Optical coherence tomography angiography (OCTA) is a powerful tool for non-invasive, label-free, three-dimensional visualization of blood vessels down to the capillary level in vivo. However, its widespread usage is hindered by the trade-off between transverse sampling rate and signal-to-noise ratio (SNR). This trade-off results in either a limited field of view (FOV) to maintain sampling density or loss of capillary details to fulfil FOV requirement. It also restricts microvascular quantifications, including flow velocimetry, which typically demand higher transverse sampling rate and SNR compared with standard qualitative OCTA. We introduce spectrally extended line field OCTA (SELF-OCTA), a cost-effective imaging modality that improves transverse sampling rate and SNR through spectrally encoded parallel sampling and increased signal acquired over longer periods, respectively. In the human skin and retina in vivo, we demonstrate its advantages in achieving significantly extended FOV without sacrificing microvascular resolution, high sensitivity to slower flow without compromising FOV, and flow velocity quantification with the highest dynamic range, emphasizing that these features can be achieved with readily available and standard OCTA hardware settings. SELF-OCTA has the potential to make wide-field, high-resolution, quantitative angiographic imaging accessible to a wider population, thereby facilitating the early detection and follow-up of vascular-related diseases.
Introduction
Microvascular network imaging in vivo provides valuable insights into the pathophysiology of a wide variety of diseases, including eye diseases, cardiovascular disorders, neurodegenerative diseases, and cancers. Specifically, microvasculature in the retina serves as a biomarker for eye disorders and offers a unique window to study the pathophysiological mechanisms of the human microcirculation1−4. Optical coherence tomography angiography (OCTA), a functional extension of optical coherence tomography (OCT), is capable of visualizing microvascular network at capillary level in three dimensions (3D) without the need for dye injection3−6. Over the past decade, there has been significant growth in its utilization3−9. Nevertheless, challenges such as restricted field of view (FOV) and limited capacity for blood flow quantification remain obstacles that slow down its widespread adoption in clinical practice5.
OCTA highlights blood vessels by detecting variations of OCT signals induced by moving red blood cells between repeated scans taken at the same position3−6. The OCTA transverse image is typically achieved through the point-scanning approach. Because of its sequential nature, there is an inherent trade-off between FOV and transverse sampling density since achievable FOV is determined by , where is total acquisition time, is A-scan rate, represents scan repeats at the same position with the minimum of 2, and is the transverse sampling density4,10. Consequently, most clinical OCTA systems have to under-sample microvasculature to expand FOV, which inevitably sacrifices capillary details4,10,11. Swept-source technology offers a solution to this issue with its significantly improved A-scan speed. However, higher imaging speed typically results in reduced OCT signal. Therefore, the speed of OCTA imaging is ultimately constrained by the signal-to-noise (SNR) requirement, which is constrained by permissible light exposure limit4.
Non-invasive blood flow quantification provides an opportunity to detect disorders before the onset of morphological changes, making it a hot and significant topic of research4,12−14. Previous studies have laid down the theoretical basis for OCTA-based velocimetry in a vascular bed15−20, which has been verified through extensive experiments using flow phantoms15−24. However, current OCTA devices, beyond merely quantifying vascular patterns and architectures, have limited capability to detect blood flow differences due to their adoption of a fixed, long interscan time (2.5–4.5 ms)15−18,21,22, which is the time interval between repeated scans taken at the same position and is approximately proportional to the OCTA signal strength. Using a long interscan time brings the OCTA signal strength close to optimal, thereby maximizing sensitivity for capillaries15−17,23,25. Whereas, the dynamic range, defined as the velocity range between the slowest detectable flow (sensitivity limit) and the fastest distinguishable flow (saturation limit), decreases as the interscan time increases and is distinct for a given interscan time4. Consequently, the velocity range of current clinical OCTA systems, which support only a fixed, long interscan time due to the insufficient transverse sampling rate, is limited to less than 0.3 mm/s4,13,15,17,25. To cover the velocity range of most human retina blood vessels (0.4–3.0 mm/s)15,23,25, multiple interscan time intervals, ranging from a couple of hundred microseconds to several milliseconds, are required. However, with current OCTA technologies, the number of distinct interscan time () for a given number of scan repeats () is determined by . This implies that, to achieve more interscan time intervals, FOV has to be sacrificed.
Furthermore, velocimetry typically demands much higher SNR compared to current morphological quantification. First, OCTA signals applicable for flow velocimetry are below the saturation limit, i.e. most capillaries present significantly lower signal strength than current, saturated OCTA. Second, to cancel the influence of vessel diameter on velocity assessment, flow velocity is computed based on the ratio between the unsaturated signal and the corresponding saturated signal12,13,16, through which errors in both signals will be transferred into the ratiometric result. Due to the requirements for interscan time and SNR, current standard OCTA technologies do not yet provide adequate temporal sampling rate to enable feasible velocimetry across the FOVs typically used in routine OCTA imaging. Although ultrafast research prototypes with a 400 kHz A-scan rate have achieved 3 interscan time intervals, with the shortest being 1.5 ms12,13,23. This is only applicable to a limited flow velocity range and 5 scan repeats were used to meet the SNR requirement for flow velocimetry12,13. Eventually, to extend the flow dynamic range, more scan repeats will be needed which will inevitably further reduce FOV, making it impractical for clinical translation.
We present a novel, low-cost technology termed spectrally extended line field OCTA (SELF-OCTA), designed to enhance both transverse sampling rate and SNR simultaneously. We have developed two SELF-OCTA devices: one operating at 1310 nm for skin imaging and another at 850 nm for retinal imaging. Our experimental findings demonstrate that SELF-OCTA offers significantly advanced solutions to the aforementioned challenges in OCTA imaging.
Results
SELF-OCTA principle
In the sample arm of a typical OCT system, a set of prism(s) is introduced in the infinite space, which disperses the polychromatic beam transversely into a line field (Fig. 1(a, b)). We split the interferometric spectrum into M bands using spectral windows equally spaced along the line field (Fig. 1(c)). The spacing r is approximately half of full-width at half maximum (FWHM) of the monochromatic transverse point-spread function (PSF) to satisfy the Nyquist sampling requirement. Spectral encoding allows simultaneous sampling of multiple transverse image positions without crosstalk26. With the line field spreading along Y (slow) axis, a 3D volumetric image, instead of a B-scan frame, is formed by mechanically scanning the line beam along the X (fast) axis. M partial-spectrum OCTA frames located at M consecutive Y-image positions could be obtained within each Y-scan cycle (Fig. 1(c)). For ease of illustration, we define Y-scan positions as Y locations illuminated by the first spectral band, Δy as the step size between 2 consecutive Y-scan positions (Fig. 1(d)). Under the Nyquist sampling condition, while the scanning step size Δy is equal to r in the point-scanning configuration, the step size Δy could be L·r in SELF configuration, where L is a positive integer no greater than M thereby increasing transverse sampling rate by a factor of L (Fig. 1(e)).
Figure 1.SELF-OCTA working principle. (a, b) Schematics of the 1300 nm system for skin imaging (a) and the sample arm of the system (b). SLD: superluminescent diode source; FC: fiber coupler; PC: polarization controller; CIR: circulator; L1–5: achromatic lenses; L6: camera lens; RM: reflective mirror; G: transmission grating; IMAQ: image acquisition card. (c) Process of splitting spectrum and generating partial-spectrum OCTA frame. K: wavenumber; M: the number of spectral bands, r: the transverse distance between two adjacent scan positions. DFT: discrete Fourier transform. (d) Schematics of signal mapping from partial-spectrum frames acquired over M/L consecutive Y-scan cycles to their Y image positions with M = 16 and L = 2. (e) Comparison of beam scanning between the point-scanning configuration where scanning step size along Y axis ∆y = r (left) and the SELF configuration with ∆y = L*r where L = 2 (right).
The slow (Y) axis scanning of SELF-OCTA is analogous to the flow production process, where multiple Y image positions are addressed in parallel, and frequency components at each image position are 'assembled' during multiple consecutive Y-scan cycles. That is, light beamlets of M/L spectral bands with distinct central wavelengths dwell at the same Y image position sequentially over M/L consecutive Y-scan cycles (Fig. 1(d)). In general, m-th (m = 1, 2, …, M) partial-spectrum OCTA frames acquired at j-th Y-scan cycle could be assigned with an index number of M·(j−1)+ m·M/L-quotient[(m−1)/L]; the final OCTA frame at i–th Y image position is the average of partial-spectrum OCTA frames with index number from (i−1)·M/L+1 to i·M/L (Fig. 1(d)). For example, when M = 16 and L = 2, the final OCTA frame at the Y image position of i = 15 is the mean of 8 partial-spectrum OCTA frames: 15th frame at 1stY-scan cycle, 13th frame at 2nd cycle, 11th frame at 3rd cycle, 9th frame at 4th cycle, 7th frame at 5th cycle, 5th frame at 6th cycle, 3rd frame at 7th cycle, and 1st frame at 8th cycle (Fig. 1(d)). The final OCTA frame at each Y image position is normalized to the mean source spectral density of corresponding spectral bands incident on the Y image position to compensate for the spectral density non-uniformity.
