could overcome the grating-based if a low-dispersion grating is used.

Compressive Raman without a priori knowledge We benchmarked the spectrometer with two different types of compressive Raman microspectroscopy schemes: with (supervised) 18–20,22,23 or without (unsupervised) 16,17 a priori knowledge about the principal components (or eigenvectors) of the hyperspectrum. In this section, we present compressive Raman results without a priori knowledge. The measurement process of the hyperspectrum is done by a combined use of the piezoscanner and the DMD, described elsewhere. 30 Briefly, the sample is illuminated in a raster scan manner over the image plane. For each spatial position, the Raman signal generated by the sample is measured using the compressive spectrometer. Instead of performing a wavelength scan over the spectral domain, the spectral content is measured using multiplexed detection with Hadamard functions generated with the DMD. One key advantage of using Hadamard basis is for increasing the SNR 14,31,32 because half the pixels are always on. This 7

ACS Paragon Plus Environment

ACS Photonics

0.8

Intensity / a.u.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 21

0.6 0.4 0.2 0

1000

1500

2000

2500

Raman shift / cm-1

3000

Figure 2: Compressive Raman imaging without a priori knowledge. Left: Raman pseudocolored image of polystyrene beads obtained by means of computational reconstruction. Right: Spectral content of both species present in the sample, water (green), and polystyrene (red). Compression: 1.7X. Scale bar: 2 µm. Effective pixel dwell time: 15 ms. Total acquisition time: 15 s. sampling scheme can be expressed in matrix form as y = Sx, where the vector y contains each one of the integrated measurements, x is the hyperspectral object (in vector form), and the whole sensing process is described by the linear operator S, which is the Kronecker product of the two sampling bases used here: S = I ⊗ H (I being the canonical spatial-scanning basis, and H the Hadamard spectral-scanning basis). In order to reduce the number of measurements, we select a random subset of rows of S to measure the hyperspectral dataset. Then, we use compressive sensing algorithms to solve the following optimization problem: 1 xk22 + τ kˆ xkT V arg min x ˆ = ky − Sˆ 2

(2)

based on the Total Variation as the regularizing function to estimate the hyperspectrum x ˆ, which has been previously reported to provide good results in multispectral imaging scenarios. 33,34 The reconstructions can be seen in Fig. 2 of a relatively small region of the sample (32x32 spatial positions and 128 spectral channels). The two average spectra show the spectral features expected for polystyrene (peaks at 1600, 2900 and 3050 cm−1 ) and water background (a continuous spectrum without sharp peaks). From the spectral information, we can distinguish between the water content of the sample (coloured green), and the polystyrene beads (red). 8

ACS Paragon Plus Environment

Page 9 of 21

polystyrene

100

B

0.12 0.08 0.04

water

0 300

500

100

300

0.2 0.16

Intensity / a.u.

A

1200

500

1800

2400

3000

Raman shift / cm-1

700

0.12

Polystyrene Silicone oil Water

0.08 0.04 0

1200

1800

2400

3000

1200

1800

2400

3000

1200

1800

2400

3000

Raman shift / cm-1

900

C 200

250nm

0 5

10

15

200

0

0

0

F1 F2 F3

4

y / μm

8

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Photonics

x / μm

Figure 3: Compressive Raman imaging with a priori knowledge. (A) Large scale imaging with 80X compression (axis are in pixel count). Raman pseudo-color images of 1 µm-polystyrene beads (red, upper left inset) and water (green, lower left inset). Spectra are presented in the inset with the optimized filters shown on top. Scale bar: 10 µm. Effective pixel dwell time: 640 µs. Total acquisition time: 292 s. (B) Comparison of optimal and non-optimal filters. The spectra used (top panel) to image with optimal (middle panel) and non-optimal (bottom panel) filters of the three species system. The filters used are presented in the bottom of the images. Scale bar: 10 µm. Effective pixel dwell time: 600 µs. Total acquisition time: 24 s. (C) High-sensitivity imaging of 88 nm-polystyrene beads. Only the polystyrene channel is presented. Scale bar: 5 µm. Effective pixel dwell time: 3 ms. Total acquisition time: 44 s.

