|Home | About | Journals | Submit | Contact Us | Français|
En face coherence microscopy or flying spot or full field optical coherence tomography or microscopy (FF-OCT/FF-OCM) belongs to the OCT family because the sectioning ability is mostly linked to the source coherence length. In this article we will focus our attention on the advantages and the drawbacks of the following approaches: en face versus B scan tomography in terms of resolution, coherent versus incoherent illumination and influence of aberrations, and scanning versus full field imaging. We then show some examples to illustrate the diverse applications of en face coherent microscopy and show that endogenous or exogenous contrasts can add valuable information to the standard morphological image. To conclude we discuss a few domains that appear promising for future development of en face coherence microscopy.
We would like first to apologize to those who contributed to the progress of this field that we have omitted to mention. As a group of experimental physicists tied to the challenges of optical measurements reaching their physical limits, we switched with enthusiasm from measuring the roughness of the mirrors for gravitational antennas (VIRGO-LIGO project ) to our first tomographic microscopy measurements . The idea of jumping by 9 orders of magnitude from the least scattering media (ultrapure silica optics with a mean free path about 100 Km) to biological samples (mean free path about 100 µm) was indeed a very stimulating jump. We have to confess that this jump required very minor changes as illustrated in Fig. 1 that shows our setups: the Wollaston prism that was used for measuring roughness with a “picometer” sensitivity common path interferometer was simply replaced by a straightforward Michelson polarization interferometer. Thankfully both were shot noise limited, but the limitations of the technique were quickly revealed when using a camera instead of a single detector, as we will discuss below. Our tomographic microscopy approach was inspired by an illuminating paper about OCM, explaining and demonstrating the gain that could be expected from such an approach  compared to confocal microscopy, as well by the first papers on OCT .
Our motivation to write this paper is not to go back through historical memories in a detailed review but rather to discuss important points that, we think, have contributed to the development of en face tomo-microscopy. Hundreds of valuable references dealing with en face tomographic microscopy can be found in a number of chapters of the books edited by W. Drexler and J. Fujimoto ; we will refer to a few of these chapters throughout this article.
When we heard first about OCT it was said that this technique was equivalent to ultrasonography (medical ultrasound) with a wavelength 1000 times shorter, so micrometer resolution ranges rather than millimeter ones were expected. Medical ultrasound displays so-called B-scan images and so did the time domain, and later the Fourier domain (spectral as well as swept source) OCT . This kind of cross-sectional depth exploration of the samples and in particular of the retina was found to be of importance to reveal pathologies and to follow up their evolution. Nevertheless doctors often asked for en face images that resembled their habitual fundus SLO (scanning laser ophthalmoscope) views, and, due to the progress in speed of the OCT systems such images were obtained from 3-D stacks of images within a reasonable amount of time, compatible with an eye examination. The main motivation to perform en face OCT is mostly to be able to get a standard microscopy resolution: for this reason we started to call it Full Field Optical Coherence Microscopy (FF-OCM) and to be sure to be adopted by the OCT community we often switched to Full Field OCT (FF-OCT) [5,6]. In this paper we will try to stick to the term OCM (full field or scanning) because the “microscopic” aspect was the main focus: while OCT features high axial resolution, OCM focuses on achieving both high lateral and axial resolution. The reader should be aware that a number of papers dealing with OCM use either the OCT or the OCM nomenclature [8,9] [10,11]. For more specific information about this technique, we refer the reader to book that is entirely dedicated to various aspects of FF-OCT .
In OCT, an axial line (A-scan) is recorded first, followed by transverse scanning in order to construct a B-scan. The required depth of field imposes working with low NA objectives that correspond to the exploration depth (e.g. to cover the entire retinal thickness), thus limiting the lateral resolution. Typical OCT lateral resolutions imposed by the depth of field lie in the 10-20 μm range. For OCM approaches there is no need to work with a large depth of field because the images are acquired at a given depth. In OCM, the axial resolution is mostly imposed by the source spectrum (1 μm with thermal sources, 5-15μm with LEDs), while typical transverse resolution with high NA objectives is around 1 μm. Nevertheless we should mention that various other approaches have successfully overcome the OCT lateral resolution limitation.