Extended field of view
With the same image acquisition settings including A-scan speed, image acquisition time, transverse sampling density, and interscan time, SELF-OCTA provides twice the FOV when L = 2 and three times the FOV when L = 3 as demonstrated in the human skin (Fig. 2, Table 1) and retina (Fig. 3, Table 2), respectively. The microvascular details are comparable between the SELF configuration and the point configuration on a one-to-one basis (Fig. 2(a, b) and Fig. 3(a−d)). The capillary details in SELF-OCTA en face projection may not be as sharp as that of the point scanning configuration (Fig. 2(a, b)), because the PSF along Y-axis is broadened by the convolution between monochromatic PSF and the spectral band. This issue is explained in Supplementary information Section 1.3 and Fig. S3. Nevertheless, this minor issue could be resolved with Y-deconvolution (Fig. 2(c) and Fig. S3).
Figure 2.Comparison of field of view in the human skin using 1300 nm system. (a–c) En face projections of vasculatures obtained with the point-scanning configuration (a), SELF configuration with M = 16 and L = 2 before (b) and after Y-deconvolution (c). From left to right: full-thickness projection (first column), capillary loops (second column), subpapillary plexus (third column), and deep vascular plexus (fourth column). Skin slabs are coded with green, yellow, and red in the full-thickness projection, respectively. (d) A schematic of skin vasculatures at different depth. (e–g) OCTA blood flow signal (red) superimposed on the OCT structural images (gray) with the point-scanning configuration (e), SELF configuration before (f) and after Y-deconvolution (g). Scale bars: 1 mm.
Table 1. Skin OCTA imaging parameters using the 1310 nm system.
Figure
Scanning approach
Optical power (mW)
A-scan speed (kHz)
A-scan volume (N×X×Y)
Imaging time (sec)
Interscan time (ms)
r (µm)*
∆y (µm)†
FOV (X×Y) (mm, pixels)
*: ∆x is step size between 2 adjacent A-scans along X (fast) axis. †: ∆y is step size between 2 consecutive Y-scan (slow axis scan) positions at the same X image position. ‡: Two different interscan time intervals are achieved with ∆t1 = 3.84 ms and ∆t2 = 7.68 ms using interlaced scanning protocol.
Fig. 2
Point
4.74
50
2×512×400
8.2
10.2
12.8
12.8
6.55×5.12 (512×400)
SELF
9.10
50
2×512×400
8.2
10.2
12.8
2×12.8
6.55×10.24 (512×800)
Fig. 4
Point
9.10
80
2×512×512
6.5
6.4
12.8
12.8
6.55×6.55 (512×512)
SELF
9.10
22
2×512×128
6.0
23.3
12.8
4×12.8
6.55×6.55 (512×512)
Fig. 5‡
SELF
9.10
50
2×384×384
2.9
-
12.8
2×12.8
4.92×9.84 (384×768)
Figure 3.Comparison of field of view in the human retina using 850 nm system at 68 kHz. (a–f) Images obtained from a female subject using 850 nm system with 90 nm spectral bandwidth: (a, b) OCTA en face projection with the point-scanning configuration over 12 mm × 4 mm (a) and the SELF-scanning configuration with M = 9 and L = 3 over 12 mm × 12 mm (b), respectively; (c, d) Corresponding zoom-in views of the fovea region acquired with the point (c) and SELF configuration (d), respectively; (e, f) Corresponding OCT structural image acquired with the point configuration (e) and SELF configuration after spectrum reconstruction (f). (g–k) Images obtained from a male subject using 850 nm system with 175 nm spectral bandwidth: (j) A partial-spectrum SELF-OCT cross-sectional image; (h, i) OCTA en face projections obtained with SELF configuration over 12 mm × 12 mm (h) and point configuration over 12 mm × 4 mm (i), respectively; (j, k) Corresponding zoom-in views of the fovea region acquired with the point (i) and SELF configuration (h), respectively. All OCT cross-sectional images are averaged over 2 consecutive images. Scale bars: 1 mm.
Table 2. Retinal OCTA imaging parameters using the 850 nm system.
Figure
Scanning approach
Optical power (mW)
A-scan speed (kHz)
A-scan volume (N×X×Y)
Imaging time (sec)
Interscan time (ms)
r (µm)*
∆y (L×r) (µm)†
FOV (X×Y) (mm, pixels)
*: ∆x is step size between 2 adjacent A-scans along X (fast) axis. †: ∆y is step size between 2 consecutive Y (slow) axis scan positions at the same X image position. ‡: Four different interscan time intervals are achieved with ∆t1 = 0.5 ms, ∆t2 = 1.0 ms, ∆t3 = 2.0 ms, and ∆t4 = 4.0 ms, where ∆t1 and ∆t2 are obtained using interlaced scanning protocol. §: The interscan time intervals are tailored using the customized interlaced scanning protocol to align with that of commercial devices operating at a comparable A-scan speed.
Fig. 3(a–f)
Point
0.84
68
2×800×460
10.8
5.9§
8.7
8.7
12×4 (800×460)
SELF
1.92
68
2×800×460
10.8
5.9§
8.7
3×8.7
12×12 (800×1380)
Fig. 3(g–j)
SELF
2.24
68
2×690×460
9.3
5.1§
8.7
3×8.7
12×12 (690×1380)
Fig. 6‡
SELF
1.92
100
3×200×400
2.4
-
8.7
1×8.7
1.74×3.48 (200×400)
With the system operating at 68 kHz, high-definition retinal angiography over 12 mm×12 mm area (45-degree angle) could be obtained in 10.8 s by utilizing the SELF technology using 2 (N)×800 (X)×460 (Y) A-lines (Fig. 3(b)). Such an imaging area traditionally requires the use of montage technique, which is troublesome and usually introduces image artifacts and misalignment. Given that L = 3, the corresponding transverse sampling density along Y-axis reaches 1380 (Y) pixels, more than 2 times higher than that of commercially mainstream devices running at 68–100 kHz A-scan speed (Fig. 3(b)). With this capability, SELF-OCTA wide field imaging allows the ready identification of the finest capillaries nurturing the fovea (Fig. 3(d)), and the precise delineation of the foveal avascular zone (FAZ), which is a key parameter for the early characterization of diseases such as diabetic retinopathy (DR) and glaucoma.
Note that SELF-OCT cross-sectional images have lower axial resolution compared with its corresponding point-scanning OCT images, because its full-spectrum bandwidth spreads into several transverse locations. Whereas, this issue can be addressed by performing spectrum reconstruction with the fast-axis scan along the line field direction, which provides axial resolving power similar to that of the point configuration (10.28 µm vs 10.07 µm for 1300 nm system and 5.3 µm vs 5.2 µm for 850 nm system, respectively) (Figs. S5(b) and S8(b),Fig. 3(e, f)). Although the partial spectrum data at each image position are acquired at different times, this time difference is small enough (several A-line periods) so that most of the dynamic particle are phase stable during fast axis scanning sequences. In addition, by utilizing a light source with a broader spectral bandwidth, SELF-OCTA could also achieve axial resolution comparable to that of standard commercial devices without spectrum reconstruction. As shown in Fig. 3(g), by employing the 175 nm full-spectrum bandwidth of the 850 nm light source, SELF-OCTA system achieves a partial-spectrum axial resolution of 7.12 µm, which enables clear recognition of each retinal layer. Similarly, within the same acquisition time (Table 2) the FOV is extended by 3 times in SELF-OCTA (Fig. 3(h)) with respect to the point scanning configuration (Fig. 3(i)) with comparable capillary details at fovea region (Fig. 3(j, k)).
Improved sensitivity to slow flow vessels with slower A-scan speed
Longer interscan times are preferred to detect slower capillary flow and microaneurysms in diseases like DR27. However, in the standard point scanning configuration longer interscan time requires longer exposure time per A-line or larger number of A-lines per frame. Consequently, achievable FOV within a certain fixation time is inevitably compromised, which goes against the requirement for wide-field imaging in disease screening and follow-up. SELF-OCTA could realize higher sensitivity to slower-flow vessels through achieving longer interscan time and exposure time with a relatively slower A-scan speed, without compromising field of view, transverse sampling density, and total image acquisition time (Fig. 4).