Compressive Raman with a priori knowledge In supervised compressive Raman, one estimates the proportions of a linear combination of M eigenspectra (P), from N measurements (x). This is achieved by projecting the input spectrum onto N optimized spectral filters (F) with the DMD. Each spatial pixel can then be written as x = F P> c, where proportions c are estimated from simple inversion of F P> . The key step for proportion estimation is to develop filters that are robust against noise. 18,21 Different from previous methods in supervised compressive Raman, 18 here we use as a metric 9

ACS Paragon Plus Environment

ACS Photonics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

the trace of the Cramer-Rao bound matrix to determine the optimized spectral filters. 21,22 A thorough description of the method, and its relation to other established methods in supervised compressive Raman, can be found elsewhere. 21,22,35 Fig. 3 presents results of large-scale and high-sensitivity chemical imaging. Fig. 3A shows an example of the compressive detection approach where a large-scale high-resolution image (0.5 Mpixel) is presented with chemical selectivity. After generation of the optimized filters (Fig. 3A, optimized filters are presented above spectra in the top-right inset), we obtain x from which we estimate c in a pixel-wise manner. Note that even though x is positive valued, c may have spurious negative values arising from the inversion procedure associated with noise in the measurements. Chemically selective images are then presented in pseudo-color images (Fig. 3A) or analysed separately in their respective channels (Fig. 3A insets shows water and polystyrene proportions). Due to high-resolution imaging, the pseudo-color images are highly contrasted (lack of yellow color). The robustness of the optimal filters is inspected by analysing a three-species chemical system. Fig. 3B shows the spectra of the species and the corresponding false-color image for two sets of filters. The color correlation of the two images demonstrates that no spurious signals are detected since both reconstructions lead to the same chemical image outcomes. However, it is obvious the robustness of the optimized filters (top image) to generate lower noise proportions reconstructions when compared to the non-optimal case (bottom image). For generating the filters, we used the constraint that each spectral bin should be measured just once, for a fair comparison among the different filters choice. 21 Pixel dwell times are defined as the total exposure time for a given spatial point taking into consideration multiple scans for various filters. To further quantify the resolution and demonstrate high-sensitivity detection, we present images of 88-nm polystyrene beads, sizes below the diffraction limit of the system, in Fig. 3C. Despite the low SNR, one observes regions of aggregated beads with high intensities, and isolated ones with much lower intensities. To estimate the resolution of the system, we use a simple deconvolution using the measured line profiles of the low-intensity beads (≈250 nm).

10

ACS Paragon Plus Environment

Page 10 of 21

Page 11 of 21

A

Lipid

Water

B

1.2

Normalized proportion

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Photonics

1

0.8 0

10

y / μm

20

30

Figure 4: High-sensitivity detection of lipid membranes in vesicles. (A) Proportions of lipids (left panel) and water (right panel). The water channel has been scaled to enhance the appearance of the signal dip. Scale bar: 10 µm. Total acquisition time: 30 s. Effective pixel dwell time: 2.4 ms. (B) Normalized (by the image mean) line profiles of the water proportion. The signal dip is due to a depletion of water from the focal volume, which allows estimation of the number of bilayers in the focal volume (see text for further details). We obtain a resolution of ≈230 nm, a value close to the diffraction limit of the system.

Sensitivity to model membranes Finally, we use the compressive spectrometer to demonstrate sensitive imaging of biological specimens with sizes below the diffraction limit. The sample chosen are vesicles made of lipids as they are model systems in cell membrane biology, 8,9 and also because the eigenspectrum is not expected to considerably change for different environments. Fig.4A shows the estimated proportions of the lipid and water response. In the lipid channel, there are clearly various bilayers wrapped around each other, a topology common to thin multilamellar vesicles. 36 In the water channel, there is a slight decrease in the overall water response due to the finite thickness of the lipid bilayers. The drop in intensity in the water channel allows us to estimate the water depletion thickness, as the Rayleigh length is much smaller than the vesicle curvature. From the intensity drop of the water channel (Fig. 4B), we estimate a maximum multilayer thickness of ≈54 nm (the strongest water proportion intensity dip), corresponding to ≈11 bilayers at that location.

