Optical coherence tomography (OCT) is a noninvasive imaging modality capable of providing high-resolution cross-section images of tissue at a high speed. These characteristics make OCT an ideal candidate for an intraoperative imaging modality. One of the most challenging microsurgical procedures that could benefit from using OCT is vitreoretinal surgery [1]. In conventional vitreoretinal surgery, the intraocular space is illuminated by a handheld fiber-optic light pipe and a surgeon has to manipulate small instruments within the confined intraocular space under microscopic view. By integrating an intraoperative OCT into a surgical microscope system, the retinal structure can be visualized, which could help guide surgical tools precisely and thereby prevent damage to delicate retinal tissues [2, 3]. However, the high cost of the intraoperative OCT system limits its availability in ophthalmic surgery [4–6] and very few operating rooms in the largest hospitals have access to it.

There have been efforts to lower the overall cost of the OCT system by either using a silicon integrated circuit [7] or inexpensive off-the-shelf components and cheaper micro electro mechanical systems (MEMS) scanner [8, 9]. However, to make it suitable for intraoperative use, these standard low-cost OCT systems need further development and adaptation that requires additional costs. On the other hand, common-path (CP) OCT can be easily built into a free-hand scanning needle imager and also integrated into microsurgical tools [10, 11] to be used for various ophthalmic procedures. It can further reduce the overall cost by eliminating the reference arm and the optical scanner. By implementing a free-hand scanning algorithm [12], B-scan images can be achieved without a high-speed galvo mirror in the system, which is another major cost reduction for the whole system. Swept-source-based CP-OCT (CP-SSOCT) systems have been developed for retinal imaging and integrated with vitreoretinal surgical tools for ophthalmic surgical guidance [13–15]. Nevertheless, the high cost of swept-source OCT engines is one of the main barriers to lowering the overall cost of the system.

In this paper, we describe the design and implementation of a low-cost, portable, spectral domain CP-OCT system using a free-hand scanning lensed fiber imaging probe designed for intraoperative use. The lensed fiber probe will be inserted into the eye through the trocar, which has already been used to gain access to the intraocular space. Our device is designed to be used to scan over a small area to quickly check the retinal tissue or identify the injection site. Compared to the standard 2D/3D scanning method, our design provides an immediate seamless imaging solution during retinal surgeries. Since free-hand scanning OCT does not require high-speed scanning, by utilizing off-the-shelf optical components and a low-frequency complementary metal–oxide–semiconductor (CMOS) sensor, the overall cost could be significantly lower than conventional OCT systems. The lensed fiber probe was fabricated in house using high index epoxy and a polyimide tube. The axial resolution and signal roll-off were analyzed. Then, the assembled system was tested using a tape phantom and demonstrated imaging capabilities on the ex-vivo porcine eyes. The results showed that the system could be used for intraoperative retinal imaging in clinical and laboratory studies.

2.1. Design of Common-path Spectral-domain OCT Engine

Here, we designed the low-cost common-path spectral-domain OCT (CP-SDOCT) engine as shown in Fig. 1 using a superluminescent diode (SLED) (EXS210031-02; Exalos, Schlieren, Switzerland) as the light source, which has a center wavelength of 840 nm and a 3-dB bandwidth of 34.2 nm. The maximum output power at the end of the imaging probe is 2.8 mW and is adjustable. A 50:50 fiber coupler (TW850R5A2; Thorlabs, NJ, USA) is used instead of a broadband circulator to form a low coherence interferometer and reduce the build cost. In the CP-OCT design, a simple needle-imaging probe composed of a single-mode fiber terminated with an epoxy lens serves as both the reference and sample arm. The reference signal comes from the fixed epoxy-glass interface at the end of the probe and the same fiber is used to collect the sample signal. Based on the refractive index of the fiber core and the epoxy, the calculated reference beam power is 0.6% of the total power based on Fresnel reflection.

Figure 1. Schematic of the spectrometer and the common-path optical coherence tomography (CP-OCT) system.