Typical OCT transverse resolutions imposed by the depth of field lie in the 10-20 µm range. Nevertheless we should mention that various approaches have successfully overcome this limitation on lateral resolution:
After using the first setup described in Fig. 1, we decided to use a different interferometer (Fig. 3, left) using 2 identical objectives that allow a much larger numerical aperture  (at that time we were not aware that it was proposed by Linnik!) and simplified the reference arm modulation by using a piezoelectric transducer.
The use of high NA microscope objectives imposes recording of en face images either by using a camera or a scanning spot (i.e. flying spot approach [25–27]), or by scanning the sample  in a plane perpendicular to the optical axis.
As mentioned before in terms of sectioning ability we expect a sub-micron axial resolution when using a broadband thermal source  with a smooth spectral distribution only achievable with sophisticated mode locked laser in the sub-five-femtosecond range [29,30] [31,32]. Single LEDs and SLDs sources offer similar axial resolution in the 10 μm range. In the case of very large NA objectives the depth of field happens to be smaller than the coherent sectioning; in this case the axial OCM sectioning is the same as that of a confocal microscope of the same numerical aperture .
When using optics of similar numerical aperture, we should be able to reach the same lateral resolution whatever the image acquisition system (i.e. flying spot C-scan or Full-Field OCM).
In terms of signal-to-noise ratio (SNR) the main difference between the various approaches is related to the energy available per voxel. We will suppose, and this is often the case in practice, that all the signals are shot noise limited. In order to provide a quantitative comparison between the different approaches we have to consider quantities such as source power, speed and complexity of the setup. Until recently 1 Mega-pixel cameras that have been used exhibit a typical full well capacity (FWC) of about 100 Ke-/pixel with speed around 100 frames/s meaning that they can only handle a low average power of about 0.1 mW for the field of view. A valuable improvement has been obtained recently with a specially designed camera made by Adimec within the framework of an EC project (CAReIOCA) led by LLTech . This camera (500-700 frames/s, FWC of 2Me- and 2 Mega-pixel), can handle a power > 1mW. Because the reference signal can be damped without S/N losses, tens of mW could impinge on the sample.
This power is one to two orders of magnitude lower than SLDs or lasers used in scanning FF-OCM that have typical power of 10 to 100 mW. A gain of 10 to 20 dB is thus expected in terms of ultimate sensitivity.
In terms of speed, fast scanners can work at 10 to 100 KHz . This corresponds to a time of 1/10 to 1/100 s for a 1 Mega-pixel B-mode image. If the acquisition times are now in the same range in terms of speed (voxel/s), as we mentioned the signal to noise ratio is expected to be higher with a flying spot system and a single detector that will not suffer from saturation as the camera pixels do.
Ultimately, the limit on power may be imposed by the sample itself: if skin can handle a few mW/mm2 the most critical part of the body is the retina that cannot accept more than around one tenth of this value in the near infrared.
One has nevertheless to keep in mind that recording a full field with a camera is always much slower than recording a spectrum on a linear array (in Fourier or spectral domain OCT) and that in presence of sample motion, the signal could be lost as acquisition of multiple images is necessary to process the modulated signal. On the contrary each line of a B-scan can be correctly positioned in post-processing. This explains why only a limited number of in vivo studies have been done so far using FF-OCM on eye [36,37] or skin .
Another successful approach of en face imaging that has been coupled to broadband spatially coherent sources in order to perform sectioning is Holographic Coherence Imaging [39–41]. Instead of working in the image plane working in a Fourier plane offers a number of advantages, such as the possibility to record either on a fast photorefractive material or on a digital camera, to be shot noise limited even with noisy laser sources by spatial filtering, and finally to be able to capture single shot images avoiding the need for phase modulation [42,43].