Figure 4.Comparison of sensitivity to slow flow in the human skin. (a) OCTA images with the point configuration at an A-scan rate of 80384 Hz giving a flyback of 14% FOV. (b) OCTA images with the SELF configuration at an A-scan rate of 22000 Hz with M = 16 and L = 4 having a flyback of 4.7% FOV. From left to right: en face OCTA projections of skin vasculatures of capillary loop (first column), subpapillary plexus (second column) and deep vascular plexus (third column). (c) Schematics of SELF-OCTA signal mapping from partial-spectrum frames acquired over M/LY-scan cycles to their Y image position with M = 16, L = 4 and ∆y = 4r. (d) Signal-to-noise ratio (SNR) as a function of exposure time. Optimal theoretical SNR (SNRTotal, orange line) was obtained when SNRel = SNRex (cyan and blue line). Black diamond and dot indicate measured total SNR at 80384 Hz and 22000 Hz A-scan rates, respectively. SNRreceiver: signal to receiver noise ratio, SNRexcess: signal to excess noise ratio, SNRshot: signal to shot noise ratio. Scale bars: 1 mm.
We validate this in human skin by conducting point-scanning configuration at 80,384 Hz A-scan rate (Fig. 4(a), Table 1) and SELF configuration with M = 16 and L = 4 at 22,000 Hz using the same optical power on tissue (Fig. 4(b, c), Table 1). OCTA image obtained with SELF configuration offers 3.65 times longer interscan time (23.3 ms vs 6.3 ms) and exposure time (45.5 µs vs 12.4 µs). As previously mentioned, as long as the OCTA signal does not reach saturation, a longer interscan time contributes to a higher signal strength. This significantly improves sensitivity to smaller vessels with slower flow, which are largely undetectable in the point-scanning case (Fig. 4(a)), but become clearly visible in the SELF-OCTA images (Fig. 4(b)). In addition, it is well-established that a longer exposure time provides better system SNR, particularly in excess noise limited systems like those powered by supercontinuum sources28−31 and receiver noise limited systems like the 1310 nm system used in this study. The SNR of this system measured to be 9.94 dB lower at 80,384 Hz than that at 22,000 Hz A-line rates (Fig. 4(d)), which can be approximately broken down to 5.84 dB drop in signal due to reduced exposure time and 4.1 dB drop in signal to excess noise ratio (SNRexcess). This SNRexcess drop significantly elevates the noise background and thus overwhelms signals from small vessels with slow flow in the point-scanning OCTA images (Fig. 4(a)).
High dynamic range towards blood flow velocimetry
Compared with the standard point-scanning approach, SELF technique supports multiple interscan time intervals with a single OCTA volume scan using a minimal number of scan repetitions. This makes it possible to realize blood flow velocimetry with high dynamic range over a feasible FOV, while maintain high sensitivity to slow flow in capillaries.
A customized interlaced scanning protocol along the fast (X) axis is designed to tailor the interscan time intervals in OCTA imaging (Fig. 5(a)). Details of this protocol are provided in Materials and methods (Section Interlaced scanning protocol in fast-axis to tailor interscan time intervals and Interleaved interscan time intervals along slow-axis scan). As exemplified in skin, there are OCTA frames of both Δt1 and Δt2 at each Y image position, which is achieved by alternating interscan time interval along Y (slow) axis scan (Fig. 5(b)). Differences in the visibility of blood vessels could be recognized between the en face projections of the two sub-volume OCTA images with different interscan time intervals (red arrows, Fig. 5(c)).
Figure 5.Multiple interscan time OCTA in the human skin using the SELF scanning configuration. (a) Schematics of customized interlaced scanning protocol. Blue and green dot lines represent A-scan points in the first and second scan repetitions, respectively. From top to bottom: interscan time intervals are shortened by reducing the A-scan points in a fast-axis run. (b) Schematics of SELF-OCTA signal mapping from partial-spectrum frames acquired over M/LY-scan cycles to their Y image position with M = 16, L = 2, and ∆y = 2r. By alternating between ∆t1 and ∆t2 along slow axis, M/L = 8 partial-spectrum OCTA frames at each Y image positions consisting of 4 frames with ∆t1 and 4 frames with ∆t2. (c) en face OCTA projections reconstructed from a single volume scan of 2 (N) × 384 (X) × 384 (Y) A-lines. Red arrows indicate differences in vascular visibility. Scale bars: 1 mm.
In the retina, we managed to realize 4 distinct interscan time intervals using SELF configuration running at 100 kHz within 2.4 s in B-M mode, which enables the reconstruction of relative flow velocity map with, to the best of our knowledge, the highest dynamic range (Fig. 6, Table 2). The scan volume is 3 (N)×200 (X)×400 (Y) A-lines over an area of 1.74 mm×3.48 mm. To achieve a wide dynamic range for blood flow velocimetry as well as a high sensitivity to capillaries, we managed the interscan time intervals ranging from 0.5 ms to 4.0 ms (Fig. 6(a)). Similar to observations in skin, capillary sensitivity improves with longer interscan time intervals (red arrows, Fig. 6(a)). Since the largest dynamic range roughly varies inversely to the shortest interscan time15, the dynamic range obtained by this 100 kHz system is approximately 10 times higher than that of commercial devices and 3 times higher than those 400 kHz research prototypes, whose shortest interscan time is ~4–5 ms4 and 1.5 ms12,13,23, respectively. With M = 9, L = 1 and N = 3, there are 9×3 spectral bands at each image position, encoding 4 different interscan time intervals (Fig. 6(f)). Building on previous established linear fitting models15 and employing the workflow to leverage signals provided by various interscan time intervals (Fig. 6(c, g)), we generate a pixel-wise merged relative flow velocity map. Details on the linear fitting model and image processing workflow are provided in Materials and methods (Section Linear fitting model and High dynamic range reconstruction model and Relative flow velocity map). Owning to the advances in dynamic range, flow velocity is distinguishable across a wide range of vessel diameters, including capillaries at the fovea and relatively larger vessels nurturing the macular area (Fig. 6(b)). In particular, the arterioles and venules can be readily recognized by their flow rate, which could be validated by the corresponding wide-field fundus photograph and OCTA image (Fig. 6(d, e)); these vessels were not distinguishable in previous studies due to their limited dynamic range13.
Figure 6.Multiple interscan time OCTA and relative flow velocity with high dynamic range in the human retina using the SELF configuration at 100 kHz. (a) En face projections of retinal vasculatures with 4 different interscan time intervals reconstructed from a single volume scan of 3 (N) × 200 (X) × 400 (Y) A-lines. (b) Corresponding pseudo color-coded relative flow velocity map with dynamic ranges contributed by ∆t1, ∆t2 and ∆t3. (c) Workflow for blood flow velocity map generation. (d, e) Corresponding fundus photograph (d) and OCTA image (e, adapted from Fig. 3(b)) from the same subject over 12 mm × 12 mm area (a 45-degree angle). Dashed box in (e) corresponding to imaging area in (a, b). (f) Schematics of SELF-OCTA signal mapping with M = 9, L = 1, and ∆y = r. Each Y image position has partial-spectrum frames encoding 4 different interscan time intervals. (g) Numerical simulations of OCTA signal and flow velocity: square root of amplitude decorrelation () as a function of square root of flow velocity () with IΔt1, IΔt2 and IΔt3 represent OCTA signal for ∆t1, ∆t2, and ∆t3, respectively. Solid lines refer to the linear range in-between the sensitivity threshold (δ) and the saturation threshold (1−δ) for ∆t1 (orang line), ∆t2 (green line) and ∆t3 (blue line), respectively. Dashed lines represent scaled and intercept-nulled decorrelation functions for ∆t1 (, orange dash line) and ∆t2 (, green dash line) with respect to ∆t3 (, blue solid line). Signals beyond the linear range are represented with dotted lines. HDR: high dynamic range. DRΔt1, DRΔt2 and DRΔt3: the corresponding flow velocity dynamic range. FAZ: foveal avascular zone, A: arterioles, V: venules.