The core component of the CP-SDOCT system is the customized spectrometer as in Fig. 1. The collected and returned reference and sample light first comes out of the cleaved fiber end and diverges. After the light travels 33 mm, a parabolic mirror (MPD01M9; Thorlabs) is used to collimate the beam and suppress the spherical aberration introduced by the grating. The collimated beam is diffracted by an 1,800 lines/mm transmission grating (HD1800/840; Wasatch, UT, USA) and then focused by a doublet with a focal length of 150 mm (AC254-150-B; Thorlabs) onto a linear CMOS sensor (S11639-01; Hamamatsu Photonics, Shizuoka, Japan). The sensor has a total of 2,048 pixels, and each pixel is 14 µm by 200 µm (width by height). The tall pixel array makes the system robust to misalignment and allows a bigger tolerance. To optimize the spot size of the signal beam on each pixel, the spectrometer was designed using OpticStudio (Zemax, WA, USA) simulation software. The distance between each component is optimized, as well as the angle between the sensor and the incoming beam. The achieved root mean square (RMS) radius of the beam is 6.987 µm at the center wavelength, and on the edge, the radii of the two beams are 8.82 µm and 9.50 µm at 880 nm and 820 nm, respectively. This is achieved by setting the CMOS sensor slightly at the defocusing point, which is 148.7 mm to the doublet, to make the edge beam size smaller. However, the edge beam size is slightly larger than the pixel size due to only one focusing lens being used. The spatial separation between the wavelength from 820 nm to 880 nm is 28.7 mm, which covers the whole sensor plane. With this setup, the maximum sensing depth was calculated to be 4.5 mm in the air. We used a mini-PC via a USB 2.0 interface to transfer and process the detected signal to achieve a maximum A-scan rate of 4 kHz.

2.2. Lensed Fiber Imaging Probe

There are two main drawbacks to using a bare fiber as an imaging probe for CP-OCT. Firstly, without any focusing optical component, the laser beam diverges, resulting in decreasing the penetration depth. Also, while the beam propagates in the sample, it diverges, causing the signal to decrease. Secondly, the reflected reference beam power depends on the refractive index of the working environment, which will affect system sensitivity. Thus, we developed a disposable lensed fiber probe as shown in Fig. 2(a) to provide a constant reference plane while working inside the ocular space.

Figure 2. Design and simulation of the lensed fiber probe. (a) Schematic of the lensed fiber probe, (b) Zemax simulation of the lensed fiber probe.

The fiber sensing probe is designed with an epoxy extension rod terminated with an epoxy semi-spherical ball lens, and the detailed design has been discussed in [16]. In this work, we propose a more robust and simplified fabrication process. To fabricate the probe, the single-mode fiber is placed into a medical polyimide tube (061-I; MicroLumen,). The tube is used to hold the epoxy adhesive for the extension area and the diameter of the tube determines the radius of the fabricated epoxy lens. Polyimide provides good protection for the fiber probe and allows effective ultraviolet (UV) curing of the epoxy inside the tube. The fabricated lens can also be put into a metal tube for additional protection. The design has been verified and optimized using Zemax as shown in Fig. 2(b). Here, the tube we used has an outer diameter of 200 µm and the epoxy has a refractive index of 1.70 (NOA170; Norland, NJ, USA). The length of the extension area is set to 450 µm and the simulated focal length of the probe is 1.85 mm inside water. The RMS radius of the beam at focus is 24.25 µm. Here, we picked the refractive index of the epoxy to be 1.7 and the index of the water as 1.33 to mimic the intraocular environment. Based on the beam size, the lateral resolution at the focal point will be 28.55 µm. Combined with the CMOS sensor, which has a maximum data acquisition rate of 4,000 lines/s, the maximum free-hand scanning speed our system can achieve is 48.5 mm/s.

We developed a simple method for fabrication of the lens and the assembled probes as shown in Fig. 3(a). The probe handle is 3D-printed and the fiber probe is fixed inside a 25-gauge metal tube. To fabricate the extension rod and ball lens, the coating of the fiber was removed and put inside the medical polyimide tubing. The inner diameter of the tube we used is 150 µm, which is slightly larger than the fiber size (125 µm). The fiber and tube were held in place by a precise translation stage and a 3D-printed holder. The length of the extension rod is determined by the offset between the fiber and the end facet of the tube. We controlled the axial and transverse positions of the fibers using x-y-z linear stages while monitoring them using digital microscopes. To set the offset, a tape phantom is placed right in front the tube, and the A-scan image is taken from the same fiber to ensure the offset is set at the predetermined value [Fig. 3(b)]. Then the high-refractive index UV curable glue was drawn into the tube by the capillary effect and the glue was cured using an ultraviolet light source to form a rigid rod [Fig. 3(c)]. After this step, the polyimide tubing and the fiber will be glued together, and the tubing will stay on the rod for additional protection. The semi-spherical lens was fabricated by applying epoxy to the extension rod end, which naturally forms a semi-spherical shaped lens due to surface tension. The amount of glue is controlled by moving the probe using the translation stage to the surface of the glue rather than inserting the probe into it. The right amount of epoxy will be transferred to the end of the probe due to surface tension. A photo of the final fabricated lensed fiber is shown in Fig. 3(d), and the radius of curvature of the half-ball lens was measured to be 103 µm using a digital microscope. The overall fabrication process is straightforward, and the probes have a consistent radius of curvature for the half-ball lens and the extension length. During tests of fabricating five probes with the same specs, the average error compared to the designed value was 3.01% +/− 2.97%. The polyimide tube protects the fiber probe and epoxy lens and allows some flexibility, which could reduce the chance of damage to tissue during use.