11

ACS Paragon Plus Environment

ACS Photonics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Discussion We have presented a new spectrometer layout designed for single-pixel spectroscopy. With this layout, we showcase two types of compressive Raman modality, either without a priori knowledge or with a priori knowledge of the eigenspectra of the hyperspectrum. The main purpose of the high-throughput spectrometer is to enable sensitive measurements of specimens that are smaller than the microscope resolution; Because the surrounding solvent generates a strong background, we reduce its intensity by tight focusing, and also increase the dynamic range of the detection system with low-noise and high-sensitivity detection. Remarkably, we could detect with molecular selectivity few bilayers with the current design. Previously, vesicle systems were mostly imaged by coherent Raman scattering techniques, 29,37–42 which are considerably more expensive and complex than the layout presented here. We estimated that the strongest region in Fig. 4A contains about 11 bilayers. Given the linearity of the spontaneous Raman process on the local concentration, we expect singlebilayer 8,9,27 imaging with further improvements in the current design, for instance, by addition of a second detector 43 and with larger clear-aperture gratings to reduce losses due to clipping. There have been various demonstrations to accelerate Raman imaging for biological applications. These include line scan methods, 6 spatial focii multiplexing, 44 and widefield microscopy. 25,45 High-sensitivity in slit-based spectrometers 6,44 is challenging as non-confocal geometries are hard to reach diffraction-limited performance. 46 In that sense, light-sheet excitation-widefield imaging 45 could probably provide the highest sensitivity, however the whole image plane has to be optically clear and static during acquisition because of the spectral/temporal sampling. The abovementioned methods all use sensitive cameras, contrary to single-pixel approaches which avoid slow and noisy measurements at high speeds (mostly read-out noise of CCDs). 22 In addition, the combination of single-pixel-spectroscopy with tight-focusing (or confocal) allows for low-power CW lasers to be used. Nevertheless, the main disadvantage of compressive Raman with a priori information (supervised) is its 12

ACS Paragon Plus Environment

Page 12 of 21

Page 13 of 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Photonics

need for at least one long scan to obtain the eigenspectra "dictionary", although a "blind" sampling approach to learn the dictionary has been proposed. 19 Note that even in biological applications one expects only a few eigenspectra, 7 which supports that supervised compressive Raman may have the fastest imaging speeds. The high-compression levels offered by supervised compressive Raman imaging opens interesting perspectives for high-throughput spectroscopic imaging. While high-resolution imaging of large field-of-views is currently very challenging with traditional Raman imaging, it is feasible for compressive Raman. 19 If we consider a 10 mm2 specimen imaged at a 0.25 µm resolution, terabytes of data are easily generated with a 1024-pixels-spectral-CCD. Furthermore, these data sizes challenges chemometric tools due to computational power available. With suitable changes of the spectrometer, compressions up to 512X (2 species, with 1024 spectral channels) are envisaged with commercially available DMDs. Therefore, while we did not demonstrate large-scale image because of the slow scanner we used, future implementations using galvanometric mirrors may. We note that the supervised approach does not use optimization routines at the detection stage but rather before the measurement itself, in contrast to the unsupervised approach which uses optimization routines for obtaining the hyperspectrum after the measurement is performed. While here we have used the supervised method on a pixel-wise manner, future work may include proportion estimation of the hyperspectrum as a whole. Previous compressive spectrometers used less sensitive detection. Quantum efficiency and dark counts of the detector are two key parameters for high-sensitivity detection as they determine the overall throughput and dynamic range for proportion estimation in supervised methods. The throughput scaling of the grating-based approach is φ4l φDMD φ2g φd , where φi denotes the throughput of the lenses (l), DMD (DM D), gratings (g) and detector (d) (by detector throughput we mean quantum efficiency). Given that lenses and gratings may have negligible losses, the largest penalty is the DMD and detector quantum efficiency. In our design, we achieve a theoretical throughput of 34% (φg = 85%, φd = 70%, φDMD =

