2.1.

Working Principle and Basic Structure of FF-OCT

FF-OCT has simplified the system structure of conventional point scanning TD-OCT and line scanning FD-OCT by performing plane detection based on full-field illumination. Meanwhile, its acquisition speed is enhanced by full-field imaging via the CCD or CMOS camera.¹⁴ The imaging principle of the FF-OCT system is shown in Fig. 1. Its basic structure is a low coherence interference microscope system with a Linnik interference structure [Fig. 1(a)], spatial incoherent broadband light source. The plane array detector acquires the en face 2D image of the sample during one exposure.²⁰ The frequently used interference structures in FF-OCT are shown in Fig. 1(b), including Michelson, Linnik, and Mirau interference structures.^{14,20,33–35} The Linnik interference structure is the most common type used in FF-OCT, whose two identical high NA microscope objectives are placed on the sample arm and the reference arm, respectively. The optical path of the two arms can be flexibly adjusted, and the optical path length as well as the focus can be modified independently. No other optical path exists between the objectives of the two arms and the imaging target, thereby extending the selectable range of NA and working distance of the objectives. Using microscope objectives with high NA, FF-OCT can achieve ultrahigh spatial resolution (axial and transverse) at the subcellular level and obtain real-time noninvasive images of the sample tissues [Fig. 1(c)], whose detection results are comparable to those of tissue biopsy. Besides the interference type, different overall structures of FF-OCT systems have been proposed, for example, external spectral modulation FF-OCT system, endoscopic FF-OCT system, and compact miniaturized FF-OCT system. In FF-OCT devices with external spectral modulation, imaging utilizing spectrally modulated, spatially incoherent illumination and a static Linnik interferometer. The first demonstration of spectrally modulated imaging through a multimode fiber optic bundle represents an important step toward achieving endoscopic FFOCM.³⁶ In a compact miniaturized FF-OCT device, the interferometer part of the setup was designed to be as compact and light as possible ($11\times 11\times 5{\mathrm{cm}}^3$, 210 g), while keeping a Linnik interferometer design, which lays the foundations for a handheld device.³⁷

Fig. 1

(a) Schematic diagram an FF-OCT. (b) Different optical set-ups of FF-OCT: Michelson-type, Linnik-type, and Mirau-type. Adapted from Refs. 19, 20, and 33. (c) FF-OCT images and tissue biopsy images of human forearm skin. Adapted from Ref. 22. (d) Tomographic reconstruction of simulated interference fringe with SPA and FPA, respectively. (e) Tomographic reconstruction of CaF2 crystal with SPA and FPA, respectively. (f) Tomographic reconstruction of embryo sample with SPA and FPA, respectively. Adapted from Ref. 19.

The tomographic images of FF-OCT are reconstructed from the interference signal acquired by its detector. The reconstruction scheme varies based on different light sources and signal detection methods. According to the light sources and signal detection methods, FF-OCT is categorized into two types, i.e., time-domain full-field optical coherence tomography (TD-FFOCT, developed from TD-OCT) and Fourier-domain full-field optical coherence tomography (FD-FFOCT, developed from swept-source OCT).

TD-FFOCT adopts the spatial incoherent broadband light source for illumination and can obtain the tomographic interference spectrum data of the sample front without transverse scanning, thereby enabling rapid 3D imaging. In TD-FFOCT, the interference signal captured by plane array detector, contains sample information and serves as the basis of tomographic image reconstruction. The signal of each pixel in the interferogram can be expressed as

In general, to balance the acquisition time and reconstruction quality, the four-step phase-shifting algorithm (FPA) is the most adopted method for reconstruct tomography; the signals of four interferograms with a phase interval of $\pi /2$ are assigned as $I_1$, $I_2$, $I_3$, and $I_4$, corresponding to $\alpha$ as 0, $\pi /2$, $\pi$, and $3\pi /2$, respectively, in Eq. (1). The intensity $A(x,y)$ and phase $\varphi (x,y)$ maps are calculated using Eqs. (2) and (3):

The five-step method is similar to FPA, except that it performs reconstruction based on five interferograms with different phases; therefore, the reconstruction speed of the five-step method is slower than that of FPA; its intensity and phase maps can be calculated using Eqs. (4) and (5):

Based on the Hilbert transform, SPA can reconstruct tomography with only two interferograms, thereby accelerating the imaging speed.⁴⁰ The signals of interferograms $I_a$ and $I_b$ with a phase difference of $\theta$ are described using Eqs. (6) and (7):

