Adaptive scan intervals in ultrawide-field OCT angiography achieved by a bidirectional interleaved approach

Abstract

Conventional unidirectional scanning rigidly couples the scan interval (t_inter) to sampling density in ultra-wide field optical coherence tomography angiography (UWF-OCTA), which impedes optimal imaging of different blood flow velocities. In this study, we implement an adaptive bidirectional interleaved scanning pattern for UWF-OCTA that enables the adjustment of t_inter while keeping the scan area, sampling density, and total acquisition time constant or optimally tuned. We implemented this pattern on our high-speed (800 kHz/1 MHz) swept-source UWF-OCTA prototypes and assessed image quality across several use cases, including pediatric patients, non-human primates, and rodents. This protocol successfully captured detailed UWF-OCTA images with well-defined microvascular patterns. We observed that, while keeping sampling density and total scan acquisition time constant, a longer t_inter enhances sensitivity to slower retinal blood flow, whereas a shorter t_inter yields more robust, motion-resistant images with clearer choroidal vasculature, as supported by quantitative contrast-to-noise ratio analysis. These findings illustrate that the optimal t_inter is dependent on the specific application and pathology, and validate the adaptive scanning pattern as a useful approach for tailoring UWF-OCTA to diverse clinical needs.

1. Introduction

Optical coherence tomography angiography (OCTA) has emerged as a critical non-invasive imaging modality in ophthalmology, providing depth-resolved visualization of slab-specific retinal and choroidal microvasculature [1–3]. The advent of ultrawide-field (UWF) OCTA imaging offers the potential to bring these capabilities to bear in assessing pathologies in the peripheral retina. This is a crucial improvement as many ocular diseases, such as diabetic retinopathy (DR), retinal vein occlusions, retinitis pigmentosa, and retinopathy of prematurity (ROP) have significant vascular involvement in the periphery [4–10].

However, the widespread clinical adoption of UWF-OCTA is limited. A primary barrier is the extended imaging time required to scan over the large retinal areas. During this extended acquisition, flow signals can be compromised by various processes such as blinks, microsaccades, drift, and bulk motion, leading to disruptive motion artifacts [11,12]. An evident solution to this challenge is to increase the A-line rate by using ultra-fast swept-source lasers to reduce the total acquisition time [13,14].

Laser speed is an oft-discussed component of UWF systems, but another fundamental concern lies in the scanning pattern, which dictates both scan efficiency and the quality of the flow signal [15–17]. OCTA detects motion contrast from blood flow. Variation in flow magnitude can only be detected when data is acquired within a limited scan interval (t_inter) regime where the signal magnitude is linear with flow speed; outside this range, the flow signal becomes either undetectable or saturates [17,18]. Consequently, optimal t_inter varies across different vascular beds with different blood flow speeds (e.g., slow-moving flow in neovascularization versus rapid flow in the choroid).

Conventional OCTA systems, however, often employ a unidirectional raster scanning pattern. In this pattern, t_inter is determined by the B-scan acquisition time (t_B-scan) and the flyback time of the galvanometer after each B-scan, which inherently takes up 10-50% of the scan time acquiring distorted data [14,15,17]. With a fixed laser A-line rate, t_B-scan dictates the number of A-lines per B-scan and the sampling density over a certain field. For instance, extending t_inter to better visualize slow capillary flow requires either increasing t_B-scan, which forces oversampling in most UWF systems, or deliberately increasing the flyback duration to introduce more dead time. Both strategies prolong the total acquisition time and reduce scan efficiency, exacerbated as laser speeds increase. As an example, achieving an isometric UWF-OCTA with extended t_inter of 4 ms using a 1-MHz unidirectional raster scan would require extending the total acquisition time to 48 seconds, which is clinically unviable. An alternative strategy of increasing the number of repeated B-scans to capture longer effective intervals is similarly problematic, as it linearly extends acquisition time.