Going back to en face imaging, spatially coherent sources can be used to illuminate the field of view: they offer a number of advantages such as single shot imaging (i.e. without modulation  or easy dark field imaging  although dark field has recently been proven using incoherent sources ). Moreover, adaptive optics for aberration correction was coupled to OCT for better resolution of the retinal structure in B scans [46,47,49] and in C scans , and for correcting the wavefront in high-resolution 2 photon microscopy . However Karamata et al. demonstrated experimentally and explained in depth in [51–53] that coherent illumination is a source of crosstalks that may quickly degrade the image quality. Crosstalk does not appear with incoherent illumination.
Here we will draw the attention of the reader to the role of spatial coherence: usually when scanning the field of view for a flying spot image, each acquisition (voxel or line) is an independent event and this is equivalent to an incoherent illumination. Surprisingly scanning and incoherent illumination are not equivalent in presence of aberrations: Indeed, in spatially incoherent full-field interferometer, the system PSF width is almost insensitive to aberrations and only the signal amplitude is reduced by the wavefront distortion (Fig. 4). This has been theoretically analyzed and compared with different coherent interferometers, and experimentally confirmed for defocus and random aberrations [54,55]. The unexpected unique merit of resolution conservation is only found in full-field spatially incoherent interferometer. If we consider the aberrated sample wavefront associated with a diffraction-limited spot in the sample arm in the presence of aberrations, the partial overlap with the corresponding diffraction spot reference wavefront damps the signal. Nevertheless the interferometric signal of the aberrated sample wavefront with a neighboring diffraction spot is damped to a greater extent. Indeed, the shift to the next spot is associated with a linear increase of of the phase shift from one side of the pupil to the other. This supplementary phase shift increases the RMS value of the aberrated wavefront leading to a strong decrease of the signal for this neighboring point.
Finally, on considering the use of en face OCM compared to other techniques based on the use of optical contrasts, it appears that non-linear microscopy is most likely to compete with FF-OCM. Indeed the use of femtosecond lasers opens the way to both two photon endogenous fluorescence signals (from NADH in cell cytoplasm, that shows cell sizes and contours, and elastin fibers in connective tissue) and second harmonic generation that is mostly induced by the crystalline structure of various kinds of collagen. Collagen and connective tissue also contribute to the backscattered signals that are at the origin of the FFOCM signals; these rather strong signals often mask the low level of backscattered light from cells that only show up using dynamic FF-OCM (see section 5). As discussed in detail in reference , dynamic and static FF-OCM are complementary modalities that can be automatically superposed on the same field of view.
The table below intends to summarize the main features of typical OCT versus various typical OCM approaches.
|3 to 10 μm|
|3 to 8 μm|
0.5 to 2 μm
|3 to 8 μm|
0.5 to 2 μm
|0.5 to 8 μm|
0.5 to 2 μm
|100 Gpixel/s||1-10 Gpixels/s >100 Gpixels/s with Spectral or Sw. source.||10-50 Gpixels/s 1 processed image||10-50 Gpixels/s|
1 processed image
|Thermal Sources LEDs|
|Typ. Power (sample)|
|10-100mW 100/110 dB||1 mW||0.5-2 mW |
|Aberrations Corrections||With Adaptive Optics||With Adaptive Optics||With Adaptive Optics||No loss of resolution (only signal reduction)|
|Sensitivity to movements||Correction possible||Correction possible||Correction unlikely||Correction unlikely|
Depth of field and aberration correction
|Synthetic aperture, Aberrations correction||Depth of field and aberration correction achieved||Numerically possible||Difficult|
|Main Drawbacks||Low resolution||Price of fs lasers||Crosstalks||Lower sensitivity: camera saturation. Sensitive to movements|
|Main Advantages||Broadly used Easy to implement||S/N. Coupling with non-linear microscopy||Possible wavefront shaping||Cheap, High resolution. Sub-cellular metabolism|
As for the various OCT domains that have been studied over the past 25 years, en face OCM has mostly been devoted to medical diagnosis. Because en face imaging gives more flexibility in terms of lateral resolution, FF-OCM or C-scan OCM has targeted complementing histology for intra operative diagnosis [57–60]. With the new fast camera, 1cm2 FF-OCM images with sub micrometer sampling (>100 Mega-pixel images) can be recorded within one minute with a suitable SNR . A large number of images of different pathologies, mostly cancers [61,62], are available on the LLTech web site as well as a number of articles that demonstrate high levels of sensitivity and specificity, often over 90% .