Notably, as shown in Figs. 1(d) and 1(e), the fringes cannot be completely eliminated in the reconstruction of FF-OCT tomographic interference images via SPA based on Hilbert transform. However, when SPA based on the Hilbert transform is adopted for the tomographic reconstruction of biological samples, almost no issue of residual fringes exists because the randomly distributed micro/nanostructures in biological samples make the fringes in the reconstructed image invisible, as shown in Fig. 1(f).¹⁹ A simplified algorithm for digital fringe analysis in two-wave interferometry with sinusoidal phase modulation was proposed in 2017. The algorithm consists of simple mathematical combinations of four frames obtained by integration by a camera of the time-varying intensity in an interference pattern during the four successive quarters of the modulation period. The algorithm is invariant by circular permutation of the four image frames. Any set of four consecutive frames can be used for the calculations, which simplifies the practical implementation of the method.⁴⁴ In addition to the above-mentioned algorithms with sinusoidal phase modulation, there is also the “Larkin” algorithm for digital envelope detection, which is a simple but highly effective nonlinear algorithm for envelope detection in white light correlograms. The method is applicable to any bandpass signals where either the envelope or the phase, or both, need to be detected.⁴⁵

FD-FFOCT adopts spatial coherent swept light source combining planar array detector to perform time-sharing detection, which can achieve parallel detection in both frequency and spatial domain. The system structure is shown in Fig. 2(a); the axial information of spatial domain can be obtained from the frequency-domain information acquired by a pixel on the plane array detector.^46,47 The image reconstruction aims to analyze the frequency-domain information of a plane (such as $512(x)\times 512(y)$] then obtain the 3D volumetric data of spatial domain (as $512(x)\times 512(y)\times 1024(z)$]; the processing flow is shown in Fig. 2(b). FD-FFOCT can conduct rapid 3D imaging without axial or transverse scanning, thereby overcoming motion artifacts caused by mechanical dither and repeated motion showing higher sensitivity and imaging speed. However, swept source is a type of spatial coherent light, and the backscattered light information of the corresponding sample points will be mapped to the wrong pixels during parallel detection, resulting in optical crosstalk and affecting the imaging results. The schematics of crosstalk generation are shown in Fig. 2(c). Reducing the spatial coherence of the light source or using multiangle illumination can diminish crosstalk and improve the imaging quality to a certain extent. Figure 2(d) shows the cross section of the 3D reconstruction of human forearm skin at different depths with reduced crosstalk.⁴⁶

Fig. 2

Schematic diagram of an FD-FFOCT system that can suppress crosstalk and speckle. (a) Optical set-up. (b) Reconstruction process of the FD-FFOCT system. (c) Crosstalk caused by incorrect backscattered ray mapping. (d) Human forearm skin images acquired by the cross-talk-free FD-FFOCT. (a) and (c) Reprinted from Ref. 46.

2.2.

Key Performance Parameters of FF-OCT

The main performance parameters of conventional OCT and FF-OCT include axial resolution, transverse resolution, penetration depth, imaging field, imaging speed, detection sensitivity, and motion artifacts. Each performance parameter usually is mutually restricted.

The axial resolution of FF-OCT mainly depends on the coherence length of the light source, whose theoretical value is calculated using Eq. (9):

The transverse resolution of FF-OCT is determined by the half peak width of the transverse point spread function, whose theoretical value is calculated using Eq. (10):

Imaging field, as a significant parameter of the system, is related to the NA of the microscope objective and the central wavelength of the light source; its theoretical value can be estimated using Eq. (11):

Light source parameters, such as wavelength, bandwidth, spatial incoherence, and power will directly impact the imaging performance of the system. Figure 3(a) shows the absorption curves in each spectral band of several common molecules or substances in biological tissues, including water, aorta, oxygenated hemoglobin (HBO), and melanin. Considering the spectral response range and tissue imaging window of the plane array detector, the central wavelength of common light sources for FF-OCT is mainly 500 to 1000 nm.⁴⁴ In this spectral band range and the NA range of 0.1 to 1.2, the transverse and axial resolutions of FF-OCT system are visualized, and the results are demonstrated in Figs. 3(b) and 3(c). The axial/transverse resolution data of typical light sources and microscope objectives are given here, as shown in Table 1. The types of light sources include light-emitting diode (LED), tungsten halogen lamp (THL), superluminescent diode (SLD), THL was first applied to the FF-OCT system in 2002.⁵⁰ In the recent decade, due to the development of LED technology, the use of LED light sources became more and more common. These light sources are covered in Table 1. The central wavelength of the light source and NA of the microscope objectives can be preliminarily determined based on these results.

Fig. 3

(a) Spectral absorption curve of common substances or molecules in biological tissues. Adapted from Ref. 48. (b) The influence of NA and center wavelength on transverse resolution. (c) The influence of spectral bandwidth and central wavelength on axial resolution.

Table 1

Axial/transverse resolution data for common light sources and microscope objectives.