To address these limitations, we apply an adaptive bidirectional interleaved scanning pattern, building on previous work [16,17]. While earlier implementations successfully increased the scan efficiency and expanded dynamic range for OCTA, their applications as a stable, variable-interval framework for UWF imaging have been limited by the physical constraints of large-angle, high-speed scanning. Here, we successfully translate this pattern to ultra-fast swept-source UWF-OCTA systems, decoupling t_inter from t_B-scan and sampling density. By utilizing both forward and backward sweeps, it eliminates flyback, thereby reducing the total acquisition time. More importantly, it allows repeats to occur across a configurable strip of N adjacent B-scan positions. This design allows the interscan interval (t_inter = N × t_B-scan) to be adjusted simply by changing N, while the scan area, sampling density, and total acquisition time remain constant, which makes UWF-OCTA at these otherwise inaccessible t_inter regimes viable within practical acquisition time.

In this study, we implemented and evaluated the adaptive bidirectional interleaved scanning pattern on a high-speed (800 kHz and 1 MHz) UWF-OCTA prototypes. We first demonstrate the ability to modify the scan interval while maintaining the same ultrawide field of view (FOV) and consistent acquisition time. Then, we show how the adjusted scan intervals affect image quality across various practical scenarios—including unsedated infant with ROP, non-human primate (NHP), and rodents.

2. Methods

2.1. Bidirectional interleaved scanning pattern design

The core of our approach is an adaptive bidirectional interleaved scanning pattern. The fast-axis (typically the X-axis) galvanometer is driven by a symmetrical triangular or sinusoidal waveform, enabling data acquisition during both forward and backward scans and eliminating flyback. To address the fast-axis galvanometer’s rapid acceleration and deceleration at extreme speeds and large angles, Pseudo Derivative Feedback with Feed Forward (PDFF) tuning was performed manually for the digital servo driver (Mach-DSP, ScannerMAX). We ensured no visible pixel offsets exist between the forward and backward scans, by iteratively fine-tuning the non-integrating and integrating servo parameters. Data from backward scans are simply computationally flipped post-acquisition. This bidirectional scanning also reduces mechanical stress on the fast-axis galvanometer by avoiding the sharp return from the sawtooth wave.

The slow-axis (typically Y-axis) galvanometer steps to the next B-scan position after each fast-axis sweep. For OCTA, B-scans are repeated multiple times (e.g., M = 3) at effectively the same location. In our interleaved pattern, instead of M repeats at a single Y-position before advancing, the system scans a “strip” of N adjacent Y-positions. After scanning over these N positions, it returns to the strip's beginning for the second set of scans, repeating M times. t_inter is thus approximately N × t_B-scan. Adjusting N allows t_inter modification (Fig. 1, Table 1). For instance, in NHP imaging (Section 3.2), with t_B-scan=1 ms and N = 2 yields t_inter=2 ms, and N = 4 gives t_inter=4 ms. Crucially, total image acquisition time and sampling density remain consistent regardless of N (as illustrated in Fig. 1 and Table 1).

Fig. 1.

Illustration of two bidirectional interleaved scanning patterns used to achieve adaptive interscan intervals. The total time duration shown runs from the starting time point (t₀) to the end of the 12^th B-scan (t₁₂). (A, C) Scan progressions from t₀ to t₁₂ of the bidirectional interleaved patterns with two (A) and four (C) B-scan positions within each repeating unit. The four distinct B-scan positions are P₁ through P₄. (B, D) Relative voltage signal applied on the slow-axis galvanometer for the patterns illustrated in (A) and (C), respectively. The interscan interval Δt of the pattern shown in (A, B) is half of what is shown in (C, D), which is 2Δt. Despite different scan intervals, both patterns complete their OCTA scans from P_(n) to P_(n + 3) simultaneously.

Table 1. Comparison of unidirectional raster scan and bidirectional interleaved scan when using a 1-MHz laser.

