2.1.

Compact Optical Coherence Tomography Spectrometer

The spectrometer in a conventional SD-OCT system is one of the most critical components, where light is dispersed onto a line scan camera to spectrally separate and measure the interferometric signal. Commercially available spectrometers demonstrate excellent performance; however, such spectrometers are often bulky and expensive, limiting their use in low-cost applications. With this regard, we developed an economical and compact spectrometer solution for SD-OCT by modifying a commercially available device (I-MON 835 OEM, Ibsen Photonics, Denmark). This spectrometer had a spectral bandwidth of 50 nm (810–860 nm) with high spectral resolution (0.08 nm). The OEM unit had dimensions of $23\text{-}\mathrm{mm}\text{height}\times 58\text{-}\mathrm{mm}\text{width}\times 76\text{-}\mathrm{mm}\text{depth}$. However, the included detector (ILX511B, 2048 pixels, Sony) had a low line-rate of only 100 Hz, hindering its application in SD-OCT as the low acquisition rate would result in motion artifacts and fringe washout during in vivo imaging. To address this, the detector was replaced with a more suitable USB 2.0 charge-coupled device (CCD) line-scan camera (TCE-1024-U, $14\text{-}\mu \mathrm{m}$ pixel size, 1024 pixels, Mightex). This camera can operate in either an 8- or 12-bit mode and provides a line rate of either 25 kHz or 10 kHz. For this study, this camera was operated at a 10-kHz acquisition rate and in 12-bit mode.

A schematic of the modified spectrometer is shown in Fig. 1(a). The light from the grating had a beam spread of 12 mm and was matched closely to the detector array size (14.3 mm). Figure 1(b) shows the measured source spectrum versus the spectrometer output, confirming that the entire source spectrum was detected by the spectrometer sensor. The final dimensions of the modified spectrometer were approximately $90\text{-}\mathrm{mm}\text{height}\times 90\text{-}\mathrm{mm}\text{width}\times 70\text{-}\mathrm{mm}\text{depth}$.

Fig. 1

Compact, high-resolution spectrometer for OCT applications. (a) Modified compact spectrometer with OEM unit (Ibsen Photonics) and CCD camera (Mightex), (b) measured spectrometer output versus source spectrum, and (c) photograph of modified spectrometer. A US quarter is shown for scale. G, grating; C, collimator; and M, concave mirror.

2.2.

Briefcase Optical Coherence Tomography System

We developed and designed this OCT system to fit into a standard briefcase ($34\text{-}\mathrm{cm}\text{height}\times 46\text{-}\mathrm{cm}\text{width}\times 13\text{-}\mathrm{cm}\text{depth}$) to be compact, easily portable, and able to be used for point-of-care applications (Fig. 2). Additionally, this briefcase enclosure also protects all components during transportation. The SD-OCT system was based on a fiber optic Michelson interferometer configuration. Light from an uncooled superluminescent diode [Exalos, EBD270005-03, ${\lambda }_c=840\mathrm{nm}$, $\mathrm{\Delta }\lambda =50\text{-}\mathrm{nm}$ full-width-half-maximum (FWHM), and 3.4-mW output power] was guided into one port of a $2\times 2$ fiber coupler and split into the reference and sample arms.

Fig. 2

Compact briefcase OCT system. (a) Schematic of the system and (b) photographs depicting the exterior and interior views of the briefcase. The top panel shows the fully enclosed briefcase OCT system. A coffee mug is shown for visual representation of the briefcase dimensions. SLD, superluminescent diode; PC, polarization controller; C, collimator; FC, fiber coupler; FPC, fiber patch cable; DM, dichroic mirror; L, lens; RM, reference mirror; and CCD, charge-coupled device camera.

In the sample arm, a handheld probe was connected via a 2-m-long fiber optic cable. The handheld probe utilized a mirror and a focusing lens ($f=60\mathrm{mm}$) to focus the beam on a sample. A compact high-speed USB 3.0 CCD camera (XIMEA MQ013CG-ON, 1.3 MP, 500 FPS) was also integrated into the handheld probe for concurrent acquisition of surface video images and mosaicking in real time. Light for surface illumination was provided by a high-power light-emitting diode that was coupled to an integrated fiber bundle. Light exited the distal fiber ends that were located in the nose cone of the probe. For acquisition of cross-sectional images, samples were manually scanned using the probe, and images were constructed based on the decorrelation of sequentially acquired A-scans.³² The incident optical power was $\sim 1.3\mathrm{mW}$, which is well below the ANSI maximum permissible exposure (MPE) limit for skin.³³