The detection sensitivity of the system can be expressed by the minimum intensity signal that can be detected by the detector, which is one of the key parameters affecting the contrast and imaging depth of the tomogram. Its theoretical value can be calculated using Eq. (12):

The detection sensitivity is defined as $10\times \log (R)$, where $K$ is a fixed constant, $\xi$ is the full-well capacity of the camera, $M$ is the superposition times of interference signals, $\chi$ is the electrical noise signal during data collection, $R_{\mathrm{inc}}$ is the light reflectivity of the sample, and $R_r$ is the light reflectivity of the reference mirror in the reference arm. The detection sensitivity of FF-OCT system can be enhanced by increasing the full-well capacity of the planar array detector, using a higher-power light source, and increasing the superposition times of interference signals.²⁷ The performance of the camera affects that of the FF-OCT system; Professor W. Y. Oh’s team first applied an indium-gallium-arsenide (InGaAs) camera to the FF-OCT system in 2006. InGaAs area array cameras with a good responsivity in the 0.9 to $1.7\mu \mathrm{m}$ wavelength range may enable imaging at longer wavelengths, thereby increasing the penetration depth of FF-OCM.⁵¹ Moreover, the use of water or oil immersion on microscope objectives not only can improve the matching degree of the microscope objective and sample refractive index but also enhances the system sensitivity to a certain extent.^19,41

3.1.

Increasing the Imaging Speed

The area array sensor and high NA microscope objectives adopted by FF-OCT can achieve submicron-level resolution while ensuring fast imaging speed. However, realizing real-time in vivo imaging remains challenging, and enhancing the overall imaging speed of the system is still a research focus. Based on the imaging principle of FF-OCT, reconstruction algorithms involving fewer interferograms as well as phase shifts with higher speed and accuracy can enhance the imaging speed of the system. The specific reconstruction algorithm of FF-OCT tomography is detailed in Sec. 2.1. The acquisition of interferograms with different phases is achieved by introducing a phase shift, which can be categorized into two methods depending on whether mechanical displacement is involved: mechanical phase shifting using the piezoelectric ceramic (PZT) phase shifters and instantaneous phase shifting without mechanical displacement using wave plates.

PZT phase shifting is the most common approach, which uses the piezoelectric effect to produce accurate small displacement to change the position of the reference mirror for phase shifting. PZT phase shifting has two methods, i.e., fixed-step phase shifting and carrier frequency phase shifting. The phase of the fixed-step approach changes discretely between intensity measurements, and the data acquisition is conducted after phase shifting. The carrier frequency method uses sine waves to enable the reciprocating motion of PZT and uses a square wave with a frequency four or five times the sine wave to trigger the detector and collect interferograms, which can accelerate the imaging speed but remains in the category of discrete phase shift.⁵⁰ By combining the PZT phase shift with the instantaneous phase shift using wave plate, two movements of PZT can achieve four-step phase shift results by constructing the optical Hilbert transform.⁵² This dual-channel structure with two detectors can enhance the overall phase shifting speed. In the genuine instantaneous phase shift without mechanical displacement, the combination of a half-wave plate and linear polarizer introduces a phase difference of $\pi$ between the optical paths of two arms; two interferograms with a phase difference of $\pi$ are simultaneously acquired by two cameras. The SPA based on the Hilbert transform is then used to reconstruct the tomography, the construction of which is shown in Fig. 4(a).

Fig. 4

(a) Schematic diagram of dual-channel instantaneous phase-shifting FF-OCT set-up. Adapted from Ref. 40. (b) Instantaneous phase-shifting FF-OCT set-up of a single channel with two divisions. Adapted from Ref. 39. (c) Instantaneous phase-shifting FF-OCT set-up of a single channel quarter, and four interferograms collected at the same time with a phase interval of $\pi /2$. Scale bar: $100\mu \mathrm{m}$. Reprinted from Ref. 41.

Industrial cameras with high well capacity and high frame rate are expensive; thus, using two cameras would increase the system cost. The dichotomy of a single channel can be achieved using a Wollaston prism, then different sensor regions of one camera can simultaneously detect two quadrature polarized light fields, as shown in Fig. 4(b).^39,43 Currently, the instantaneous phase-shifting structure with a quartered single channel has also been achieved, as shown in Fig. 4(c), which can simultaneously acquire four phase-shifting images using a single camera. Compared with the PZT four-step phase shift, this method can significantly enhance the imaging speed. Instantaneous phase-shifting approaches with a dichotomized or quartered single channel can reduce the system cost and enhance the imaging speed, but it would also decrease the number of pixels available for single imaging on an industrial camera, hence rendering a certain impact on the transverse imaging range of the system.

Each phase shifting approach varies in phase shifting speed and quality, economic cost, system complexity, and imaging field. Several phase shifting implementation schemes are listed in Table 2, which can serve as the reference for selecting phase-shifting approach during system design. The number of asterisks in Table 1 represents the degree of excellence regarding the performance, and more asterisks indicate a better performance.

Table 2

Features of different FF-OCT structures.

3.2.

Enhancing Imaging Quality

To enhance imaging quality, the following aspects can be considered: reducing external noise to improve the accuracy and sensitivity of signal acquisition, reducing aberration to improve resolution, and reducing image noise to increase contrast. The FF-OCT image quality can be enhanced by either hardware or by algorithm and software.