Scanning pattern (3 repeats OCTA)	Interscan interval (tinter)	B-scans/repeating unit (N)	A-lines/B-scan	20 × 20-mm Sampling density	6 × 6-mm Sampling density	Acquisition time
Unidirectional raster scan	2 ms	1	2000	10 µm/pixel	3 µm/pixel	12 s
	4 ms	1	4000	5 µm/pixel	1.5 µm/pixel	48 s
Bidirectional interleaved scan	2 ms	2	500	20 µm/pixel	12 µm/pixel	0.75 s
	4 ms	4	500	20 µm/pixel	12 µm/pixel	0.75 s

2.2. Swept-source OCTA prototypes

This scanning protocol was implemented on our two custom-built swept-source OCTA prototypes. For pediatric retinal imaging, a handheld UWF-OCTA prototype was used [19,20]. It features an 800-kHz swept-source laser (1060-nm center wavelength, 70-nm bandwidth), providing a 140° FOV and 46.4 µm lateral resolution. For animal studies, a modified portable UWF-OCTA system [21] with a 1-MHz A-line rate swept-source laser was employed. When imaging rhesus macaques, it achieved a 105° FOV (∼20 × 20-mm retinal area, 40-µm lateral resolution). For rats, it provided a 108° FOV (∼5.2 × 5.2-mm area, 8.9-µm lateral resolution).

2.3. Study subjects

The ROP pediatric patient was enrolled at the Casey Eye Institute of Oregon Health & Science University (OHSU). Written informed consent from the parent/guardian was obtained prior to initiating the study. The research was conducted with approval from the OHSU Institutional Review Board (IRB), in compliance with the guidelines outlined in the Declaration of Helsinki.

For the NHP study, a rhesus macaque (Macaca mulatta) from the Oregon National Primate Research Center (ONPRC) colony was imaged under anesthesia. The subject was a genetic model of Bardet-Biedl syndrome (BBS7) [22]. The specific eye imaged in this study had previously received Adeno-Associated Virus (AAV) gene therapy [23,24], resulting partial correction of the retinal phenotype.

In the rodent study, the subjects were Brown Norway rats. The oxygen-induced retinopathy (OIR) model was induced using a modified 50%/10% oxygen exposure paradigm to model ROP [21,25].

All procedures involving animal experiments were reviewed, approved, and under the supervision of the Institutional Animal Care and Use Committee (IACUC) of OHSU, and were performed in compliance with the National Institutes of Health Guide for Care and Use of Laboratory Animals and the Association for Research in Vision and Ophthalmology (ARVO) Statement for the use of Animals in Ophthalmic and Vision Research.

2.4. Imaging protocols

To enable direct comparison, OCTA data for the different t_inter settings were acquired back-to-back in each subject, ensuring that the scanning locations for each test were virtually identical. The specific imaging parameters for each subject are provided in Table 1.

For the ROP study, the 800 kHz handheld UWF-OCTA was used to image an infant in the Neonatal Intensive Care Unit without anesthesia. Two t_inter settings were tested: 4-ms (N = 4) and 8-ms (N = 8). Each volume has 800 A-lines per B-scan, 800 B-scan locations, and M = 3 repeats. With t_B-scan=1 ms, the total acquisition time was 2.4 seconds.

The 1 MHz UWF-OCTA prototype was used to image both NHP and rodent subjects under different settings (Table 2). For the NHP study, sampling density was approximately 20 µm/pixel (1000 A-lines/B-scan). Tested t_inter values included 2 ms, 4 ms, and 8 ms, with a total image time of 3 seconds (M = 3). The rhesus macaque was prepared for imaging using intramuscular Telazol (3.5–5.0 mg/kg) for anesthetization, followed by intubation and maintenance on a mixture of isoflurane (1–2%) and pure oxygen. Prior to the scan, 1% tropicamide and 2.5% phenylephrine drops were applied to fully dilate the pupils.

Table 2. Imaging protocols for different subjects.