OCTA is non-invasive and capable of resolving vascular details down to the capillary level, being able to identify capillary impairments that are typically undetectable using the invasive, dye-based examination3,4. Due to the constraints in transverse sampling rate and SNR in current standard OCTA practice, blood vessels usually have to be under-sampled in the transverse plane to achieve clinically feasible field of view, which compromises OCTA image resolution at risks of creating illusion of vessel drop-out or missing critical vascular details4,10,11. SELF-OCTA enables simultaneous signal acquisition from multiple transverse image positions. This parallelization makes transverse sampling much faster than the point-scanning configuration running with the same A-scan speed. While achieving faster transverse sampling rate, one of the important features of SELF-OCTA over the standard point-scanning OCTA is that the power dispersed into multiple points in a line field, so that it allows higher total input power and thus provides higher SNR. It is worth mentioning that this does not translate to a higher power density at any individual point on the sample. The fundamental limitation on power density per point remains. The improvement in SNR is mainly due to increased signal acquired over longer periods, rather than an increase in instantaneous power allowed to deliver to each point.
Towards wide field imaging for comprehensive assessment of retina disorders, the capability to mitigate the constraints in FOV, sampling density, and interscan time will make OCTA a more powerful tool, paving the way for its adoption as a first-line examination. This is particularly beneficial in busy clinical scenarios where invasive, time-consuming examinations like fluoresceine angiography are not practical3,32. It will be particularly useful in the diagnosis, monitoring, and treatment of diseases such as DR, which affects one-third of patients with diabetes. DR affects both the central and peripheral regions of the retina, requiring routine follow-up with high resolution, wide-field, and variable interscan time angiographic imaging tools such as SELF-OCTA. Additionally, it is also helpful in managing ocular emergencies like retinal artery occlusion.
Towards flow velocity quantification, SELF-OCTA offers the possibility to provide multiple interscan time intervals with minimal scan repeats. This unique capability is attributed to the improvement in both the transverse sampling rate and SNR. With the SELF configuration, we managed to achieve 4 interscan time intervals using an A-scan speed commonly used in clinical settings and realized relative blood flow velocity map at the capillary level in the human retina in vivo with the highest dynamic range to the best of our knowledge12,13,23. Although dynamic range is primarily determined by the shortest interscan time interval, multiple, distinct interscan time intervals remain helpful in recognizing and understanding hemodynamic abnormalities because each interval offers the best SNR for a distinct flow velocity range. Therefore, various intervals may facilitate the readily detection of regions with existing hemodynamic disorders, reducing the learning curve for OCTA-based flow velocity assessment. These benefits of SELF-OCTA in aiding flow velocity quantification will ultimately help in understanding the pathogenesis of diseases that progress through different stages of flow impairment rather than a rapid progression to microvascular atrophy, including but not limited to macular degeneration, diabetic retinopathy and glaucoma12,13,33.
In this pilot study, flow velocity map is generated using a 100 kHz system within 2.4 s over an area of 1.72 mm×3.44 mm, where FOV is limited by the step response of Galvo scanners, a practical constraint on achievable shortest interscan time interval. This could be mitigated by reducing the step size of interlaced fast axis scan with higher A-scan rate or use of 2 or more fast-axis Galvo scanners. Moving forward, it is desirable to extend the FOV, making velocity evaluation accessible over areas beyond the macula. Also, as observed in previous studies, abrupt fluctuations could be observed in certain vessels from the relative flow velocity map, likely attributed to artifacts induced by eye motion during in vivo imaging12,13.
There is a trade-off between the axial resolution and transverse resolution along the direction of the line field, since the spectrum is dispersed in the transverse direction, the spectral bandwidth at each transverse position is reduced in the SELF configuration. Nevertheless, by aligning the fast axis scan with the direction of the line field, the axial resolution and resolving power of SELF-OCT could be reconstructed and are comparable to that of the point-scanning configuration. The resolution shortcomings with respect to the standard clinical devices can also be compensated by use of commercially available light sources with broader bandwidth34,35. For example, by employing the full ~175 nm spectral bandwidth of the light source used for retina imaging in this study, all retina layers could be clearly resolved from the partial-spectrum OCT structural images. Other than approaches mentioned above, alternative solutions like machine learning powered spectral processing36,37 and deconvolution38 can potentially restore axial image resolution of partial-spectrum data.
Conclusions
The SELF-OCTA platform represents a significant step forward towards next-generation angiographic imaging. This includes, but is not limited to, wide field of view for comprehensive disease assessment, high sampling density to detect subtle vascular changes, and quantitative functional imaging to detect disorders prior to morphological changes. More importantly, these OCTA features could be realized in devices with limited A-scan speed by overcoming the limitations in transverse sampling rate without sacrificing system sensitivity. This capability is particularly advantageous for SD-OCTA systems, which are relatively more affordable and thus widely used but typically have limited A-scan speed. Besides, SELF-OCTA overcomes the bottleneck for flow velocity quantification in vivo, making this important functional imaging parameter available without the need for an ultrafast system, which is expensive and primarily used as a research prototype. With all the aforementioned capabilities, we believe SELF technology would significantly increase the accessibility of high-quality, non-invasive qualitative and quantitative OCTA imaging to broader populations. These advancements hold the potential to enhance the understanding, screening, and management of microcirculation-related disorders in the retina and beyond, including various systems such as the cardiovascular system.
Materials and methods
1310 nm OCT system for skin angiographic imaging
Optical design and characterization of the line field
The 1310 nm spectral domain OCT (SD-OCT) system uses a superluminescent diode with spectral output from ~1240 nm to ~1360 nm (−6 dB). For the ease of comparison, the system can conveniently switch to the point-scanning configuration by replacing the prisms with a mirror through a manual translation stage (Fig. 1(a)). Detailed information on system design and construction is provided in Supplementary information Section 1.1.
In the SELF configuration, the collimated sample beam passes through 3 identical prisms (N-SF11, apex angle: 30°, angular spacing of ~56.9°) with the incident angle of 26.9°. With an objective lens of focal length 50 mm, line field length at the focal plane of the objective lens is ~252 µm corresponding to the spectral bandwidth of ~152.3 nm detected by 1024 pixels of the spectrometer. According to the Zemax model, the field angle of the sample beam along the Y-axis is approximately a linear function of wave number (Fig. S1), and the transverse distance (Y-axis) to wave number conversion ratio was ~4.51×10−5 mm/cm−1. Detailed information of the linear relation between Y coordinates and the wavenumber is provided in Supplementary information Section 1.2.
We use M = 16 Hamming windows with size of 263 pixels (64.72 µm) and spacing of 52 pixels (12.8 µm) to split the full spectrum of 1024 pixels (Fig. S2). In the wave number space, the spectral range of each Hamming window is ~1,437 cm−1 (~39.1 nm). The FWMH spectral bandwidth of the Hamming window is ~20.5 nm, corresponding to a transverse distance of 33.9 μm, which results in a 52% broadening of the monochromatic PSF (Fig. S3). Details of the effect of spectral filtering on transverse resolution is provided in Supplementary information 1.3.
Maximum permissible exposure for skin imaging
As shown in Table 1, different scanning speed were used for skin imaging. It is understood that scanning speed, number of scan repeats, and sampling density are crucial factors for determination of MPE. For our study, we follow the more restrictive guideline by IEC 60825-139: classifying our scanning instrument based on a stationary beam. Following the 'Most Restrictive Ratio' method40, the angular subtense in Y-axis is ~3.176 mrad, so that the extended source correction factor CE is 1.558741. Assuming 4.74 mW as the MPE limit for skin in the point-scanning configuration, the optical power for the SELF configuration corresponding to the simulated maximum permissible exposure (MPE) of the point-scanning configuration is (4.74 mW × CE) / 0.799 = 9.25 mW39−41, since the normalized partial power within the angular subtense was 0.799 (Fig. S4). This value is almost 2 times of that of the point-scanning configuration. A detailed method to determine the corrected MPE limit is provided in Supplementary information Section 1.4.
SNR Characterization of 1310-nm system
SNR was compared at 50 kHz A-scan rate with a partial reflector (−38.74 dB)42,43. The optical power at the fiber tip of the sample arm was the same for both configurations. In the SELF configuration, we performed fast-axis scanning along the direction of the line field with a scanning step size of 12.8 µm. There are 16 partial-spectrum A-scans acquired during 16 consecutive Y-scan cycles at each Y image position. We coherently combined all 16 partial-spectrum interference fringes at each image position before the standard signal processing procedures. Since Galvo scanning may introduce changes in optical path-length between the partial reflector and the reference reflector, we corrected the phase difference between interferometric signals of consecutive A-scans following the method described by An et al.44 before the coherent combination. The dispersion imbalance induced by the prisms was compensated numerically following the previously established method45. Axial profiles are obtained by Fourier transform of the spectral interferometric data, with which signal strength, noise floor, and the axial resolution of both the full-spectrum and partial-spectrum OCT signals were measured.