Phase shifting and sample dithering during FF-OCT imaging may render motion artifacts and confuse image signals and noise, thereby impairing signal acquisition accuracy.^53,54 The instantaneous phase shift without mechanical displacement mentioned in Sec. 3.1 is an approach to enhance the speed and accuracy of phase shifting as well as to reduce motion artifacts.⁵⁵ And methods with two frames from the two outputs of the interferometer, en face tomographic images are calculated as a combination of two phase-opposed interferometric images acquired simultaneously by two CCD cameras placed at both outputs of the interferometer, with a spatial resolution of $0.8\mu \mathrm{m}\times 1.6\mu \mathrm{m}$ (axial × transverse). This new FF-OCT system, which uses instantaneous phase shifting with nonpolarizing optics and pulsed illumination, can suppress most of the artifacts that arise due to sample motion.⁵³ Motion artifacts can also be reduced via image processing; as shown in Fig. 5(a). The method based on singular value decomposition (SVD) can effectively filter the motion artifacts in collagen fibers, rendering a clearer display of cells.^56,60,61 When imaging biological tissue, an interface would occur between cover glass and tissues. If the refractive indexes of the two materials do not match, a strong “specular reflection” would occur, which can severely interfere with the recognition of backscattered signals of biological tissue. The dark-field FF-OCT can block specular reflections by the help of an opaque disk in the pupil-conjugated plane; the reference mirror is replaced by a blazed grating, which eliminates a walk-off between the sample and the reference beams on a camera that otherwise limits the imaging FOV, which suppresses specular reflection by at least two orders of magnitude.^49,62 Therefore, high-contrast tomography can be captured at the cover glass–tissue interface. The results of dark-field detection are shown in Fig. 5(b), exhibiting a significantly increased contrast.

Fig. 5

Image enhancement methods for FF-OCT. (a) The method based on SVD can effectively filter the motion artifacts in collagen fibers, rendering a clearer display of cells, pointed by the red arrow in the figure; (b) FF-OCT image and dark-field FF-OCT image recorded at the interface of cover glass and sheep brain tissue slice. Adapted from Refs. 49 and 56. (c) The process of deep learning network used to remove noise signals in FF-OCT images. Adapted from Ref. 57. (d) Comparison of FF-OCT images of a Ficus leaf before and after sample self-induced aberration was corrected when imaging at a depth of $75\mu \mathrm{m}$. Adapted from Ref. 58. (e) The difference in refractive index between the microscope objective lens and air causes the coherence plane and the focal plane to separate. Adapted from Ref. 59.

The coherence plane and the focus plane of the FF-OCT system are usually in the same plane. However, when imaging biological tissues with a dry objective, the difference between the refractive indexes of light in air ($n_{\mathrm{im}}$) and biological tissues ($n$) would separate the coherence plane and the focus plane, leading to defocus. As shown in Fig. 5(e), defocus would blur images reconstructed by FF-OCT. The differences in refractive index between the sample and the immersion medium of the microscope objectives may degrade the FFOCM image quality because of focus defect and optical dispersion mismatch. Dubois⁵⁹ studied the effects of focusing defects and dispersion mismatch in full-field optical coherence microscopy. Performing water immersion or oil immersion objectives can ensure that $n_{\mathrm{im}}$ and $n$ are close, for example, water immersion objectives can be used to mitigate defocus for imaging macrophages in Petri dishes.^19,41 The multiscale aberration structure of biological tissues would diminish the imaging resolution of different depths. The adaptive optics (AO) system and OCT can be combined to correct the inherent aberration of biological tissues, in which AO devices can compensate for the signal loss caused by tissue wavefront distortion as well as improve the image clarity. More structure information can be observed due to AO correction, as the image shown in Fig. 5(d).^58,63–65

Inherent speckle noise exists in FF-OCT, which would generate a grainy pattern and impair image quality. When it cannot be eliminated by hardware, certain image-processing algorithms can be adopted to reduce speckle noise, such as Gaussian filtering, median filtering, deep learning denoising, and other algorithms.^66,67 In recent years, deep learning algorithms have achieved satisfying results in the noise reduction of FF-OCT images and paved a new way to improve FF-OCT image quality. The fundamental approach for noise elimination based on deep learning is to distinguish noise from signals by combining spatial mixing and noise spectrum prediction technology.⁵⁷ As shown in Fig. 5(c), a neural network of two-phase image-spectrum conversion is adopted to reduce the noise of the FF-OCT image. By averaging adjacent B-scan images, speckles were reduced and the denoise database was obtained. A denoising convolution neural network based on the spatial-compounding (SC-DnCNN) was trained for noise recognition in the denoise database. After training, SC-DnCNN is used to eliminate the speckle noise of FF-OCT and output clean images.

3.3.