Subject	FOV (scan angle)	A-lines/B-scan	B-scan locations	t B-scan	Tested tinter	Repeats	Acquisition time
ROP	140°	800	800	1 ms	4 ms / 8 ms	3	2.4 s
NHP	105°	1000	1000	1 ms	2 ms / 4 ms / 8 ms	3	3 s
Rat	108°	1200	1200	1.2 ms	2.4 ms / 4.8 ms / 9.6 ms	3	4.3 s
Rat (Choroid)	62°	800	800	0.8 ms	1.6 ms / 3.2 ms	8	5.2 s

For the rodent studies, the sampling density of the 108° UWF-OCTA scan was set to approximately 4.3 µm/pixel (1200 A-lines/B-scan; near the Nyquist frequency for lateral resolution). Tested t_inter for this pattern were 2.4 ms, 4.8 ms, and 9.6 ms, with a total scan time of 4.3 seconds (M = 3). To further explore the optimal visualization of the choroid, we implemented an 8-repeat B-scans focused on the central 3 × 3 mm area (62° FOV) sampled at 3.75 µm/pixel (800 A-lines/B-scan), with a total acquisition time of 5.2 seconds. Intervals of 1.6 ms and 3.2 ms were tested in this protocol. Pre-imaging preparation included 1% tropicamide for mydriasis and lubricant gel to maintain optical clarity. During the scan, subjects remained under 2.5% isoflurane anesthesia, with body temperature sustained by a warm water blanket (37°C).

2.5. Image processing and analysis

The acquired OCT/OCTA signal was processed and displayed in real-time with custom-developed software [26], accelerated by a NVIDIA RTX 4070 Ti SUPER graphics card. Post-acquisition, backward fast-axis scans were flipped. Numerical dispersion compensation was performed for different subjects to correct the differences in group velocity dispersion. Repeated B-scans were processed for OCTA using split-spectrum amplitude decorrelation angiography (SSADA) [27]. Different retinal layers were automatically segmented via a deep learning network [28]. The en face OCTA projections of the inner retina were generated from the slab between the inner limiting membrane and the outer nuclear layer, while the choroidal layer was generated from the retinal pigment epithelium to the bottom of the imaging depth.

Angiograms from different t_inter values were qualitatively compared for vascular detail, background, neovascularization visibility, and motion artifacts. To quantitatively evaluate the impact of the interscan interval on angiogram signal and background noise, the contrast-to-noise ratio (CNR) was calculated for the acquired angiograms. To ensure a standardized comparison across different t_inter settings, the back-to-back acquisitions were spatially co-registered. Within a given field of view, targeted regions of interest (ROI) were manually delineated, encompassing distinct vascular segments to determine the mean signal intensity ( $\overline{I_s}$ ). Corresponding avascular regions within the inter-capillary spaces, free from motion artifact striping, were selected to determine the mean background intensity ( $\overline{I_{bk}}$ ) and its standard deviation ( ${\sigma }_{bk}$ ). The CNR was then calculated using the standard formula [29]:

$$CNR=\frac{\overline{I_s}-\overline{I_{bk}}}{{\sigma }_{bk}}$$

3. Results

The adaptive scanning pattern enabled flexible adjustment of t_inter values for angiograms. For each subject, these scans were performed with constant total imaging time and sampling density. The resulting differences in angiogram quality for each application are presented below.

3.1. ROP pediatric patient

Figure 2 shows representative UWF-OCTA images from a ROP eye, comparing 4-ms and 8-ms t_inter. The en face OCTA of the inner retina generated from both intervals displayed similar vasculatures in regions with normal circulations. In the area of the neovascular ridge, however, the 8-ms t_inter provided a more prominent depiction (Fig. 2(F)), whereas the 4-ms t_inter exhibited lower apparent contrast of the ridge neovascularization, though it was still discernible.

Fig. 2.