Figure 3. Assemble and farbrication of the imaging probe. (a) Assembled lensed fiber imaging probe. Fabrication stages of the lens: (b) offset set, (c) extension area fabricated, and (d) half-ball lens fabricated.

2.3. System Assembly and Software Development

The final assembled system is shown in Fig. 4(a), which includes the mini-PC, main CP-SDOCT chassis, and portable display. The main chassis of the CP-SDOCT features a double-side mounting design that conserves space and cost. The customized spectrometer and the SLED are mounted on the top of the chassis as in Fig. 4(b). All the optical components were mounted on adjustable positioners that have two degrees of freedom for fine alignment of the components. The power source, fiber coupler, and CMOS driver are mounted on the back side [Fig. 4(c)]. A compact, low-noise cooling fan is set right next to the power supply. The overall size of the system is 270 mm × 80 mm × 180 mm (W × H × D), which is smaller than a shoebox.

Figure 4. Assembled low-cost CP-SDOCT. Overal view (a), top view (b), and bottom view (c) of the system interior. CP-SDOCT, common-path spectral-domain optical coherence tomography.

Since free-hand scanning does not require high-speed scanning and the maximum acquisition speed of the CMOS sensor we used is only 4 kHz, we can significantly lower the overall cost of the system by choosing a cheaper and low-end customized desktop (BNUC11TNBI50000, 8GB RAM; Intel). The signal reconstruction algorithm using this desktop was developed for the CP-SDOCT system as follows. The input spectrum is 1 by 2,048 pixels, and remapping is conducted first to convert the wavelength-uniform OCT spectrum data to a K-space uniform set. Then the data is converted into a complex number and complex-to-complex (C2C) inverse Fourier transfer is performed to generate complex A-scan data. After that, the A-scan is converted back to the real number and the log scaling and threshold are applied. To catch up with the acquisition speed, we developed parallel computing algorithms to provide real-time imaging processing using the OpenCL framework in C++ and the time to reconstruct and transfer a single A-scan was measured to be 0.083 ms. To form a B-scan image while the probe is scanned by hand across the surface of the sample, we analyzed the lateral displacement between adjacent A-scans and rearranged them into equal spacing using the advanced speckle model, which was developed in [12]. The 4kHz A-scan line rate is sufficient to match the free-hand scanning speed and reconstruct the B-scan properly.

3.1. Cost-effective Analysis of the CP-SDOCT System and Imaging Probe

The cost of this common-path OCT system is significantly lower compared to conventional commercial OCT systems, because the spectrometer uses a low-cost low-frequency USB CMOS camera for signal detection without a frame grabber and the system uses a manually scanned common path lensed fiber probe for imaging to eliminate imaging optics, scanning device and electronic control for scanning. Also, the material required for fabrication of each lensed fiber probe is less than $ 10. The total component cost at the retail price for the fully functional system was under $ 5,000, and we expect it to be less than $ 2,000 for volume production with a minimal cost for the labor to assemble the system. We plan to offer the final clinical grade system at around $ 10,000. With the bulk purchase of the components and fabrication, we expect the build cost to decrease significantly while other costs associated with the regulatory and certification process increase. Current intraoperative OCT (iOCT) systems are only available as a module to surgical microscope systems and the total cost is in excess of $ 300,000, whereas the iOCT system is roughly 1/3 of the total cost. Thus, our proposed system allows iOCT capabilities at around $ 10,000 but also allows the use of existing surgical microscope systems. The overall size of the system is compact and can be easily fitted into the operating room and does not require separate installation.

3.2. Evaluation of the CP-SDOCT System

To test the system impulse response, we first used a glass slide as a target where the glass slide could be considered as having a delta-function reflectivity profile and evaluated the point spread function of the CP-SDOCT system as shown in Fig. 5(a). We fitted the system response peak in the A-scan image with a Gaussian function and characterized the axial resolution of the CP-SDOCT system, which was measured to be 11.5 µm. Based on the shape of the input SLED power spectrum and 3 dB bandwidth [17], the theoretic axial resolution of the system was calculated to be 10.8 µm. This indicates that the spectrometer is well assembled, and the reconstruction algorithm works as expected. The system roll-off was evaluated by placing the glass slide at different imaging depths. A total of 10 positions were used, and the peaks were observed as shown in Fig. 5(b). Here, the maximum SNR achieved at the end of the probe is around 57 dB, which is similar to the standard CP-SSOCT system. At an imaging depth of 1.2 mm, the signal is still 30 dB higher than the noise floor considering the diverging beam emitted from the bare fiber. Besides the main peaks, there are also small peaks in Fig. 5(b), and this is from the multiple reflections between the surface of the glass and the reference plane due to the high reflectivity of the glass.