Improving Imaging Throughput (FOV and Depth)

Large imaging range and high spatial resolution are the two goals expected to be simultaneously achieved for biomedical imaging, but mutual restraint exists between these two goals. To expand the imaging FOV of the system on the premise of ensuring a high spatial resolution of FF-OCT, a scanning mechanism can be added to the detection unit of the system for transverse stitching of tomograms. Alternatively, a multiconjugate AO structure can also be used to obtain a wider imaging FOV.^{63,65,68–70} The disadvantages of these approaches are as follows: transverse stitching of multiple elongates the imaging time, and the accuracy of image stitching is easily influenced by motion artifacts. The multiconjugate AO device increases the system complexity and shrinks the actual effective imaging field of a single AO device. Currently, coherent shaping is adopted to expand the effective imaging FOV of the FF-OCT system: the eye serving as the objectives breaks the symmetry of the Linnik interferometer in the FF-OCT system during retinal imaging, forming a curved coherent gate. Therefore, the effective imaging FOV of the FF-OCT system becomes smaller.^71–73 The reference arm is equipped with an electric displacement platform to adjust the position of the coherent gate, and the sample arm is equipped with an optical window element to correct the geometry of the coherent gate. As shown in Fig. 6(a), the geometry of the coherent gate matches the curvature of the imaging sample, thereby expanding the effective imaging FOV. In retinal imaging, the effective imaging FOV after coherent gate shaping is three to four times larger than that before shaping [Fig. 6(b)].

Fig. 6

(a) Schematic showing different degrees of curvature for the coherence gate and their consequence on the FOV of the FF-OCT image, the left picture shows the coherence gate curvature does not perfectly match the retinal curvature, the middle picture shows the coherence gate is reshaped, although the curvature does not completely match the retinal curvature, a full FOV is still generated, as the coherence gate has a given thickness defined by the axial resolution. Adapted from Ref. 72. (b) The retina FF-OCT images before and after correcting for the coherence gate curvature. Adapted from Ref. 72. (c) FF-OCT imaging system based on coherent reflection matrix. Adapted from Ref. 74. (d) Samples with high scattering coefficient to verify deepening effect of imaging depth. Adapted from Ref. 74.

Owing to the existence of aberration and multiple scattering, the imaging depth of FF-OCT for biological tissues is mostly limited to $\sim 1\mathrm{mm}$, whereas the optimal imaging depth is limited to a few hundred microns. There are currently two approaches to increase the imaging depth of the FF-OCT in the scattering medium as much as possible. One approach utilizes optical clearing or other hardware, such as a light source with appropriate central wavelength based on the scattering properties of the sample or optical devices to reduce the imaging aberration. Another approach adopts a computational matrix, such as ballistic wave identification combining matrix calculation and iterative time reversal, which theoretically and experimentally confirmed that the imaging depth can be increased by at least two times.⁷⁵ Chen’s group at University of California, Irvine (UCI) proposed a calculation method based on coherent reflection matrix measurement, which confirmed that the imaging depth can be extended. The reflection matrix is obtained by point-by-point scanning and phase shift, then SVD is conducted to retrieve the single backscattered light.⁷⁴ Combining the reflection matrix measurement and wide-field heterodyne detection, satisfactory tomography of the microbeads through highly scattering medium is obtained with a doubled imaging depth; the system principle and imaging results are shown in Figs. 6(c) and 6(d).

3.4.

Functional Imaging Expansion

Like OCT with spot scanning, FF-OCT can also develop multimodal composite imaging, which can be combined with one or more technologies such as dynamic imaging,³⁰ fluorescence microscopic imaging,^24,25 elastography,⁷⁶ polarization-sensitive,⁷⁷ and photothermal imaging⁷⁸ and hyperspectral imaging.⁷⁹ Therefore, multimodal images of one sample can be collected at the same position, which can not only provide structure information with subcellular-level resolution but also render data regarding tissue function.^23,25,31

As an extension of FF-OCT, dynamic functional imaging detects intracellular motility by analyzing the signal fluctuation. It provides a method of label-free cell imaging, which is a mostly a nondestructive detection method for tissue or cell dynamics compared with FM. As shown in Fig. 7(a) shows the signal acquisition process of dynamic FF-OCT, where the camera collects the interference images of the sample in temporal sequence and obtains a 3D data matrix. Temporal signal on each pixel is collected, and it is converted from the time domain to the frequency domain for power spectrum analysis. Dynamic FF-OCT uses cell motility as the endogenous contrast agent and exploits metabolically driven dynamic scattering changes of cellular structures to enhance contrast.³⁰ Dynamic FF-OCT images serve as a useful supplement to static FF-OCT images, as static FF-OCT has high imaging contrast for static tissues such as fibers, blood vessels, and collagen, while dynamic FF-OCT is sensitive to cell membrane floating and cell surface reconstruction. Therefore, the superposition of static and dynamic FF-OCT images can enhance the overall visibility of tissue morphology. Due to the motion discrepancy between cytoplasm and connective tissue, dynamic FF-OCT images can clearly display the cell boundaries and demonstrate the hidden cells in static FF-OCT with high contrast. As shown in Fig. 7(b), the dynamic image of liver tissue exhibited clearer cell structures and boundaries than those in static images. The superposition of dynamic and static FF-OCT images of mouse brain tissue are shown in Fig. 7(c), which has significantly enriched the information capacity and contrast of the images.³⁰