13

ACS Paragon Plus Environment

ACS Photonics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

70%). The analysis here assumes that the whole spectrum is detected (not modulated), as different methods (supervised vs. unsupervised) may have very different spectral amplitude modulation, e.g. supervised methods exploiting two detectors 43 can have φDMD = 70% compared to φDMD = 35% for Hadamard spectroscopy. A higher spectrometer throughput value (50%) is expected if operating in the near-IR as optical elements are commercially available. Note, however, while the sensitivity might be higher, the overall signal will be lower as the Raman cross-section scales as λ−4 . An alternative grating-based approach is to fold the design (back-reflect the light into the same grating), 15 however this would be even more lossy as it does not allow to use two-detectors strategy for supervised methods. 43 Finally, the lens-based approach may reach overall throughput of 43% after meticulous design of the optical elements.

Conclusion High-sensitivity Raman imaging means detecting faint signals over strong idle backgrounds. To address that, we have presented two inexpensive layouts of a high-throughput spectrometer for single-pixel compressive Raman microspectroscopy, and demonstrated its application with two different levels of a priori information used for computational reconstruction. Such layout allowed for high-sensitivity imaging of few-bilayers vesicle in few seconds, therefore demonstrating that compressive Raman is a doable framework for nanoscopic sources detection such as found in cell biology and nanophotonics applications.

Acknowledgement We thank Tom Sperber for fruitful discussions. H.B.A. was supported by LabEX ENS-ICFP: ANR-10-LABX-0010/ANR-10-IDEX-0001-02 PSL*. This works was supported by European Research Council (ERC) (724473), Universitat Jaume I (PREDOC/2013/32), Generalitat Valenciana (PROMETEO/2016/079), Ministerio de Economia y Competitividad (MINECO, 14

ACS Paragon Plus Environment

Page 14 of 21

Page 15 of 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Photonics

FIS2016-75618-R), Agence Nationale de la Recherche (ANR) France Bio Imaging (ANR-10INSB-04-01) and France Life Imaging (ANR-11-INSB-0006) infrastructure networks and Plan cancer INSERM (PC201508 and 18CP128-00). S.G. is a member of the Institut Universitaire de France.

References (1) Kong, K.; Kendall, C.; Stone, N.; Notingher, I. Raman spectroscopy for medical diagnostics–From in-vitro biofluid assays to in-vivo cancer detection. Adv. Drug Deliv. Rev. 2015, 89, 121–134. (2) Harmsen, S. et al. Surface-enhanced resonance Raman scattering nanostars for highprecision cancer imaging. Sci. Transl. Medicine 2015, 7, 271ra7. (3) Jermyn, M.; Mok, K.; Mercier, J.; Desroches, J.; Pichette, J.; Saint-Arnaud, K.; Bernstein, L.; Guiot, M.-C.; Petrecca, K.; Leblond, F. Intraoperative brain cancer detection with Raman spectroscopy in humans. Sci. Transl. Medicine 2015, 7, 274ra19. (4) Lorenz, B.; Wichmann, C.; Stöckel, S.; Rösch, P.; Popp, J. Cultivation-free Raman spectroscopic investigations of bacteria. Trends Microbiol. 2017, 25, 413–424. (5) Bodelón, G.; Montes-García, V.; López-Puente, V.; Hill, E. H.; Hamon, C.; SanzOrtiz, M. N.; Rodal-Cedeira, S.; Costas, C.; Celiksoy, S.; Pérez-Juste, I.; Scarabelli, L.; La Porta, A.; Pérez-Juste, J.; Pastoriza-Santos, I.; Liz-Marzán, L. M. Detection and imaging of quorum sensing in Pseudomonas aeruginosa biofilm communities by surfaceenhanced resonance Raman scattering. Nat. Mater. 2016, 15, 1203. (6) Okada, M.; Smith, N. I.; Palonpon, A. F.; Endo, H.; Kawata, S.; Sodeoka, M.; Fujita, K. Label-free Raman observation of cytochrome c dynamics during apoptosis. Proc. Natl. Acad. Sci. U. S. A. 2012, 109, 28–32.