The measured signal strength of the SELF configuration is ~17% lower than the point-scanning configuration, which is mainly due to the round-trip reflection loss at the prism surfaces (~13.5%) (Fig. 5(a, b)). The noise in SELF configuration is ~15% lower than the point-scanning configuration (Fig. S5(a)), because in the coherent addition of partial-spectrum fringe data, noise at each spectral bin is averaged over multiple partial-spectrum fringe data acquired at different time. As a result, the measured SNR of the SELF configuration is 106.1 dB, comparable to that of the point-scanning configuration (106.3 dB). Whereas, since the thermal damage threshold of an extended source is higher than the corresponding point source41,46−49, the corrected MPE limit of the line field of the 1310 nm is almost 2 times larger than the point configuration, equivalent to ~3 dB advantage in SNR.
Resolution characterization
Isotropic spatial resolutions are designed for OCTA signal analysis in SELF configuration. The line field length is split into M = 16 spectral bands spacing at r = 12.8 µm, which corresponds to the distance between two adjacent Y image positions. The axial resolution with the full-spectral bandwidth is measured to be 10.28 µm in air, comparable to that of the point-scanning configuration which is 10.07 µm in air (Fig. S5(b)). After the spectrum is filtered with a spectral window, the axial resolution of a partial-spectrum band is measured to be ~28.5 µm in the skin with refractive index = 1.38 (Fig. S5(c)).
The measured monochromatic transverse resolution along X-axis is 26.2 µm (10%−90% edge width) or 24.1 µm (FWHM); the polychromatic transverse resolution along Y-axis is measured to be 39.6 µm (10%–90% edge width) or 36.4 µm (FWHM), broadened by 52% through convoluting the monochromatic transverse PSF with the spectral window. We conduct deconvolution on en face projections of angiograms along the Y direction using the Lucy-Richardson method (deconvlucy in MATLAB®), and is restored to 28.8 µm (Fig. S5(d)), which is corroborated by the en face images of resolution chart: after Y-deconvolution, element 5 in group 4 of the resolution chart with a line width of 19.69 µm can be clearly resolved in both Y-axis and X-axis (Fig. S5(e)).
Skin image acquisition
OCTA images from the palm side of the proximal interphalangeal joint of the middle finger in healthy human subjects were obtained with N = 2 scan repeats at the same position. We used a lab jack (L200, Thorlabs Inc.) as the vertical hand-rest and a cage plate (CP33T/M, Thorlabs Inc.) as the horizontal hand-rest to minimize motion. Ultrasound transmission gel (Aqua Sonic 100) was applied to avoid the high reflection signal from the skin surface. Hand-rest was in place to minimize involuntary hand movement. Details of imaging parameters are listed in Table 1. This study is approved by the Institutional Review Board (IRB) of Nanyang Technological University (IRB-2016-10-015).
OCT and OCTA image processing in skin
To generate OCTA images, we use the same Hamming windows to split the interferometric data acquired with both the point and SELF configuration. For SELF case, OCTA images are obtained by averaging M/L partial-spectrum decorrelation signals at the same transverse position as illustrated in Fig. 2(b). In Y-deconvolution, we use the Hamming window mentioned above as the input point-spread function and a damping threshold of 2. For data obtained with point-scanning configuration, OCTA images are generated following the split spectrum amplitude decorrelation angiography (SSADA) algorithm using the same Hamming windows as for the SELF configuration50. We manually segmented the cross-sectional OCTA images into 3 slabs according to image depth with respect to the skin surface and obtained the en face projection of vascular network of each slab using ImageJ.
For data obtained with the SELF configuration, we coherently combine all the partial-spectrum interference fringes acquired at the same transverse image position to reconstruct OCT structural image. There are M/L partial-spectrum fringe data at each transverse position acquired during M/L consecutive Y-scan cycles (M = 16). Since data at the same position is obtained at different time, we correct bulk motion among all the partial-spectrum interference fringes before coherent combination, following the method described by An et al44. All the data was processed using MATLAB (R2021, MathWorks, Inc.).
850 nm OCT system for retina angiographic imaging
Optical design and characterization of the line field
We developed an 850 nm system capable of operating at both point-scanning and SELF configuration (Fig. S6). The detailed system construction is provided in Supplementary information Section 2.1. The 850 nm SD-OCT system uses a superluminescent diode array (T850, Superlum Diodes, Ireland) with a spectral range from 755 nm to 930 nm. In all the retina angiographic imaging, we limit the spectral range of the input light to 90 nm (~805 nm to 895 nm) by switching off the diode 1 and use of a dichroic filter (TSP01-887-25x36, Semrock) in the sample arm except for Fig. 3(g–k) where the full 175 nm spectrum is used. The following characterization is based on the system with 90-nm bandwidth.
In the SELF configuration, we use a prism (N-SF11, apex angle: 30°) with the incident angle of ~27.13° for line field generation. The corresponding line field length at the retina is measured to be approximately 94 µm assuming an eye focal length of 17.2 mm. We used M = 9 Hamming windows with size of 650 pixels (~413 cm–1 in wavenumber space and ~30 µm in the spatial coordinates along the Y-axis) and spacing of 190 pixels (~121 cm−1) to split the full spectrum of 2048 pixels. Transverse spacing r between two adjacent spectral bands at the retina is calibrated to be ~8.7 µm with a USAF 1951 resolution chart.
Maximum permissible exposure for 850 nm system
As shown in Table 2, different scanning speed and number of scan repeats, and sampling density were used for retinal imaging, which are crucial factors for determination of MPE. For our study, we follow the more restrictive guideline by ANSI Z80.36-202149: classifying the OCTA device as a Group 1 instrument for which no potential light hazard exists. Following the 'Most Restrictive Ratio' method40,49, the angular subtense in Y-axis is ~5.372 mrad and the extended source correction factor CE is calculated to be ~2.2941 (Fig. S7(a, b)). The ratio between the power on the cornea and the power of the collimator output is 77% and 83% for the point scanning configuration and SELF configuration, respectively. Assuming 0.84 mW on the cornea as the MPE of the of the point-scanning configuration, the optical power for the SELF configuration corresponding to the simulated MPE was larger than 1.92 mW40,49. A detailed method to determine the sample power for SELF configuration is provided in Supplementary information Seciton 2.2.
SNR characterization of 850 nm system
SNR is compared at 100 kHz A-scan rate with a partial reflector (−33.56 dB). The measured signal strength of SELF configuration is comparable to that of the point-scanning configuration (Fig. S8(a)). The noise in SELF configuration is ~27.7% lower than the point-scanning configuration because noise at each spectral bin is averaged over multiple partial-spectrum fringe data acquired at different time (Fig. S8(a)). As a result, the measured SNR of the SELF configuration is slightly higher than the point configuration with the same power on the cornea: 2.02 mW (99.1 dB vs 97.4 dB). Since the thermal damage threshold of an extended source is higher than the corresponding point source41,46−49, the corrected MPE limit of the line field of the 850 nm is almost 2 times larger than the point configuration, equivalent to ~3 dB advantage in SNR.
Resolution characterization
Using an eye model with a focal length of 17.2 mm, the full-spectrum line filed length is simulated to be ~94 µm with the Zemax® model, which is split into M = 9 spectral bands with spacing r = 8.7 µm. The axial resolution was measured to be 5.2 µm for point configuration and 5.3 µm for SELF configuration with the full spectral bandwidth in air (Fig. S8(b)), which was ~16.5 µm in tissue for a single spectral band after above-mentioned spectral window was applied (refractive index = 1.38) (Fig. S8(c)). For the point configuration, the X-Y transverse resolution is measured to be 13.7 µm at 10%−90% edge width (orange dashed line, Fig. S8(d)). For SELF configuration, the transverse resolution in X-axis is monochromatic, measured with 13.7 µm at 10%−90% edge width; since the polychromatic spectrum is broadened along Y-axis by 51.6%, the Y-axis transverse resolution is measured to be ~20.77 µm at 10%−90% edge width (blue solid line, Fig. S8(d)), which is restored to 13.7 µm after Y-deconvolution is performed (black dotted line, Fig. S8(d)).
Retina image acquisition
The study of human retinal imaging is approved by the Institutional Review Board (IRB) of Nanyang Technological University (IRB-2019-05-050). We imaged the retina in healthy human subjects using N = 2 scan repetitions at the same position for OCTA signal analysis, except for flow velocity analysis in Fig. 6 where N = 3 was used. Details of retinal imaging parameters are provided in Table 2.