Fig. 7

(a) Signal acquisition process of dynamic FF-OCT. Adapted from Ref. 27. (b) Static and dynamic FF-OCT images and partial enlarged images of rat liver. Adapted from Ref. 30. (c) Static/dynamic composite FF-OCT images of mouse brain cortex. Adapted from Ref. 30. (d) Dynamic images are computed in the HSV color space; hue is computed with the mean frequency, from blue (low temporal frequencies) to red (high temporal frequencies); saturation is computed as the inverse of the frequency bandwidth. The value is computed as the running STD. Adapted from Ref. 27. (e) Computation of the mean frequency (hue); frequency bandwidth (saturation); dynamic (value) grayscale image; dynamic FF-OCT of retinal organoid in the HSV color space. Adapted from Ref. 27.

Dynamic FF-OCT imaging is calculated based on the standard deviation (STD) to obtain the interference signal fluctuation over time.³⁰ The time-lapse sequence signal of the intensity $I(x,y,t_i)$ is acquired and processed using Eq. (13):

To further enhance the display of dynamic images, the signal obtained by the STD method is analyzed by power spectrum to generate “frequency-power spectrum” data. The analysis results of STD power spectrum are displayed in different color spaces, reflecting cell motility. For instance, the cell dynamics can be divided into three normalized frequency bands by RGB color space, in which R, G, and B represent low, medium, and high frequency cell movements, respectively.^80,81 Alternatively, the hue–saturation–value (HSV) color space can also be used, and the image brightness reflects the amplitude of cell movement while the image color is related to the cell motion speed.^27,29 The HSV color display approach and the results of dynamic FF-OCT imaging of retinal organoids are demonstrated in Figs. 7(d) and 7(e). Colored dynamic images can demonstrate the overall activity distribution and local abnormal activity distribution of the sample.

In addition, dynamic imaging signals can also be analyzed using autocorrelation (AUTO),⁸² logarithmic intensity variance (LIV),^82–85 and other methods. The specific calculation principles are provided in Eq. (14) and Eq. (15):

Fig. 8

Cancer organoid imaging using different dynamic OCT method. (a) SD-OCT image of colorectal cancer organoids which are cultured in Matrigel. (b) Dynamic OCT image with LIV method. (c) Dynamic OCT image with STD method. (d) Dynamic OCT image with AUTO method. (d) Dynamic images in the HSV color space with STD method (256 fps SD-OCT images is used for calculation), where the orange-red background in the picture is due to the rapid movement of the inherent speckle noise of SD-OCT. (f) Dynamic images in the HSV color space with STD method with background noise removed using a mask. Scale bar: $125\mu \mathrm{m}$.

3.5.

Combination of FF-OCT Images and Deep Learning

Intelligent recognition of the important features or regions in FF-OCT images as well as the pathological or physiological information contained in images is a research focus in this field, which will bring revolutionary improvement to the clinical diagnostic application of FF-OCT.⁸⁶

In the latest research, deep learning has been exploited in the target recognition of FF-OCT images. For instance, U-net neural networks are used to identify red blood cells in skin microvessels from FF-OCT images with a subcellular-level resolution;⁸⁷ multidirectional CNN is applied for predicting the possible pixels of epidermis in 3D data to enhance the accuracy of distinguishing dermal–epidermal junction (DEJ);⁸⁸ the OCT image data of enhanced capillary network generated by automatic simulation algorithm are combined with the U-net neural network to achieve automatic segmentation of skin capillaries.⁸⁹ The recognition of specific targets, particularly the recognition of structural features related to pathology, is crucial for the early diagnosis of severe diseases (e.g., cancer) and the in-depth understanding of tissue physiology and pathology. The combination of FF-OCT and deep learning can also enhance the accuracy of tissue segmentation [Fig. 9(c)]. A Pruned-ResNet-18 classifier based on the CNN network can be combined with FF-OCT images with subcellular-level resolution to detect skin squamous cell carcinoma, which has an accuracy of 80% in the early detection of this disease; the recognition results are shown in Fig. 9(f).⁹⁰ Moreover, the automatic classification algorithm based on a support vector machine is exploited to distinguish the adipose tissue in the hollow structure of breast tissue; combined with the texture features from ultrahigh-resolution FF-OCT image, the invasive ductal carcinoma (IDC) tissue, and normal fibrous matrix can be further distinguished.⁹¹

Fig. 9

FF-OCT image processing using deep-learning techniques. (a) FF-OCT image of the DEJ of human skin. Adapted from Ref. 88. (b) Hematoxylin–eosin staining (H&E) image of DEJ of human skin. Adapted from Ref. 88. (c) Deep learning algorithm performs DEJ detection on FF-OCT images of human skin, where the red line is the DEJ manually marked by the doctor, and the green one is the automatic recognition result by the deep learning algorithm. Adapted from Ref. 88. (d) FF-OCT image of mouse squamous cell carcinoma skin, where the red region is the stratum corneum (SC), the purple region is the epidermis (EPI), and the green region is the nucleus (NUC). Adapted from Ref. 90. (e) H&E image of mouse squamous cell carcinoma skin; adapted from Ref. 90. (f) The result of deep learning tissue classification algorithm for squamous cell carcinoma detection in FFOCT images of mouse skin, where the yellow circles represent the identified cancerous cells. Adapted from Ref. 90.