Figure 5. System evaluation of resolution and signal roll-off. (a) Axial point spread function of OCT imaging with a full-width half maximum of 11.5 µm, and (b) CP-OCT signals obtained from a glass surface at different imaging depths. OCT, optical coherence tomography; CP-OCT common-path OCT.

A 10-layer Scotch tape phantom was first used to further test the system. To better mimic the surgical environment, the tape phantom was also placed under the water where the medium between the fiber and phantom changed from air to water. The imaging probe was scanned by hand across the surface of the phantom. We can see that the system has the capability to resolve all 10 individual tape layers when the phantom is placed either in the air [Fig. 6(a)] or under the water [Fig. 6(b)].

Figure 6. System evaluation on tape and porcine eye. Free-hand scan optical coherence tomography (OCT) image of 10-layer scotch tape, (a) in the air, (b) underwater, (c) lensed fiber probe tested on the porcine eye, and (d) OCT image of the porcine retina.

Lastly, the system was tested on ex-vivo porcine eyes, and the cadaveric tissues were directly purchased from Wagner’s Meat, Mt Airy. The experiment was done on the same day that the eyes were harvested. To mimic the surgical setup, a 25-gauge trocar was mounted on the side of the eye as shown in Fig. 6(c). The lensed fiber probe was then inserted into the eye, and we rotated the probe by hand to scan across the porcine retina. In the image of the retina as in Fig. 6(d), we can identify the exact position of the surface of the lens in front of the imaging probe. This helped us to maintain a safe distance from the retina to avoid unwanted damage, although the image quality was best when the probe touched the tissue due to the signal roll-off. The layers of the retina can still be identified in the free-hand scan images when the probe stayed away from the retina. We can also see that the hand motion is not noticeable during our scan. Here, we can successfully identify and label the retinal pigment epithelium (RPE) layer in the image, which could be useful for subretinal injection. A video of the probe scanning inside the eye with the real-time OCT image display is included in the supplementary material. In practice, the imaging probe could be used to identify the injection site or used during injection by free-hand scanning above the needle tip or glass pipe to guide the surgery. It could also be used for the evaluation of the induced bleb. In Fig. 6(d), the signal from the RPE layer is slightly lower than usual. One reason is that the experiment was carried out on ex-vivo porcine tissue whose quality was not as good as in-vivo tissue. One can see that the cornea of the eye exhibited clouding, indicating that the tissue quality was not that great even though this was harvested on the same day. This led to poor image quality and a rough retinal surface. For all the tests, multiple probes were fabricated and evaluated. We measured the SNR at the RPE location from five images acquired using different probes. The average SNR is 14.7 dB +/−2.65 dB, which proves the consistency of the fabrication process.

In this work, we proposed and implemented a low-cost free-hand scanning CP-SDOCT system. We also designed and fabricated a disposable epoxy lensed fiber imaging probe to provide better imaging depth while working in the intraocular space. A real-time parallel algorithm for reconstructing the OCT imaging and rearranging the free-hand scanned B-scan image was also developed. The performance of the system was evaluated and the measured axial resolution is 11.5 µm. The imaging capability was also validated using a tape phantom and ex-vivo porcine eye. The portable and free-hand scanning design makes the system easy to use and can be readily adapted to a wide range of different surgical scenarios. Easy fabrication of the cost-effective imaging probe could be made as the single-use disposable probes do not require sterilization. In summary, the CP-SDOCT system demonstrates the ability to image the retina in the intraocular space and meets the requirements for intraoperative use. Future work on the system will focus on further lowering the cost. A cheaper SLED source and CMOS sensor will be explored and tested to compare the performance with the current system. The design of the lensed fiber probe will be fine-tuned by using different glues and parameters to extend the working distance to further reduce the risk of touching the retinal tissue during the scan. To address the potential effect of hand motion, we will explore integrating a motion compensation mechanism into our handheld tool as described in [18]. More experiments and in-vivo studies will be carried out to test the stability of the system and optimize the imaging quality. The potential of our simple low-cost design has not been fully realized and our device is the first step toward this goal. More applications will also be evaluated in dermatological and dental applications for both diagnostic and surgical purposes.

The authors declare no conflict of interest.

Data underlying the results presented in this paper are not publicly available at the time of publication, but may be obtained from the authors upon reasonable request.

National Science Foundation (NSF SBIR Phase II grant #1951120).