We would like to provide here a few examples outside the well-covered field of medical diagnosis. En face OCT has been used for a number of biological applications competing with phase contrast microscopy . For instance the kind of sectioning of the bed bug shown in Fig. 5 would have been impossible in transmission phase contrast microscopy.
OCT also proved very useful in material characterization and non-destructive testing [65,66], while FF-OCM with its high sensitivity and high sectioning ability was used to check local defects in optical interference coatings  and holography for analyzing various kinds of structures [39–41].
At the limit between characterization and biometrics, OCT has been used to provide internal fingerprint images that cannot be counterfeited [68–71]. In the case of B-mode OCT an en face image must be deduced from a 3-D acquisition. On the contrary FF-OCM is able to directly capture the en face image of fingerprints  as can be seen in Fig. 6 where internal and external fingerprints are compared.
Even if biomedical applications of OCT dominate in terms of publications, the field of OCT for art is very active  and an up to date web site can be consulted . As most art objects are designed for an en face view, FF-OCT seems a natural way to examine them. Latour et al. used FF-OCM to analyze the structure of paint layers at high resolution  and various objects such as wood, lacquer (Fig. 7), amber etc. have been examined .
Despite worldwide success in the field of biomedical studies, OCT has often been associated with another modality in order to add information and to increase the sensitivity and specificity of the diagnosis.
As we have mentioned and we will see later many en face studies were inspired by similar preceding studies using OCT. On the other hand, to the best of our knowledge, the first multimodal imaging system that combined OCT and fluorescence was an en face one that mixed en face flying spot coherent microscopy with two photon fluorescence of calcium green dye  in rat brain. If OCT in general, and en-face OCT in particular, can provide a high-resolution, high sensitivity structural image of a tissue, it suffers from a lack of cellular and molecular specificity. Even performed with high numerical aperture objectives, en face OCT can only discriminate two cell populations based on their shapes, but not on their molecular content, or on their function. Fluorescence techniques can reveal or complement the OCT information that is obtained from a cell or a structure [77–80] using various approaches: single or 2 photon fluorescence, superficial observation or endoscopy [81–84]. Following similar developments in OCT [85–88], several multimodal setups associating full field microscopy and various fluorescence techniques have been developed [89–91]. While confocal or multi-photon microscopy can be naturally associated with point scanning OCT systems, it is less natural for full field coherence microscopy, where widefield fluorescence techniques are preferred. Therefore, most of the setups combining Full-Field OCM and fluorescence microscopy used Structured Illumination Microscopy (SIM) in the fluorescence channel to generate an optical section. This consists of conjugating a physical grid, often named a Ronchi ruling, with the focal plane of the objective. Fluorescent structures in focus are multiplied by the grid pattern intensity, whereas the grid is blurred outside of the focal plane. In this way, similarly to FF-OCM phase stepping, the out-of-focus fluorescence can be excluded by sequentially moving the grid in three, four, or multiple positions and computing the analog of the FF-OCM amplitude with similar algorithms. Ideally, two cameras are required for combining FF-OCM and SIM, as technological requirements of FF-OCM and fluorescence cameras are almost opposite. References  and  have proposed a similar configuration, in which a flip mirror enables acquisition of either an FF-OCT or a SIM image. Thanks to the sequential imaging of the two modalities, there are less spectral limitations for the two modalities combined. In practice, a white halogen lamp can illuminate the FF-OCM setup to provide an optical section as thin as 1 µm, while any kind of visible fluorophore can be used for fluorescence in parallel, because the two images are acquired one after the other. On the contrary , suggested a simultaneous configuration, in which a dichroic filter combines the fluorescence and the FF-OCM paths. Both modalities are therefore captured at the same time. This