The combination of high-resolution FF-OCT images and deep learning has especially revolutionized histopathological diagnosis, but it still has the shortcoming of manual marking, which is time- and labor-intensive, and the marking accuracy is influenced by precise identification of tissue characteristic margins. The colored images of cells and tissues with high resolution and contrast from dynamic FF-OCT can directly serve as characteristic marking images to train the deep learning network. Currently, based on feather engineering and the automatic diagnosis algorithm of CNN, some researchers performed the auxiliary diagnosis of cancer with an accuracy of nearly 100% using the high-resolution colored image of D-FFOCT at both small scale ($2{\mathrm{mm}}^2$) and macroscale ($2{\mathrm{cm}}^2$). This has verified the huge potential of combining D-FFOCT and deep-learning for fast, automatic, and accurate histopathological diagnosis.⁹²

4.1.

Biomedical Imaging Applications

FF-OCT can perform feature recognition from the microstructural level to cell level in ophthalmology, dermatology, oncology, and other fields. OCT has been most widely used in ophthalmology; compared with currently commercialized FD-OCT, FF-OCT can characterize cells and tissues as well as present pathological structure details.^{31,54,65,71,72,96–100} In the early stage of FF-OCT development, high-quality, high-resolution tomographic images can be obtained in some fields. For example, for retinal imaging, as early as 2004, some researchers used FF-OCT detected high-resolution retinal images and achieved the effect of distinguishing different structures.¹⁵ In addition, in other early applications, it also included tissue engineering scaffolds, etc.¹⁰¹ Even though high-quality imaging cannot be achieved due to hardware limits, FF-OCT opened a door for unique imaging performance, which made an important contribution to the development and application of FF-OCT. In recent years, imaging quality of FF-OCT has observed a significant improvement, Fig. 10(a) depicts the nerve fiber layer, ganglion cell layer (GCL), inner plexiform layer, outer plexiform layer (OPL), and outer nuclear layer (ONL) of macaque retina. Besides, due to its subcellular resolution, FF-OCT can detect the cellular characteristics of each layer within the retina such as the details of ganglion cell axons and cell bodies (yellow box) demonstrated in Figs. 10(a) and 10(b). Wide-field images are important for the early diagnosis of retinal diseases in clinical settings. Therefore, FF-OCT with a large imaging FOV, high resolution, and fast imaging speed has a prospective future in the field of ophthalmic detection. As shown in Fig. 10(c), the AO technology was adopted to obtain the retinal image of $5\mathrm{deg}\times 5\mathrm{deg}$ FOV, then five images were stitched into a $12\mathrm{deg}\times 12\mathrm{deg}$ FOV via a certain stitching method, in which the color map represents the calculated photoreceptor density. Both coherent gate shaping and AO technology can be used to expand the imaging field of FF-OCT during retinal detection. Moreover, optical phase modulation via movement of the eye itself to retrieve tomographic images of TD-FFOCT has also been achieved. The combination of this method with coherent gate shaping technology has the potential to facilitate the clinical application of FF-OCT in the ophthalmology field.⁵⁴

Fig. 10

(a) Merged static (red) and dynamic (green) FF-OCT images on the macaque retina to identify stationary (fiber, blood vessel, collagen) and dynamic (cell) structures. Adapted from Ref. 28. (b) The FF-OCT image shows the details of the cell axons and cell bodies of the human retinal GCL, which can be used to check the details of the damage. Adapted from Ref. 96. (c) Combine AO and stitching methods to get the FF-OCT image of retina with a larger FOV. Adapted from Ref. 71.

Combining FF-OCT with other imaging technologies to perform multimodal image acquisition at the same position of the sample can significantly enrich imaging contrast and provide functional information besides structure. Therefore, FF-OCT performance in medical diagnosis can be greatly enhanced, exhibiting clinical potential in early cancer diagnosis and serving as guidance in tumor resection. As shown in Figs. 11(a)–11(d), the reported applications include, but are not limited to, the discrimination of benign and malignant tissues of skin cancer,^35,103,105 breast invasive ductal cancer,^102,106 head and neck cancer,¹⁰⁴ as well as optical biopsy of intestinal tissues,^107,108 detection of esophageal mucosa, liver, and gastric mucosa.^{86,102,106,109–113} Besides, FF-OCT incorporating artificial intelligence can benefit deep mining of the abundant image data, which can reduce the difficulty of human eye recognition, improve the accuracy of information recognition, and promote the efficiency of FF-OCT histological diagnosis.