In the SELF configuration, to achieve full-spectrum SELF-OCT images (Fig. 3(f)), we conducted a cube scan with the fast axis along the line field direction, right after the OCTA volume scan in Fig. 3(b). The time interval between partial-spectrum data at the same position is no more than (M−1) A-scan periods, where M is the number of spectral windows. This ensures that phase errors induced by eye motion are negligible. Over the 12 mm×12 mm imaging area, the Y (fast) axis scanning step size is set to be r = 8.7 µm and X (slow) axis step size is 87 µm.
Retinal OCTA and OCT image processing
We processed the retinal data with the complex-based algorithm rather than the intensity-based decorrelation to generate OCTA images, eliminating the need to address the decorrelation artifacts in the low intensity regions for convenience purposes51. For data obtained with the SELF configuration, M/L partial-spectrum fringe data at each transverse image position is acquired during M/L consecutive Y-scan cycles. After bulk motion correction44, we subtract the raw spectrum in the wave number space from its repeat, following algorithm similar to OMAG52, and split the differential data into M = 9 spectral bands and Fourier transform them to obtain 9 partial-spectrum axial profiles. The resultant M/L partial-spectrum axial profiles located at the same transverse image position are incoherently averaged to generate the OCTA axial profiles, whose contrast is enhanced following a previous study53. For data obtained with the point-scanning configuration, we adopted the same approach for processing data obtained with the SELF configuration, except that M partial-spectrum axial profiles at each image position are all from the same A-scan. We manually segmented the inner retinal layers and obtained the retinal vascular network by en face projection of the full thickness of the segmented region using ImageJ.
For the data acquired with the additional cube scan along the fast axis, like in skin imaging (Materials and methods, Section SNR Characterization of 1310-nm system), we first removed the phase difference between two adjacent A-scans caused by the Galvo scanning, and coherently combined all M partial-spectrum interferometric data at the same transverse position into a full-spectrum interferometric data before generating structural through Fourier Transform. All the data was processed using MATLAB (R2021, MathWorks, Inc.).
Interlaced scanning protocol in fast-axis to tailor interscan time intervals
To obtain shorter interscan time without decreasing sampling density for a given A-scan speed, we tailored the A-scan points of a fast-axis run within a B-frame through "interlaced scanning" sequences (Fig. 5(a)). Specifically, assuming we used a system operating at 100 kHz with 200 A-sans per B-frame (), if the fast axis galvo-scanner addressed all the 200 A-scan points in a fast-axis run followed by its repetition, an interscan time interval of 2 ms could be obtained, which is typical for a B-M scan mode in conventional OCTA imaging. If we divided the 200 sampling points of the B-frame into 2 fast-axis run cycles (100 points for each run), conducting the first series of fast-axis run and its repetition, followed by the second series, we could achieve an interscan time of 1.0 ms. Similarly, we could tailor the time intervals into 0.5 ms by dividing the sampling points in the B-frame into 4 (50 points for each run).
With interlaced scanning protocol, we tailored the interscan time intervals of 12 mm×12 mm retinal OCTA scan shown in Fig. 3 to improve sampling density without prolonging interscan time interval, thereby making the intervals comparable to those of commercial devices.
Interleaved interscan time intervals along slow-axis scan
Based on SELF platform, we could encode multiple intervals within a single OCTA volume scan, by interleaving different interscan time intervals along Y (slow) axis scan cycles (Figs. 5(b), 6(f)).
In the skin angiographic imaging (Fig. 5(b, c)), with scan repetitions at the same position N = 2, we interleaved the interscan time intervals of ∆t1 = 3.84 ms and ∆t2 = 7.68 ms along Y (slow) axis scan (Fig. 5(b)). Given that M = 16 and L = 2, there were 8 partial-spectrum OCTA frames at each image position, with 4 frames encoded with each of the two interscan time intervals (Fig. 5(b)).
In the retina angiographic imaging (Fig. 6(a, f)), we managed to realize 4 distinct interscan time intervals within one OCTA volume scan where ∆t1 = 0.5 ms, ∆t2 = 1.0 ms, ∆t3 = 2.0 ms and ∆t4 = 4.0 ms. With the interlaced scanning protocol in fast axis, fundamental interscan time intervals of ∆t1 and ∆t3 interleaving along Y (slow) axis were obtained between 2 consecutive scan repetitions (Fig. 6(f)). Given that scan repetitions N = 3, intervals of ∆t2 and ∆t4 could be obtained between the first and third scan repetitions (Fig. 6(f)). Therefore, with M = 9 and L = 1, at each image position there were 9×3 partial-spectrum OCTA frames encoding the 4 interscan time intervals (Fig. 6(f)).
Linear fitting model and High dynamic range reconstruction model
In general, there is a sigmoid relation between amplitude decorrelation signals and blood flow velocity which can be used for blood flow velocity quantification15. In this study, we obtained amplitude decorrelation signals using a complex (phasor) based algorithm similar to optical microangiography (OMAG)17,44,52. This is because complex signal based algorithm results in fewer uncertainties of flow estimation than the amplitude based algorithm20. Recalling the principle of OMAG, which takes the function and can be re-written as17
where and are the amplitudes at time t and t + ∆t, is the phase difference. The expression in the square bracket of Eq. (1) denotes decorrelation where is the amplitude autocorrelation function. represents the difference between output and input beam wave-vectors, is the mean-square displacement of the scatter in time ∆t and can be described as in case of random flow. Here, is flow velocity and ∆t represents interscan time.
The amplitude terms and can be cancelled by normalizing the OCTA signals (∆t1 = 0.5 ms, ∆t2 = 1.0 ms, and ∆t3 = 2.0 ms) with the corresponding saturated one (∆t4 = 4.0 ms). Assuming the phase difference is governed by the Gaussian random distribution with a mean phase-difference of the averaged, normalized OCTA signal is the square root of amplitude decorrelation17,54
where is the amplitude decorrelation and ∆t may take ∆t1, ∆t2, or ∆t3. Similar to the linear model established between amplitude decorrelation and velocity15, the normalized OCTA signal () and the square root of velocity () is approximately linearly related (Fig. 6(g)):
where is the slope parameter which is linearly dependent on the interscan time15, and is the intercept parameter.
We numerically simulated the OCTA signal as a function of according to Eq. (2) where is arbitrarily chosen (Fig. 6(g)). We define the dynamic range of the OCTA signal to be [δ, 1−δ], where δ is the sensitivity threshold and 1−δ is the saturation threshold15. We linearly fit the functions within the dynamic range, and confirm the following relation
The intercept parameter can be broken down to
where is related to the decorrelation caused by receiver noise, excess noise, and shot noise, is related to the decorrelation caused by Brownian motion which is neglected in this study as they are much smaller than the convection flow, and is obtained by linear fitting of the OCTA data within the dynamic range.
As shown in Fig. 6(g), the linear range of the normalized OCTA signals are illustrated as and (solid orange, solid green and solid blue curve, respectively). In this study, we empirically set the sensitivity threshold δ = 15%. The corresponding is −0.2985, which is the same for OCTA signals of different interscan times based on the linear simulation model and is also independent of according to Eq. (2). To leverage the dynamic range provided by different interscan time intervals, we multiply the OCTA signals of ∆t1 and ∆t2 with and , respectively, so that the rescaled signals ( and ) have the same slope as (orange dash and green dash lines, Fig. 6(g)), allowing high dynamic range (HDR) reconstruction.
Relative flow velocity map
Prior to flow velocity assessment in OCTA images, we developed a binary vessel mask to remove noises in non-vessel areas to highlight the vascular architecture within the tissue, i.e., retina55. En face angiograms of the inner retina is generated by maximum projection of the OCTA signal, which is then denoised using Block-matching and 3D filtering, processed using global threshold, followed by Hessian filter and local adaptive filter following the previously reported method55. The global threshold determined by the noise level at the foveal avascular zone (FAZ) was manually measured. The adaptive filter adopted the local threshold defined by the mean value within a 3×3 window to generate the binary vessel map55. The resulting image thereafter serves as the mask, setting pixels between blood vessels to zero (Fig. 6(f), S9(a)). All the data was processed in MATLAB (R2021, MathWorks, Inc.).