configuration can help to decrease the acquisition time, but the FF-OCM source and the fluorescence spectra should not overlap. We recently built a similar simultaneous configuration and demonstrated its interest for dynamic measurements including dynamic FF-OCM (see below) . Moreover, using high NA objectives enabled us to rely on both the spatial optical sectioning, and the temporal optical sectioning . Using this setup, structures invisible in FF-OCM were revealed [93,94]. With this multimodal setup, we have been able to add specific contrast to the structural FF-OCT image: Fig. 8 shows an example revealing nerves and nerve terminals at different depths in a mouse cornea using a fluorescent label that specifically labels nerves in a transgenic mouse line. In this example, FF-OCM can clearly identify the corresponding corneal layers and their associated structures that help to localize the nerves in 3D. We could show that most of the fine sub-basal nerves in the cornea were located at the surface of Bowman’s layer, as expected from literature [95,96]. Interestingly, the thickest nerve fibers can be detected with FF-OCM at Bowman’s layer, but it fails to detect thinner nerves, the nerve terminals, and most of the thick stromal nerves.
Motion is a key factor in living systems and one can use FF-OCM to assess the evolution of structures over time. Blood flow  can be seen and analyzed in this way and individual erythrocytes can be tracked if the frame rate of the camera is fast enough compared to the rate of flow of red blood cells . Dynamic FF-OCM  aims to highlight another kind of motion, i.e. motion taking place inside cells and contributing to activity To extract dynamic FF-OCM signals, we record a short movie (about 1 second) of en face tomographic images and then process the time series obtained for each pixel. The fluctuations of the backscattered intensity are linked to the local metabolism inside the sample. Dynamic FF-OCM characterizes the local metabolism by the amplitude of the fluctuations and their characteristic time scale (Fig. 9).
We could also take advantage of this dynamic nature of the cells in fresh tissue to add new contrast in retinal imaging. Thanks to its high spatial resolution, FF-OCM can reveal valuable details in the retina . Notably, in the ganglion cell layer, we are able to separate the contribution of the strongly scattering ganglion cell axons from that of the capillary walls, and from the weakest scattering structures, i.e. the cell somas.
Figure 10 left presents a typical FF-OCM image of the retinal ganglion cell layer in mouse. First, as presented on the right in Fig. 10, simply looking at the amplitude of the temporal fluctuations, we can hide the signal from stationary axons and blood vessels and add an intracellular contrast, which allows the detection of some cell nuclei and enhances the signal from the red blood cells trapped inside the capillaries, thanks to their highly fluctuating membranes. One can then consider the relative frequencies of the temporal fluctuations. We can emphasize faster moving structures, or intracellular components. For example, in Fig. 10 center, we have created an RGB image from the slowly varying structures (< 0.5 Hz) displayed in red, the intermediate structures in green, and the fastest structures (>10 Hz) in blue. This display enables the identification of two cell populations: round cells with almost no intracellular inhomogeneity, and fewer cells with a faster cytoplasm, and a slower dark center, that could correspond to the nucleus. We postulate that these two cell populations could correspond to different cell types , which would offer a way to discriminate specific cells without labeling in live retinal explants.
Whether using OCM  or Holography [101–103] we have to underline the importance of the dynamic approach and emphasize that similar dynamic approaches can track the evolution of cellular activity  or action of drugs [105,106] down to the nucleus scale .
Knowing the elastic properties of tissues has proved to be valuable information to complement ultrasound imaging. Since the pioneering work of Schmitt  followed by the analysis of arterial stiffness of Bouma’s group [109,110] an increasing number of studies combining OCT and elastography emerged . For small strains, the variations in phase from usual B-scan images are sufficient to obtain the required information and in vivo experiments have been carried out successfully . For a time dependent tissue deformation Doppler-like signal processing was used [111,112].