In the reference arm, a variable neutral density filter was placed before the lens ($f=60\mathrm{mm}$) to optimize system sensitivity. The light reflected from both arms was recombined in the $2\times 2$ fiber coupler, and the resulting spectral interference signal was captured by the modified compact OCT spectrometer. Spectral data from the line scan CCD camera of the spectrometer were acquired by a laptop (HP OMEN, CE019DX, i7-7700HQ, 8 GB RAM) via a USB 2.0 connection, sampled, and rescaled as a function of the wavenumber to display A-scans in real time. During image acquisition, the line rate of the CCD camera was set to 10 kHz and the camera exposure time was set to $100\mu \mathrm{s}$. The complete cost of this briefcase OCT system, including the surface imaging module, was $\sim \mathrm{USD}\$8000$ (Table 1), and the entire system weighed around 9 kg (20 lbs.). The cost of parts for the system was based on off-the-shelf prices for a consumer, and even lower costs are possible by further integrating components and purchasing components in volume.

Table 1

Cost of briefcase OCT system.

Figure 3 shows photographs of the portable briefcase OCT system along with a high-end, cart-based low-coherence interferometry (LCI) system [Fig. 3(a)] previously developed by our group and used in several clinical studies.^17,18 For comparison, the dimensions of the cart-based LCI system were $107\text{-}\mathrm{cm}\text{height}\times 46\text{-}\mathrm{cm}\text{width}\times 66\text{-}\mathrm{cm}\text{depth}$ and cost $\sim \mathrm{USD}\$\mathrm{50,000}$. Figures 3(b) and 3(c) show the portability of the briefcase OCT system where a user/physician can easily carry the system to the primary care and/or remote setting. Photographs of human subjects and operators were obtained and used with their consent.

Fig. 3

Portability of the briefcase OCT system. (a) Comparison of the briefcase OCT system with a high-end, cart-based LCI system. (b–c) Photographs illustrating the portability of briefcase OCT where a physician can easily carry such system for examination in primary care and point-of-care settings.

2.3.

Real-Time Mosaicking and B-scan Formation

The algorithm used for OCT acquisition, image assembly, and surface image mosaicking is shown in the flowchart in Fig. 4. The processing was performed using a custom-designed graphical user interface (LabVIEW 2015) on a laptop. Initially, background spectra and resampling indices for linearization in the wave number were precalculated in the initialization step. Standard OCT processing was then performed by background subtraction, cubic B-spline interpolation for linearization, and dispersion compensation. This was followed by A-scan reconstruction by taking the Fourier transform of the recorded spectral signal. Reconstruction of the cross-sectional images was implemented with a manual scanning approach previously reported by our group.³² The use of this manual scanning method to generate cross-sectional images was implemented to minimize cost, to eliminate a lateral scanning mechanism, and to enable imaging over larger areas of tissues.

Fig. 4

Flowchart representation of the real-time OCT acquisition and surface image mosaicking. The algorithm performs OCT acquisition, B-scan formation based on the correlation coefficients, and mosaicking of surface images all in real time.

A-scan assembly was based on a defined cross-correlation threshold value set by the user, where the cross-correlation of two sequential A-scans ($i$ and $i+1$, where $i$ corresponds to the first A-scan) is computed. The A-scans are appended to the assembled images when the correlation coefficient falls below the threshold (implying that the new A-scan is from a new lateral location), and excluded from the assembled images when the correlation coefficient is greater than the threshold (implying that the new A-scan is from the same lateral location). The decorrelation of the A-scans effectively confirms displacement of the probe beam or a change in the sample structure, which otherwise would have a higher correlation coefficient. A threshold value for the correlation coefficient in this study was empirically set to 0.8 for image assembly.

The developed algorithm also simultaneously acquires surface images of the sample and performs mosaicking in real time. For efficient data flow, a producer/consumer architecture in LabVIEW was implemented where the producer loop acquires surface images and the consumer loop does the mosaicking of acquired images in real time. The consumer loop starts only when the correlation coefficient is below the threshold level and is notified using a notifier (Fig. 4). The notifier also synchronizes the acquired A-scans for each surface image in the mosaic. This allows for the determination of the lateral two-dimensional (2-D) position of the OCT beam, and subsequently for more standard B-scan images to be extracted from regions of straight paths identified from within the total path of the beam.

For real-time mosaicking, out-of-focus or blurred images were eliminated to form a high-quality mosaic. Blurry images could also result in an error in the registration algorithm. Our algorithm incorporated a command to eliminate out-of-focus images in real time by measuring the sharpness of the image. The sharpness value (or mean value, $\overline{x}$) of the image was measured by taking the mean of the convoluted image using a linear filter, which effectively highlighted the features of the image. For real-time processing, the threshold value was set to correspond to the sharpness value at a depth of 0 and 1.5 mm [Fig. 5(d)], and any images acquired outside of these depths were excluded.

