Toward optical coherence tomography on a chip: in vivo three-dimensional human retinal imaging using photonic integrated circuit-based arrayed waveguide gratings

Optical coherence tomography (OCT) is a standard, noninvasive ophthalmic imaging modality whose performance has improved markedly in resolution and speed. However, conventional spectral-domain OCT (SD-OCT) systems remain bulky (~1 m³) and expensive (up to ~$100,000), creating a need for miniaturization and cost reduction. The study investigates whether photonic integrated circuits (PICs), specifically arrayed waveguide gratings (AWGs), can replace free-space diffraction gratings in SD-OCT to enable compact, robust, CMOS-compatible, and potentially mass-producible OCT systems. The research aims to demonstrate in vivo three-dimensional human retinal imaging and angiography using PIC-based AWGs and to benchmark performance against a commercial clinical SD-OCT.

Prior PIC developments for OCT have focused largely on the 1300–1550 nm range, including integrated Michelson and Mach–Zehnder interferometers, multimode interferometers, and polarization splitters, achieving sensitivities up to ~91 dB and axial resolutions down to ~13 µm in various implementations. For AWG-based OCT spectrometers, earlier demonstrations include: Nguyen et al. (2011) with a 195-channel SiON AWG at 1300 nm (SNR 75 dB, 19 µm axial resolution); Akca et al. (2012, 2013) with SiON AWGs at 800 and 1300 nm (125 and 195 channels; 25 µm and 20 µm axial resolution; first in vivo skin tomogram with 74 dB sensitivity at 47 kHz); and Ruis et al. (2019) with a TriPlex Si3N4 cascaded 512-channel AWG at ~850 nm (77 dB sensitivity at 1 kHz, 5.9 µm axial resolution, 6-dB roll-off at 400 µm). These systems often required complex packaging and lacked sufficient sensitivity for high-speed in vivo ophthalmic imaging without averaging. The present work addresses these gaps by demonstrating 256-channel AWGs fabricated in a fully CMOS-compatible process and achieving in vivo human retinal imaging without time-intensive averaging.

Two compact 256-channel silicon nitride AWGs (each ~13 × 14 mm² on a 20 × 20 mm² chip) were designed for nominal 850 nm operation but fabricated devices exhibited shifted central wavelengths: AWG 1 centered at 794 nm with 0.09 nm channel spacing (22 nm bandwidth; 782–804 nm) and AWG 2 centered at 875 nm with 0.19 nm spacing (48 nm bandwidth; 850–898 nm). Waveguides were fabricated using a fully CMOS-compatible process. For optical transmission characterization, a tunable Ti:sapphire laser (800–900 nm) was coupled via polarization-maintaining fiber into an inverted taper (tip ~160 × 160 nm²), with ~2.5 dB loss per fiber–PIC coupling event and ~0.5 dB/cm TM-mode propagation loss. Output from selected AWG channels (1, 8, 16, …, 256; 33 points) was collected with a single-mode fiber and measured with a power meter; mean transmissions were −15.51 dB (AWG 1) and −11.64 dB (AWG 2), normalized to the laser’s wavelength-dependent power. The OCT systems combined each AWG with a fiber-based interferometer. On the AWG output side, light was projected using achromatic lenses (reducing transmission losses by ~1–2 dB relative to fiber–fiber characterization). Sensitivity was measured using a neutral density filter in the sample arm (measured attenuations: 17.3 dB for AWG 1 and 15.4 dB for AWG 2) in front of a focusing lens and mirror; the point spread function SNR was determined and sensitivities computed by adding twice the ND attenuation to account for double-pass loss. A-scan rates and powers on the eye were optimized per system due to the wavelength shift and component differences: AWG 1 at 34 kHz and 67 kHz (830 µW on eye), AWG 2 at 20 kHz (480 µW). Axial resolution was measured from PSF FWHM at multiple depths, converted from air to tissue assuming refractive index 1.3549. Sensitivity roll-off and imaging depth were measured as a function of distance from zero delay. In vivo imaging of the right eye of a healthy volunteer was performed under institutional approval, using each AWG system to acquire three-dimensional retinal datasets; tomograms were compared qualitatively with those from a commercial clinical SD-OCT as a benchmark.