Fig. 11

(a) D-FFOCT image of esophageal mucosa at a $10\mu \mathrm{m}$ depth, where the epithelial cell cytoplasm (white arrows) and nuclei (blue arrows), pericellular lining (blue arrowheads), lamina propria papillae (white arrowheads). Adapted from Ref. 109. (b) D-FFOCT image of invasive ductal carcinoma of breast tissue, in which nested cells with malignant features have local low intensity. Adapted from Ref. 102. (c) FF-OCT image of mouse skin squamous cell carcinoma where the cancerous tissue in the black box. Adapted from Ref. 103. (d) FF-OCT image of head and neck cancer, where the yellow dotted arrow points to the stromal reaction, the yellow cross mark points to the cancerous tissue, and the yellow asterisks points to the keratin pearls. Adapted from Ref. 104. (e) In the first line, D-FFOCT monitors the dynamic profile (the boundary between the two cells) changes during cell differentiation; the second line, the very active area in the organoid, is composed of cells that exhibit high dynamics, and may be undergoing apoptosis. Adapted from Ref. 27. (f) D-FFOCT monitors the activity and morphology of Hela cells over time. Adapted from Ref. 32.

FF-OCT can provide high-resolution structural and functional information of 2D or 3D cell cultures and 3D-printed cell scaffolds under a normal culture environment.^15,34,114 Researchers have used dynamic FF-OCT to conduct long-term longitudinal comparative studies on organoid samples, including long-term observation of cell apoptosis, proliferation, or migration as well as observation of the dynamic contour changes of tissues during organoid cell differentiation [Fig. 11(e)]. Moreover, the dynamic and static dual-mode of FF-OCT detection has also been exploited to monitor the changes in cell viability in cancer cells (e.g., Hela cells) and complex culture systems [Fig. 11(f)]. The ability of FF-OCT to track the growth and development of cells in vitro can monitor the cell viability change in disease modeling or artificial tissue transplantation as well as reveal the mechanism of structural and functional reconstruction of 3D cultured tissues such as organoids and tumor microspheres, thereby facilitating research on the mechanism of human development and disease.^{15,26,27,32,34,114} The functional imaging extension of FF-OCT is also applied in spectroscopic FF-OCT, with different pseudocolors showing melanocytes in ex vivo Xenopus laevis tadpole embryos at high resolution.^115,116 In addition, it also includes superimposing the fluorescence microscopic image of the mouse colon and the FF-OCT image, so more information can be displayed in one image;¹¹⁷ the images obtained at the same time at the center wavelength of 800 and 1200 nm are given different colors, respectively. Superimposed multimodal imaging was applied to detect rabbit trachea, which can increase the contrast of different tissue structures, significant improvement in imaging penetration depth was demonstrated at 1200 nm compared to 800 nm, and a color image representation was applied to fuse both images into a single image for better comparison of wavelength effects.¹¹⁸

4.2.

Expansion of FF-OCT Applications

Benefiting from the enhancement of the overall performance of FF-OCT system, its potential in scientific research and application of materials science, security, and identification has gradually been valued by researchers. FF-OCT can nondestructively acquire 2D tomography of a certain depth with high axial and transverse resolution.^119–121 Therefore, this technology can detect the specific spatial location and detailed characteristics of material damage to analyze the degree and cause of the damage. For instance, FF-OCT can be used to assess the subsurface (dozens of microns below the surface) damage of optical glass [Fig. 12(a)] and observe the internal and external damage of hair [Fig. 12(b)].^122,123

Fig. 12

(a) Subsurface damage of optical glass using FF-OCT. Adapted from Ref. 122. (b) Human hair damage detection. Adapted from Ref. 123. (c) FF-OCT cross-sectional views of different depths of banknotes, used for authenticity identification. Adapted from Ref. 124. (d) FF-OCT fingerprint patterns at different depths, from top to bottom are external fingerprints ($0\mu \mathrm{m}$), internal fingerprints ($300\mu \mathrm{m}$), and sweat pores ($200\mu \mathrm{m}$). Adapted from Ref. 125.

The unique internal characteristics of security objects can be adopted as new indicators of anticounterfeiting measures. Noninvasive detection of the internal details of banknotes by FF-OCT can be used to identify their authenticity [Fig. 12(c)].¹²⁴ The internal fingerprint detection can function as identity recognition or a password [Fig. 12(d)], which is more secure than the surface fingerprint. Modern forgery is highly specialized, and the appearance of imitations may be mistaken as genuine.^124–127 Therefore, FF-OCT is attracting increased attention as an effective tool for authenticity identification of certificates, paintings, and porcelain, exhibiting a broad range of application prospects in anticounterfeiting science.