Fig. 5

Characterization of the briefcase OCT system. (a) Measured system sensitivity roll-off using a fixed optical density filter in the sample arm. (b) Measured axial resolution at FWHM of the point spread function. (c) Lateral resolution over various depths measured with a beam profiler and compared with calculated values. (d) Representative surface images and the corresponding sharpness value used for discarding out-of-focus images in real-time.

The surface image registration algorithm determines the translational offset between two consecutively acquired frames. A fast registration algorithm was implemented as this approach requires a small fraction of computational time and memory requirement.^34,35 The fast registration algorithm performed a cross-correlation using the discrete Fourier transform between consecutive images. The registered images were subsequently inserted into a predefined zero matrix (also called a canvas), where the first image is copied to the center of the canvas and subsequent frames are copied with an offset value. The stitching of images was based on the dead-leaf approach,³⁶ where a new frame overwrites an existing frame. The canvas size was approximately five times larger (FOV: $15\mathrm{mm}\times 15\mathrm{mm}$) than a singly acquired frame, and the time required for image registration was $\sim 10\mathrm{ms}$. The program saved the surface image mosaics as 12-bit PNG images, and the assembled A-scans both as raw data and as a 12-bit PNG image. The real-time data procssing time of the consumer loop was $\sim 15\mathrm{ms}$. The measured loop time is system (laptop) dependent and mainly relies on the usage of the CPU during data acquisition.

3.1.

System Characterization

The sensitivity of the system was measured by placing a mirror and an optical density filter ($\mathrm{OD}=2.1$, round trip loss 42 dB) in the sample arm beam. Figure 5(a) shows the measured sensitivity of the system to a depth of 2.7 mm in air. The measured sensitivity at zero optical path difference (OPD) was $\sim 98.89\mathrm{dB}$, and dropped by 6 dB at a depth of 1 mm in air. For in vivo imaging (shown later), a depth range of 1.5 mm (85 dB) was used as the sensitivity was degraded over larger depths and was closer to the noise floor. Furthermore, the low optical power from the uncooled SLD that is incident on scattering samples will limit the sensitivity of the system, which will degrade with increased A-scan rate. The OCT images were acquired at a 10 kHz A-scan rate and the image quality was sufficient to differentiate the structural features. We measured the output power over several hours (at different intervals) and found minimal fluctuations (microwatts) in output power. Future studies will investigate the dependence of output power on ambient temperature fluctuations.

Figure 5(b) shows the measured and calculated axial resolutions of the system. The calculated axial resolution based on the SLD specifications from the vendor was around $6.2\mu \mathrm{m}$. The axial resolution was measured at the FWHM of the point spread function and was around $8.1\mu \mathrm{m}$ at zero OPD. The $2\text{-}\mu \mathrm{m}$ discrepancy in the axial resolution may be due to slight uncertainty in dispersion correction. The measured axial resolution deteriorates with larger imaging depths (over 2 mm). Nonetheless, for an imaging range up to 1.5 mm, the system had minimal degradation compared with the zero OPD. Figure 5(c) shows the measured and calculated lateral resolutions of the system. The calculated lateral resolution was based on the focal length of the objective ($f=60\mathrm{mm}$) and the measured beam diameter before the objective (3 mm) and was around $21.4\mu \mathrm{m}$. For imaging, we placed the focus roughly at $600\text{-}\mu \mathrm{m}$ depth. The Gaussian beam spot size (lateral resolution) was measured using a beam profiler (Thorlabs, BP209-VIS) as depicted.

The developed system will be used for imaging various tissue sites that are commonly examined in primary care clinical settings, such as the tympanic membrane of the ear. Considering this application, a longer working distance would be required, and the subsequently lower (poorer) lateral resolution would be an inherent trade-off. Further, in the designed handheld probe, the lens was fixed inside the 3-D printed case, which along with a speculum tip, comprised a total length of $\sim 55\mathrm{mm}$. Hence, a lens with a 60-mm focal length was used. The reported system is not completely optimized for tympanic membrane (TM) imaging; however, we are working to optimize this system, along with improving the lateral resolution of the probe. Figure 5(d) shows the representative surface images and sharpness values for the corresponding depths, and lateral resolutions are shown in Fig. 5(c). At focus, the image demonstrated a maximum sharpness value ($\overline{x}$) that decreases away from focus.

3.2.