- First demonstration of chip-based OCT and OCT angiography enabling in vivo 3D human retinal imaging without time-intensive averaging using PIC-based AWGs with 256 channels. - AWG 1 (center 794 nm, 22 nm bandwidth, 0.09 nm/channel): measured sensitivity up to 91 dB at 34 kHz A-scan rate with 830 µW on the eye (88 dB at 67 kHz), axial resolution 10.7 µm in tissue (14.5 µm in air), imaging depth 1123 µm, ~6 dB roll-off at ~625 µm, mean transmission −15.51 dB. - AWG 2 (center 875 nm, 48 nm bandwidth, 0.19 nm/channel): measured sensitivity 90 dB at 20 kHz A-scan rate with 480 µW on the eye, axial resolution 6.5 µm in tissue (8.8 µm in air), imaging depth 645 µm, ~6 dB roll-off at ~380 µm, mean transmission −11.64 dB. - CMOS-compatible silicon nitride platform enables the prospect of monolithic co-integration of photodiodes and read-out electronics, potentially eliminating external cameras and easing packaging. - Comparison with a commercial clinical SD-OCT showed similar tomogram features, indicating strong clinical potential of AWG-based PIC spectrometers for ophthalmic OCT.

Replacing free-space diffraction gratings with PIC-based AWGs yielded a compact spectrometer compatible with in vivo ophthalmic imaging. The two AWG designs demonstrated that bandwidth and channel spacing trade-offs directly affect axial resolution and imaging depth/roll-off: the broader 48 nm bandwidth (AWG 2) improved axial resolution to 6.5 µm but exhibited a shorter imaging depth and faster roll-off, whereas the narrower 22 nm bandwidth (AWG 1) provided longer imaging depth at lower axial resolution. Achieved sensitivities (up to 91 dB) at eye-safe powers and practical A-scan rates enabled 3D retinal imaging without averaging, addressing prior limitations of AWG-OCT systems. The CMOS-compatible SiN platform supports future co-integration of photodiodes and read-out electronics, which can improve robustness, reduce alignment and packaging complexity, and ultimately lower size and costs compared with conventional SD-OCT. The qualitative similarity of tomograms to those from a commercial clinical system supports the viability of AWG-PIC spectrometers as a path toward OCT-on-a-chip solutions suitable for clinical environments.

This work demonstrates the first in vivo 3D human retinal OCT and OCTA measurements using PIC-based AWGs with 256 channels, achieving sensitivities up to 91 dB and axial resolutions down to 6.5 µm. The CMOS-compatible silicon nitride platform, together with the measured performance, indicates strong potential for fully integrated, compact, and robust OCT systems. Future directions include: monolithic integration of dedicated photodiodes and complete read-out electronics to remove external cameras; heterogeneous integration of light sources; optimization of AWG design to widen usable bandwidth while improving roll-off and imaging depth; reduction of coupling and propagation losses; and further packaging and system-level integration toward point-of-care clinical devices.

- Fabrication deviations led to shifted central wavelengths for both AWGs relative to the 850 nm design target, necessitating different system optimizations (coupler ratios, power levels, A-scan rates) and complicating direct dynamic-range comparisons. - Measured mean transmissions (−15.51 dB for AWG 1; −11.64 dB for AWG 2) and coupling losses (~2.5 dB per fiber–PIC event) indicate non-negligible insertion losses. - Sensitivity roll-off and imaging depth were limited (6 dB roll-off at ~380–625 µm; imaging depths 645–1123 µm), especially for the broader-bandwidth AWG 2. - Current demonstrations still rely on external fiber-based interferometer components and free-space projection optics at the AWG output; full monolithic integration (detectors, electronics, source) was not implemented in this study. - Transmission characterization sampled 33 of 256 channels, providing averaged estimates rather than per-channel characterization across the full spectrum.