Building on the linear fitting model established earlier, we attempted to create a relative flow velocity map by leveraging the dynamic range offered by different interscan time intervals (∆t1, ∆t2 and ∆t3). First, to eliminate the effect of noise , we subtracted the mean signal measured from the FAZ from the corresponding en face projection (mean projection), which is obtained through manually segmenting the FAZ in the cross-sectional OCTA images and carefully excluding those affected by ocular motion. The estimated parameter are measured to be 0.115, 0.125, 0.124, and 0.148 for ∆t1 = 0.5 ms, ∆t2 = 1.0 ms, ∆t3 = 2.0 ms, and ∆t4 = 4.0 ms, respectively. The contribution of Brownian motion is neglected as mentioned above. Second, note that, to eliminate the influence of vessel diameter on velocimetry, the noise-corrected OCTA signals of ∆t1, ∆t2 and ∆t3 were normalized with that of ∆t4 on a pixel-by-pixel basis, after averaged over a connected domain of 21×21 pixels in the vessel mask12,13,16 (Fig. S9(b)). Third, OCTA signal outside the dynamic range [δ, 1−δ] was excluded from the processing where δ = 15% which results in =−0.2985. Fourth, normalized OCTA data including and were scaled following and . Fifth, the were nulled and resultant OCTA images of 3 interscan time intervals were averaged pixel-wise. Lastly, the OCTA signal was squared to generate the relative flow velocity map with merged dynamic range (Fig. 6(b)).
[1] CYL Cheung, CheungCYL, IkramMK, al SabanayagamCet, MK Ikram, CheungCYL, IkramMK, al SabanayagamCet, C Sabanayagam et al. Retinal microvasculature as a model to study the manifestations of hypertension. Hypertension, 60, 1094-1103(2012).
[2] HS Shi, ShiHS, KoronyoY, al FuchsDTet, Y Koronyo, ShiHS, KoronyoY, al FuchsDTet, DT Fuchs et al. Retinal capillary degeneration and blood-retinal barrier disruption in murine models of Alzheimer’s disease. Acta Neuropathol Commun, 8, 202(2020).
[3] AH Kashani, KashaniAH, ChenCL, al GahmJKet, CL Chen, KashaniAH, ChenCL, al GahmJKet, JK Gahm et al. Optical coherence tomography angiography: a comprehensive review of current methods and clinical applications. Prog Retin Eye Res, 60, 66-100(2017).
[4] RF Spaide, SpaideRF, FujimotoJG, al WaheedNKet, JG Fujimoto, SpaideRF, FujimotoJG, al WaheedNKet, NK Waheed et al. Optical coherence tomography angiography. Prog Retin Eye Res, 64, 1-55(2018).
[5] M Niederleithner, NiederleithnerM, L deSisternes, al StinoHet, Sisternes L de, NiederleithnerM, L deSisternes, al StinoHet, H Stino et al. Ultra-widefield OCT angiography. IEEE Trans Med Imaging, 42, 1009-1020(2023).
[6] DM Sampson, SampsonDM, DubisAM, al ChenFKet, AM Dubis, SampsonDM, DubisAM, al ChenFKet, FK Chen et al. Towards standardizing retinal optical coherence tomography angiography: a review. Light Sci Appl, 11, 63(2022).
[7] E Borrelli, BorrelliE, SarrafD, al FreundKBet, D Sarraf, BorrelliE, SarrafD, al FreundKBet, KB Freund et al. OCT angiography and evaluation of the choroid and choroidal vascular disorders. Prog Retin Eye Res, 67, 30-55(2018).
[8] BJ Vakoc, VakocBJ, LanningRM, al TyrrellJAet, RM Lanning, VakocBJ, LanningRM, al TyrrellJAet, JA Tyrrell et al. Three-dimensional microscopy of the tumor microenvironment in vivo using optical frequency domain imaging. Nat Med, 15, 1219-1223(2009).
[9] KC Liang, LiangKC, WangZ, al AhsenOOet, Z Wang, LiangKC, WangZ, al AhsenOOet, OO Ahsen et al. Cycloid scanning for wide field optical coherence tomography endomicroscopy and angiography in vivo. Optica, 5, 36-43(2018).
[10] SS Gao, GaoSS, JiaYL, al ZhangMet, YL Jia, GaoSS, JiaYL, al ZhangMet, M Zhang et al. Optical coherence tomography angiography. Invest Ophthalmol Vis Sci, 57, OCT27-OCT36(2016).
[11] J Ho, HoJ, DansK, Set YouQ, K Dans, HoJ, DansK, Set YouQ, Q S You et al. Comparison of 3 mm × 3 mm versus 6 mm × 6 mm optical coherence tomography angiography scan sizes in the evaluation of non-proliferative diabetic retinopathy. Retina, 39, 259-264(2019).
[12] W Choi, ChoiW, MoultEM, al WaheedNKet, EM Moult, ChoiW, MoultEM, al WaheedNKet, NK Waheed et al. Ultrahigh-speed, swept-source optical coherence tomography angiography in nonexudative age-related macular degeneration with geographic atrophy. Ophthalmology, 122, 2532-2544(2015).
[13] SB Ploner, PlonerSB, MoultEM, al ChoiWet, EM Moult, PlonerSB, MoultEM, al ChoiWet, W Choi et al. Toward quantitative optical coherence tomography angiography: visualizing blood flow speeds in ocular pathology using variable interscan time analysis. Retina, 36, S118-S126(2016).
[14] B Braaf, BraafB, GräfeMGO, al Uribe-PatarroyoNet, MGO Gräfe, BraafB, GräfeMGO, al Uribe-PatarroyoNet, N Uribe-Patarroyo et al. OCT-based velocimetry for blood flow quantification. High Resolution Imaging in Microscopy and Ophthalmology: New Frontiers in Biomedical Optics, 161-179(2019).
[15] J Tokayer, TokayerJ, JiaYL, al DhallaAHet, YL Jia, TokayerJ, JiaYL, al DhallaAHet, AH Dhalla et al. Blood flow velocity quantification using split-spectrum amplitude-decorrelation angiography with optical coherence tomography. Biomed Opt Express, 4, 1909-1924(2013).
[16] JP Su, SuJP, ChandwaniR, al GaoSSet, R Chandwani, SuJP, ChandwaniR, al GaoSSet, SS Gao et al. Calibration of optical coherence tomography angiography with a microfluidic chip. J Biomed Opt, 21, 086015(2016).
[17] WJ Choi, ChoiWJ, QinW, al ChenCLet, W Qin, ChoiWJ, QinW, al ChenCLet, CL Chen et al. Characterizing relationship between optical microangiography signals and capillary flow using microfluidic channels. Biomed Opt Express, 7, 2709-2728(2016).
[18] N Uribe-Patarroyo, Uribe-PatarroyoN, BoumaBE, BE Bouma. Velocity gradients in spatially resolved laser Doppler flowmetry and dynamic light scattering with confocal and coherence gating. Phys Rev E, 94, 022604(2016).
[19] N Uribe-Patarroyo, Uribe-PatarroyoN, VilligerM, BoumaBE, M Villiger, Uribe-PatarroyoN, VilligerM, BoumaBE, BE Bouma. Quantitative technique for robust and noise-tolerant speed measurements based on speckle decorrelation in optical coherence tomography. Opt Express, 22, 24411-24429(2014).
[20] MGO Gräfe, GräfeMGO, NadiarnykhO, JF DeBoer, O Nadiarnykh, GräfeMGO, NadiarnykhO, JF DeBoer, Boer JF De. Optical coherence tomography velocimetry based on decorrelation estimation of phasor pair ratios (DEPPAIR). Biomed Opt Express, 10, 5470-5485(2019).
[21] J Lee, LeeJ, WuWC, al JiangJYet, WC Wu, LeeJ, WuWC, al JiangJYet, JY Jiang et al. Dynamic light scattering optical coherence tomography. Opt Express, 20, 22262-22277(2012).
[22] T Durduran, DurduranT, ChoeR, al BakerWBet, R Choe, DurduranT, ChoeR, al BakerWBet, WB Baker et al. Diffuse optics for tissue monitoring and tomography. Rep Prog Phys, 73, 076701(2010).
[23] X Wei, WeiX, HormelTT, al PiSHet, TT Hormel, WeiX, HormelTT, al PiSHet, SH Pi et al. High dynamic range optical coherence tomography angiography (HDR-OCTA). Biomed Opt Express, 10, 3560-3571(2019).
[24] JL Yang, YangJL, SuJ, al WangJet, J Su, YangJL, SuJ, al WangJet, J Wang et al. Hematocrit dependence of flow signal in optical coherence tomography angiography. Biomed Opt Express, 8, 776-789(2017).