Comparison of 140° OCTA images generated from 4-ms and 8-ms scan intervals in a ROP patient. (A, B) Inner retina en face projections from the 4-ms and 8-ms t_inter, respectively. The ridge area marked by the blue boxes is enlarged in (E) and (F), and blue arrows mark the locations of neovascularization. Neovascularization is difficult to visually ascertain in the image acquired with a 4-ms t_inter scan pattern (E) compared to the 8-ms t_inter scan pattern, possibly because of slow blood flow speed in these vessels. However, the motion artifacts are more conspicuous for this scan interval. (C, D) Choroid en face projections from the 4-ms and 8-ms t_inter, respectively. The blue boxes highlight an area with dense choroidal vasculature, enlarged in (G) and (H). The 8-ms t_inter zoomed view (H) appears saturated and indistinct with dilated vessels, whereas in (G) the 4-ms t_inter preserves sharp vessel boundaries and clearly differentiated intervascular spaces.

A clear difference was also observed in the choroid. The 4-ms t_inter provided a sharper image with better differentiation of choroidal intervascular space. Due to a higher blood flow speed in the choroid, the 8-ms t_inter failed to provide any benefit of increased vessel sensitivity as it did in retina, but the vessels appeared indistinct and saturated (Fig. 2(H)). Besides that, the 8-ms t_inter also exhibited higher susceptibility to background noise and motion artifacts compared to the 4-ms t_inter, especially in the choroidal projections, manifesting bright and distorted stripes (Fig. 2(D)). Overall, the 8-ms t_inter did a better job in highlighting subtle neovascularization, while the 4-ms t_inter could provide a more stable and robust performance against motion and excellent delineation of choroidal vessels while maintaining decent visualization of ROP induced neovascularization.

3.2. Non-human primate (NHP) imaging

UWF-OCTA images of the retina and choroid of a rhesus macaque were acquired using 2-ms, 4-ms, and 8-ms t_inter. For the inner retina, all three intervals provided good visualization of the vascular network across the 105° FOV, as shown in the images from the subject (Fig. 3). The 8-ms t_inter image showed a slightly noisier background (Fig. 3(C)). Closer inspection (e.g., foveal avascular zone) revealed marginal improvements in fine capillary delineation and vessel connectivity with the 4-ms t_inter (Figs. 3(B, E)), while the 2-ms t_inter images presented the cleanest background and clearest inter-capillary spaces (Figs. 3(A, D)).

Fig. 3.

Comparison of 105° retinal OCTA images generated from 2-ms, 4-ms, and 8-ms scan intervals in a rhesus macaque. (A-C) Retinal en face OCTA projections. Central macular regions marked by white boxes are enlarged in (D-F), respectively. All intervals provided good visualization of the overall vascular network. The 2-ms t_inter images (A, D) presented the cleanest background and clearest inter-capillary spaces. The 4-ms t_inter (B, E) has more apparent background noise and offered a marginal improvement in fine capillary delineation, as indicated in the locations marked by the white and green arrows, showing improved vascular connectivity and less fragmentation compared to the 2-ms t_inter. The 8-ms t_inter (C, F) exhibited the most background noise, making the capillaries harder to identify relative to the shorter t_inter counterparts.

In the choroid, visualizations of structures like vortex veins were possible with the UWF-OCTA system (Fig. 4) from the same subject. Results evidently showed an interval-dependent trend in vessel definition and artifacts. The 2-ms t_inter scans provided the sharpest vessel definition and the most clearly defined intervascular spaces within small-to-medium choroidal vessels (Figs. 4(A, D)). With increasing t_inter, choroidal vessels exhibited an apparent, progressive dilation or blurring (Figs. 4(B, C, E, F)). This artifact was consistent with the general observation in choroidal visualization, also seen in the ROP subject.

Fig. 4.

Comparison of 105° choroidal OCTA images generated from 2-ms, 4-ms, and 8-ms scan intervals in a rhesus macaque. (A-C) Choroidal en face OCTA projections. Areas with dense choroidal vasculature marked by blue boxes are enlarged in (D-F), respectively. (G-I) Corresponding cross-sectional OCT B-scans in the locations marked by the green dashes, with OCTA flow signal overlay, demonstrating true flow signal within the deep choroidal vessels. The 2-ms interval (A, D) provided the sharpest vessel definition and the most clearly defined intervascular spaces. With increasing t_inter, choroidal vessels exhibited an apparent, progressive blurring and dilation (B, E and C, F).