However, in soft tissues, deformation occurs in 3D. 3D images are therefore recorded before and after deformation, and the strains are deduced from the 3D correlation of the two fields [113–115]. This was also the approach that we used using OCM at the micrometer scale, however, the effective resolution is unfortunately degraded by the correlation procedure (Fig. 11) .
A different way to measure stiffness is to induce strain waves that propagate through the tissue. This approach has been performed by a number of groups and often called dynamic OCT [117,118]. Surface waves [119,120] that could be generated without contact are often used for the cornea but also for tumor characterization . We have used OCM coupled to an ultrafast camera to follow strain waves during their propagation .
All these dynamic studies are limited in terms of spatial resolution by the size and/or the frequency of the source that induces the waves.
We will end our examples of multimodal approaches; other examples can be found in reference 4 and include in particular photothermal effects that become important with the contrast induced by various kinds of nano particles [123,124].
We have tried in this article to review the main differences between en face and B-mode coherence imaging approaches; from the hardware point of view the most important point is the high lateral resolution, achieved by using high numerical aperture microscope objectives for en face imaging. We have then shown that full field coherence microscopy is able to capture millions of pixels images without scanning but carries drawbacks linked to the saturation level of cameras. Flying spot coherence microscopy is less limited in terms of usable power, flexibility to use Bessel beams etc.
Despite some notable technological progress, the frontline of the field, in our opinion, lies rather in the mathematical/digital methods that improve SNR and available depth, along with the possibility of adding complementary contrasts to the OCT/OCM morphological information.
Concerning the complementary contrasts we have pointed out a number of fields of application where en face images are more likely to be used and shown that, as in B scan approaches, multimodal imaging using a combination of contrasts is highly valuable to complement the structural information that we usually obtain.
Nevertheless, a major limitation of FF-OCM compared to B-scan OCT remains its penetration depth. Along with peers involved in coherent imaging, we think that this limitation on accessible imaging depth is due to multiple scattering that happens in highly scattering media [125,126]. Confocal filtering in flying spot OCT acts as a filter for the multi-scattered light but an equivalent filtering happens in (33) in full field detection. The fact that en face OCM is more sensitive to multiple scattering is linked to the geometrical aberrations that are more likely to show up at large aperture. Indeed it is demonstrated [130, supplementary] that, at a given depth, the ratio of Single-Scattering/Multiple-Scattering scales as the Strehl ratio (diffraction limited PSF area/aberrated PSF area). If we consider low order aberrations such as defocus or spherical aberration, the wavefront distortion scales as the square or the forth power of the numerical aperture. Because of the associated damping of the ballistic photons it is then more likely that multiple scattering affects large numerical aperture experiments.
Inspired by experiments in ultrasonic acoustics [116–118], based on the acquisition of a time dependent reflection matrix and theory based on singular value decomposition of this matrix, an en face microscopy matrix approach to optical imaging has recently been proposed. The aim of such an approach that is able to more efficiently filter the multiple scattered light is to push back the imaging depth limit . By combining matrix discrimination of ballistic waves and iterative time reversal, it was demonstrated both theoretically and experimentally that an extension of the imaging-depth limit by a factor larger than two was possible, rendering greater depth than OCT attainable while keeping the transverse resolution of OCM.
We warmly thank Egidijus Auksorius for the fingerprint images, Jean-Marie Chassot and Romain Garrouste for the bug OCM images, Nacim Bouheraoua and Djida Ghoubay for the corneal and retinal tissues.
This work was supported by grants from the European Research Council SYNERGY Grant scheme (HELMHOLTZ, ERC Grant Agreement# 610110; European Research Council, Europe) and LABEX WIFI (Laboratory of Excellence ANR-10-LABEX-24) as part of the French Program ‘Investments for the Future’ under reference ANR-10-IDEX-0001-02 PSL.