[25] B Braaf, BraafB, VermeerKA, al VienolaKVet, KA Vermeer, BraafB, VermeerKA, al VienolaKVet, KV Vienola et al. Angiography of the retina and the choroid with phase-resolved OCT using interval-optimized backstitched B-scans. Opt Express, 20, 20516-20534(2012).
[26] B Karamata, KaramataB, LaubscherM, al LeuteneggerMet, M Laubscher, KaramataB, LaubscherM, al LeuteneggerMet, M Leutenegger et al. Multiple scattering in optical coherence tomography. I. Investigation and modeling. J Opt Soc Am A Opt Image Sci Vis, 22, 1369-1379(2005).
[27] Y Kaizu, KaizuY, NakaoS, al SodaTet, S Nakao, KaizuY, NakaoS, al SodaTet, T Soda et al. Longer interscan times in OCT angiography detect slower capillary flow in diabetic retinopathy. Ophthalmol Sci, 2, 100181(2022).
[28] LB Liu, LiuLB, GardeckiJA, al NadkarniSKet, JA Gardecki, LiuLB, GardeckiJA, al NadkarniSKet, SK Nadkarni et al. Imaging the subcellular structure of human coronary atherosclerosis using micro–optical coherence tomography. Nat Med, 17, 1010-1014(2011).
[29] S Chen, ChenS, LiuXY, al WangNSet, XY Liu, ChenS, LiuXY, al WangNSet, NS Wang et al. Visualizing micro-anatomical structures of the posterior cornea with micro-optical coherence tomography. Sci Rep, 7, 10752(2017).
[30] J Yi, YiJ, LiuWZ, al ChenSYet, WZ Liu, YiJ, LiuWZ, al ChenSYet, SY Chen et al. Visible light optical coherence tomography measures retinal oxygen metabolic response to systemic oxygenation. Light Sci Appl, 4, e334(2015).
[31] S Chen, ChenS, GeX, al LiuXYet, X Ge, ChenS, GeX, al LiuXYet, XY Liu et al. Understanding optical reflectance contrast for real-time characterization of epithelial precursor lesions. Bioeng Transl Med, 4, e10137(2019).
[32] A Couturier, CouturierA, ReyPA, al ErginayAet, PA Rey, CouturierA, ReyPA, al ErginayAet, A Erginay et al. Widefield OCT-angiography and fluorescein angiography assessments of nonperfusion in diabetic retinopathy and edema treated with anti-vascular endothelial growth factor. Ophthalmology, 126, 1685-1694(2019).
[33] AY Chen, ChenAY, LiuL, al WangJet, L Liu, ChenAY, LiuL, al WangJet, J Wang et al. Measuring glaucomatous focal perfusion loss in the peripapillary retina using OCT angiography. Ophthalmology, 127, 484-491(2020).
[34] DY Cui, CuiDY, LiuXY, al ZhangJet, XY Liu, CuiDY, LiuXY, al ZhangJet, J Zhang et al. Dual spectrometer system with spectral compounding for 1-μm optical coherence tomography in vivo. Opt Lett, 39, 6727-6730(2014).
[35] RDS Shreesha, ShreeshaRDS, JensenM, al Grüner-NielsenLet, M Jensen, ShreeshaRDS, JensenM, al Grüner-NielsenLet, L Grüner-Nielsen et al. Shot-noise limited, supercontinuum-based optical coherence tomography. Light Sci Appl, 10, 133(2021).
[36] KC Liang, LiangKC, LiuXY, al ChenSet, XY Liu, LiangKC, LiuXY, al ChenSet, S Chen et al. Resolution enhancement and realistic speckle recovery with generative adversarial modeling of micro-optical coherence tomography. Biomed Opt Express, 11, 7236-7252(2020).
[37] JQ Hu et al. RSPSSL: a novel high-fidelity Raman spectral preprocessing scheme to enhance biomedical applications and chemical resolution visualization. Light Sci Appl, 13, 52(2024).
[38] X Ge, GeX, ChenS, al LinKet, S Chen, GeX, ChenS, al LinKet, K Lin et al. Deblurring, artifact-free optical coherence tomography with deconvolution-random phase modulation. Opto-Electron Sci, 3, 230020(2024).
[39] International Electrotechnical Commission . Safety of laser products - Part 1: Equipment classification, requirements, and user’s guide. IEC 60825–1 Edition 1.2(2001).
[40] K Schulmeister, SchulmeisterK, GilberR, al SeiserBet, R Gilber, SchulmeisterK, GilberR, al SeiserBet, B Seiser et al. Retinal thermal laser damage thresholds for different beam profiles and scanned exposure. Proc SPIE, 6844, 68441L(2008).
[41] ANSI . American National Standard for safe use of lasers (ANSI136. 1). ANSI 136.1–2000(2000).
[42] SH Yun, YunSH, TearneyGJ, al BoumaBEet, GJ Tearney, YunSH, TearneyGJ, al BoumaBEet, BE Bouma et al. High-speed spectral-domain optical coherence tomography at 1.3 µm wavelength. Opt Express, 11, 3598-3604(2003).
[43] R Leitgeb, LeitgebR, HitzenbergerCK, FercherAF, CK Hitzenberger, LeitgebR, HitzenbergerCK, FercherAF, AF Fercher. Performance of fourier domain vs. time domain optical coherence tomography. Opt Express, 11, 889-894(2003).
[44] L An, AnL, WangRK, RK Wang. In vivo volumetric imaging of vascular perfusion within human retina and choroids with optical micro-angiography. Opt Express, 16, 11438-11452(2008).
[45] M Wojtkowski, WojtkowskiM, SrinivasanVJ, al KoTHet, VJ Srinivasan, WojtkowskiM, SrinivasanVJ, al KoTHet, TH Ko et al. Ultrahigh-resolution, high-speed, Fourier domain optical coherence tomography and methods for dispersion compensation. Opt Express, 12, 2404-2422(2004).
[46] JJ Vos. A theory of retinal burns. Bull Math Biophys, 24, 115-128(1962).
[47] JA Zuclich, ZuclichJA, LundDJ, al EdsallPRet, DJ Lund, ZuclichJA, LundDJ, al EdsallPRet, PR Edsall et al. Laser-induced retinal damage threshold as a function of retinal image size. Proc SPIE, 3591, 335-343(1999).
[48] ES Beatrice, BeatriceES, FrischGD, GD Frisch. Retinal laser damage thresholds as a function of image diameter. Arch Environ Health Int J, 27, 322-326(1973).
[49] ANSI . Z80.36–2021 for Light Hazard Protection for Ophthalmic Instruments(2021).
[50] YL Jia, JiaYL, TanO, al TokayerJet, O Tan, JiaYL, TanO, al TokayerJet, J Tokayer et al. Split-spectrum amplitude-decorrelation angiography with optical coherence tomography. Opt Express, 20, 4710-4725(2012).
[51] AQ Zhang, ZhangAQ, ZhangQQ, al ChenCLet, QQ Zhang, ZhangAQ, ZhangQQ, al ChenCLet, CL Chen et al. Methods and algorithms for optical coherence tomography-based angiography: a review and comparison. J Biomed Opt, 20, 100901(2015).
[52] L An, AnL, QinJ, WangRK, J Qin, AnL, QinJ, WangRK, RK Wang. Ultrahigh sensitive optical microangiography for in vivo imaging of microcirculations within human skin tissue beds. Opt Express, 18, 8220-8228(2010).
[53] T Schmoll, SchmollT, BagheriniaH, RenHG, H Bagherinia, SchmollT, BagheriniaH, RenHG, HG Ren. OCTA flow signal enhancement by reducing residual structural signal. Invest Ophthalmol Visual Sci, 60, PB048(2019).
[54] D Richter, RichterD, FardAM, al StraubJet, AM Fard, RichterD, FardAM, al StraubJet, J Straub et al. Relative retinal flow velocity detection using optical coherence tomography angiography imaging. Biomed Opt Express, 11, 6710-6720(2020).
[55] ZD Chu, ChuZD, LinJ, al GaoCet, J Lin, ChuZD, LinJ, al GaoCet, C Gao et al. Quantitative assessment of the retinal microvasculature using optical coherence tomography angiography. J Biomed Opt, 21, 66008(2016).
Tools
Get Citation
Copy Citation Text
Si Chen, Kan Lin, Xi Chen, Yukun Wang, Chen Hsin Sun, Jia Qu, Xin Ge, Xiaokun Wang, Linbo Liu. Spectrally extended line field optical coherence tomography angiography[J]. Opto-Electronic Advances, 2025, 8(5): 240293