15

ACS Paragon Plus Environment

ACS Photonics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

(7) Klein, K.; Gigler, A. M.; Aschenbrenner, T.; Monetti, R.; Bunk, W.; Jamitzky, F.; Morfill, G.; Stark, R. W.; Schlegel, J. Label-free live-cell imaging with confocal Raman microscopy. Biophys. J. 2012, 102, 360–368. (8) Ando, J.; Kinoshita, M.; Cui, J.; Yamakoshi, H.; Dodo, K.; Fujita, K.; Murata, M.; Sodeoka, M. Sphingomyelin distribution in lipid rafts of artificial monolayer membranes visualized by Raman microscopy. Proc. Natl. Acad. Sci. U. S. A. 2015, 112, 4558–4563. (9) Donaldson Jr, S. H.; de Aguiar, H. B. Molecular Imaging of Cholesterol and Lipid Distributions in Model Membranes. J. Phys. Chem. Lett. 2018, 9, 1528–1533. (10) Shipp, D. W.; Sinjab, F.; Notingher, I. Raman spectroscopy: Techniques and applications in the life sciences. Adv. Opt. Photonics 2017, 9, 315–428. (11) Duarte, M. F.; Davenport, M. A.; Takhar, D.; Laska, J. N.; Sun, T.; Kelly, K. E.; Baraniuk, R. G. Single-pixel imaging via compressive sampling. IEEE Signal Process. Mag. 2008, 25, 83. (12) Berto, P.; Scotte, C.; Galland, F.; Rigneault, H.; de Aguiar, H. B. Programmable singlepixel-based broadband stimulated Raman scattering. Opt. Lett. 2017, 42, 1696–1699. (13) Chan, W. L.; Charan, K.; Takhar, D.; Kelly, K. F.; Baraniuk, R. G.; Mittleman, D. M. A single-pixel terahertz imaging system based on compressed sensing. Appl. Phys. Lett. 2008, 93, 121105. (14) August, Y.; Vachman, C.; Rivenson, Y.; Stern, A. Compressive hyperspectral imaging by random separable projections in both the spatial and the spectral domains. Appl. Opt. 2013, 52, D46–D54. (15) Davis, B. M.; Hemphill, A. J.; Cebeci Maltas, D.; Zipper, M. A.; Wang, P.; BenAmotz, D. Multivariate hyperspectral Raman imaging using compressive detection. Anal. Chem. 2011, 83, 5086–5092. 16