Cross-Sectional Imaging Using a Motorized Translation Stage

To first test the feasibility of cross-sectional imaging using the briefcase OCT system, an image of layered tape (S-9783, 3M 810 Scotch Tape) was obtained by translating the sample using a motorized translation stage. During imaging, the probe was fixed on a mount and a sample was placed on a sample holder. Figure 6(a) shows a cross-sectional image of the layered tape wherein $>15$ layers are clearly visualized. The dimensions of the acquired B-mode image were $7\mathrm{mm}\times 1.5\mathrm{mm}$, with 512 axial points in depth along each A-scan, and with 1000 A-scans of data making up the cross-sectional B-mode image. The image of the layered tape shows that the current imaging resolution and depth are sufficient for imaging biological samples.

Fig. 6

Representative cross-sectional images obtained by translating the sample with a motorized stage. (a) Reconstructed OCT image of a layered roll of tape, (b) in vivo imaging of the human palm region, (c) fingernail plate and fold, and (d) fingertip. The scale bar represents $500\mu \mathrm{m}$.

We further imaged human skin sites including the palm region, nail fold, and fingertip. The dimensions of the acquired B-mode images were the same as for the layered tape. Human subject imaging was performed under a protocol approved by the Institutional Review Board at the University of Illinois at Urbana-Champaign. Figures 6(b)–6(d) show representative cross-sectional images of these sites. Our system was able to resolve intricate features in the palm and fingertip such as the stratum corneum, stratum spinosum, sweat ducts, and the dermis layer. The image of the nail fold revealed features such as the epidermis, dermis, and nail plate. The missing A-lines in Fig. 6(d) were due to the movement of the fingertip during acquisition ($\sim 10\text{-}\mathrm{mm}$ scan). Acquisition of the B-mode images using a motorized stage was done to compare image quality with cross-sectional images assembled using the manual scanning approach.

3.3.

Manual Scanning: Mosaicking and Cross-Sectional Imaging Approach

After implementing manual scanning and real-time mosaicking, we further tested the feasibility of this approach by once again imaging in vivo human skin. Figure 7 shows the OCT and surface image mosaic of skin from the human palm. The real-time surface mosaicking showed precise image registration, resolving the papillary ridges (fingerprint) of the skin in the palm region [Fig. 7(a)]. We also observed the misregistration of the mosaic in Fig. 7(a) at the angled surface of the palm. This could be due to the defocused image limiting the accuracy of the registration algorithm. In the present study, we had implemented a sharpness-based metric to eliminate out-of-focus images and saved real-time mosaics as they were formed.

Fig. 7

Cross-sectional beam path image and surface mosaic of the in vivo human palm. (a) Surface mosaic of the palm region with the corresponding OCT beam position shown in (b). (c) Cross-sectional image of the palm constructed from the corresponding OCT beam path. (d) Representative B-scans are formed by extracting adjacent A-scans from a straight path in the manual scan shown by the solid and dashed box outlines. The scale bar represents $500\mu \mathrm{m}$.

Figure 7(b) shows the plot of the OCT beam position during the real-time scan. This position plot was based on the measured offset value of the image registration corresponding to the acquired A-scan position. The cross-sectional beam path image generated for the integrated path of the OCT beam position is shown in Fig. 7(c). The represented image shows well-differentiated structures such as the stratum corneum, sweat ducts, and the dermis. This image has a lateral dimension of over 10 mm and was obtained from both dimensions ($X$ and $Y$) in the beam path position. This system allows a user to manually and arbitrarily scan larger FOVs. Furthermore, the recorded path of the beam position enables the construction of a thickness map that could be overlaid on the surface image mosaic. Figure 7(d) shows representative B-scans obtained from the same dataset at two different straight paths highlighted in Fig. 7(b). The B-scans obtained from this manual scanning approach were comparable with the B-scans obtained using the motorized stage. The results from imaging the palm show that our system can clearly resolve microstructural features in scattering samples.

We further imaged the human fingertip and forearm regions. The OCT and surface image mosaics from both tissue sites resolved optimal surface features and depth-resolved structures as shown in Figs. 8(a) and 8(b). Cross-sectional images and the corresponding surface mosaic of a US dollar bill are shown in Fig. 8(c). A dollar bill was imaged to highlight the accuracy of the image registration and the real-time mosaicking algorithm as fine features can be resolved in the surface image mosaic. The cross-sectional image of the dollar bill, however, lacked structural features, and largely only a surface reflection could be visualized by OCT. From these results, it can be concluded that the briefcase OCT system with manual scanning and real-time mosaicking can accurately and simultaneously construct surface and depth-resolved cross-sectional images in real time.

Fig. 8

Cross-sectional beam path images and surface mosaics of human skin tissue sites and a dollar bill. Surface mosaics, OCT beam position plots, and corresponding cross-sectional beam path and B-scan images are shown for (a) human fingertip, (b) forearm skin, and (c) US dollar bill. The representative OCT B-scan images are extracted from the straight path regions as shown by the solid box outlines. The scale bars represent $500\mu \mathrm{m}$.