ACS Paragon Plus Environment

Page 16 of 21

Page 17 of 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Photonics

(16) Galvis-Carreño, D. F.; Mejía-Melgarejo, Y. H.; Arguello-Fuentes, H. Efficient reconstruction of Raman spectroscopy imaging based on compressive sensing. Dyna 2014, 81, 116–124. (17) Thompson, J. V.; Bixler, J. N.; Hokr, B. H.; Noojin, G. D.; Scully, M. O.; Yakovlev, V. V. Single-shot chemical detection and identification with compressed hyperspectral Raman imaging. Opt. Lett. 2017, 42, 2169–2172. (18) Wilcox, D. S.; Buzzard, G. T.; Lucier, B. J.; Wang, P.; Ben-Amotz, D. Photon level chemical classification using digital compressive detection. Anal. Chimica Acta 2012, 755, 17–27. (19) Wilcox, D. S.; Buzzard, G. T.; Lucier, B. J.; Rehrauer, O. G.; Wang, P.; Ben-Amotz, D. Digital compressive chemical quantitation and hyperspectral imaging. Analyst 2013, 138, 4982–4990. (20) Wei, D.; Chen, S.; Ong, Y. H.; Perlaki, C.; Liu, Q. Fast wide-field Raman spectroscopic imaging based on simultaneous multi-channel image acquisition and Wiener estimation. Opt. Lett. 2016, 41, 2783–2786. (21) Réfrégier, P.; Scotté, C.; de Aguiar, H. B.; Rigneault, H.; Galland, F. Precision of proportion estimation with binary compressed Raman spectrum. J. Opt. Soc. Am. A 2018, 35, 125–134. (22) Scotté, C.; de Aguiar, H. B.; Marguet, D.; Green, E.; Bouzy, P.; Vergnole, S.; Winlove, P.; Stone, N.; Rigneault, H. Assessment of compressive Raman versus hyperspectral Raman for microcalcification chemical imaging. Anal. Chem. 2018, 90, 7197–7203. (23) Corcoran, T. C. Compressive Detection of Highly Overlapped Spectra Using Walsh– Hadamard-Based Filter Functions. Appl. Spectrosc. 2018, 72, 392–403.

17

ACS Paragon Plus Environment

ACS Photonics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

(24) Pavillon, N.; Smith, N. Compressed sensing laser scanning microscopy. Opt. Exp. 2016, 24, 30038–30052. (25) Wadduwage, D. N.; Singh, V. R.; Choi, H.; Yaqoob, Z.; Heemskerk, H.; Matsudaira, P.; So, P. T. Near-common-path interferometer for imaging Fourier-transform spectroscopy in wide-field microscopy. Optica 2017, 4, 546–556. (26) Zhang, S.; Song, Z.; Godaliyadda, G. D. P.; Ye, D. H.; Chowdhury, A. U.; Sengupta, A.; Buzzard, G. T.; Bouman, C. A.; Simpson, G. J. Dynamic Sparse Sampling for Confocal Raman Microscopy. Anal. Chem. 2018, 90, 4461–4469. (27) Rieger, S.; Grill, D.; Gerke, V.; Fallnich, C. Quantitative spontaneous Raman scattering spectroscopy in artificial binary lipid membranes. J. Raman Spectrosc. 2017, 48, 1264– 1269, jrs.5205. (28) Bélanger, E.; Bégin, S.; Laffray, S.; Koninck, Y. D.; Vallée, R.; Côté, D. Quantitative myelin imaging with coherent anti-Stokes Raman scattering microscopy: alleviating the excitation polarization dependence with circularly polarized laser beams. Opt. Express 2009, 17, 18419–18432. (29) Gasecka, P.; Jaouen, A.; Bioud, F.-Z.; De Aguiar, H. B.; Duboisset, J.; Ferrand, P.; Rigneault, H.; Balla, N. K.; Debarbieux, F.; Brasselet, S. Lipid order degradation in autoimmune demyelination probed by polarization resolved coherent Raman microscopy. Biophys. J. 2017, 113, 1520–1530. (30) Soldevila, F.; Dong, J.; Tajahuerce, E.; Gigan, S.; de Aguiar, H. B. Fast compressive Raman bio-imaging via matrix completion. Optica 2019, 6, 341–346. (31) Decker, J. A. Experimental realization of the multiplex advantage with a Hadamardtransform spectrometer. Appl. Opt. 1971, 10, 510–514.

18

ACS Paragon Plus Environment

Page 18 of 21

Page 19 of 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Photonics

(32) DeVerse, R.; Hammaker, R.; Fateley, W. Realization of the Hadamard multiplex advantage using a programmable optical mask in a dispersive flat-field near-infrared spectrometer. Appl. Spectros. 2000, 54, 1751–1758. (33) Bioucas-Dias, J. M.; Figueiredo, M. A. A new TwIST: two-step iterative shrinkage/thresholding algorithms for image restoration. IEEE Trans. Image Process. 2007, 16, 2992–3004. (34) Rajwade, A.; Kittle, D.; Tsai, T.-H.; Brady, D.; Carin, L. Coded hyperspectral imaging and blind compressive sensing. SIAM J. Imaging Sci. 2013, 6, 782–812. (35) Cebeci, D.; Mankani, B.; Ben-Amotz, D. Recent Trends in Compressive Raman Spectroscopy Using DMD-Based Binary Detection. J. Imaging 2019, 5, 1. (36) Moreaux, L.; Sandre, O.; Blanchard-Desce, M.; Mertz, J. Membrane imaging by simultaneous second-harmonic generation and two-photon microscopy. Opt. Lett. 2000, 25, 320–322. (37) Müller, M.; Schins, J. M. Imaging the Thermodynamic State of Lipid Membranes with Multiplex CARS Microscopy. J. Phys. Chem. B 2002, 106, 3715–3723. (38) Potma, E. O.; Xie, X. S. Detection of single lipid bilayers with coherent anti-Stokes Raman scattering (CARS) microscopy. J. Raman Spectrosc. 2003, 34, 642–650. (39) Li, L.; Wang, H.; Cheng, J.-X. Quantitative coherent anti-Stokes Raman scattering imaging of lipid distribution in coexisting domains. Biophys. J. 2005, 89, 3480–3490. (40) Cheng, J.; Pautot, S.; Weitz, D.; Xie, X. Ordering of water molecules between phospholipid bilayers visualized by coherent anti-Stokes Raman scattering microscopy. Proc. Natl. Acad. Sci. U. S. A. 2003, 100, 9826–9830.

19

ACS Paragon Plus Environment

ACS Photonics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

(41) Chen, B.-C.; Sung, J.; Lim, S.-H. Chemical Imaging with Frequency Modulation Coherent Anti-Stokes Raman Scattering Microscopy at the Vibrational Fingerprint Region. J. Phys. Chem. B 2010, 114, 16871–16880. (42) Shen, Y.; Zhao, Z.; Zhang, L.; Shi, L.; Shahriar, S.; Chan, R. B.; Di Paolo, G.; Min, W. Metabolic activity induces membrane phase separation in endoplasmic reticulum. Proc. Natl. Acad. Sci. U. S. A. 2017, 114, 13394–13399. (43) Rehrauer, O. G.; Dinh, V. C.; Mankani, B. R.; Buzzard, G. T.; Lucier, B. J.; BenAmotz, D. Binary Complementary Filters for Compressive Raman Spectroscopy. Appl. Spectrosc. 2018, 72, 69–78. (44) Kong, L.; Navas-Moreno, M.; Chan, J. W. Fast Confocal Raman Imaging Using a 2-D Multifocal Array for Parallel Hyperspectral Detection. Anal. Chem. 2016, 88, 1281– 1285. (45) Müller, W.; Kielhorn, M.; Schmitt, M.; Popp, J.; Heintzmann, R. Light sheet Raman micro-spectroscopy. Optica 2016, 3, 452–457. (46) Schlücker, S.; Schaeberle, M. D.; Huffman, S. W.; Levin, I. W. Raman microspectroscopy: a comparison of point, line, and wide-field imaging methodologies. Anal. Chem. 2003, 75, 4312–4318.

20

ACS Paragon Plus Environment

Page 20 of 21

Page 21 of 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Photonics

For Table of Contents Use Only High-sensitivity high-speed compressive spectrometer for Raman imaging Benneth Sturm, Fernando Soldevila, Enrique Tajahuerce, Sylvain Gigan, Herve Rigneault, Hilton B. de Aguiar compressive spectrometer

Raman spectrum

A conceptual drawing summarizing main outcomes: using a programmable and compressive spectrometer to perform fast and sensitive detection of lipid bilayers.

21

ACS Paragon Plus Environment