3.3. Healthy adult rat

We first imaged six healthy adult rats with the 108° UWF scanning protocol using 2.4-ms, 4.8-ms, and 9.6-ms t_inter. In a representative subject’s inner retina (Fig. 5), all three intervals yielded largely comparable angiograms, all showing good depictions of the dense retinal vasculature. Upon magnification, subtle enhancements in capillary definition were observed in localized regions with 4.8-ms or 9.6-ms t_inter compared to 2.4-ms t_inter. The marginal gains are likely limited by the rodent’s inherently faster blood flow [30,31]. The 9.6-ms t_inter images, however, were visibly noisier and showed obvious strip-like motion artifacts. This qualitative observation was corroborated by quantitative analysis. The CNR of the vascular networks degraded as t_inter extended. The shortest t_inter of 2.4 ms yielded the highest CNR at 30, which decreased to 25 at the 4.8-ms t_inter, and dropped substantially to 18 at the longest 9.6-ms t_inter.

Fig. 5.

Comparison of 108° retinal OCTA images generated from 2.4-ms, 4.8-ms, and 9.6-ms scan intervals in a healthy adult rat. (A-C) Retinal en face OCTA projections. Areas with dense capillary beds marked by white boxes are enlarged in (D-F), respectively. All three t_inter provided clear depictions of the dense retinal vasculature. Slight localized enhancements in capillary visualization can be observed at 4.8-ms and 9.6-ms compared to 2.4-ms, as indicated by the blue arrows. However, the 9.6-ms t_inter images (C) were visibly noisier and exhibited obvious strip-like motion artifacts.

A separate protocol, using 8 B-scan repeats (M = 8) focused on the central 3 × 3 mm (62° FOV), was then used to explore choroidal OCTA performance with 1.6-ms and 3.2-ms t_inter (Fig. 6). With this high-repeat scan, speckle noise was minimized, and intervascular spaces within small-to-medium choroidal vessels were clearly depicted. A similar trend to what we saw in the ROP patient and NHP was observed, where the shorter 1.6-ms t_inter produced sharper, better-differentiated intervascular spaces in the choroid compared to the 3.2-ms t_inter. Quantitative analysis also confirmed this visual assessment. The 1.6-ms t_inter yielded a CNR of 8.5, which decreased by more than half to 3.6 when t_inter was extended to 3.2 ms.

Fig. 6.

Comparison of 62° choroidal OCTA images generated from 1.6-ms, and 3.2-ms scan intervals in a healthy adult rat. (A, B) Choroidal en face OCTA projections. Areas with dense choroidal vasculatures marked by white boxes are enlarged in (C, D), respectively. Both intervals clearly depicted intervascular spaces within small-to-medium choroidal vessels. The shorter 1.6-ms t_inter (A, C) produced sharper and more well-differentiated intervascular spaces compared to the 3.2-ms t_inter (B, D).

3.4. Oxygen-induced retinopathy (OIR) rat

Next, we conducted bidirectional interleaved scans in four OIR rats at P22. While it is reasonable to expect that longer scan intervals would provide superior sensitivity to slow-flow neovascularization of the OIR rat, we observed that the 2.4-ms t_inter was sufficient to visualize all neovascular structures that are discernible in the longer t_inter. In the representative images, the 4.8-ms and 9.6-ms t_inter showed slightly improved delineation and capillary connectivity in the periphery (Fig. 7). However, we again observed that motion artifacts were more prevalent in the longer-scan-interval images. Not only that, in the areas where motion artifacts occurred, the 9.6-ms t_inter had a much more distracting bright band that washed out the vascular details, whereas in the 2.4-ms t_inter OCTA the general shape of the neovascularization is still apparent (Fig. 7(D)).

Fig. 7.

Comparison of 108° retinal en face OCTA images of inner retina generated from 2.4-ms, 4.8-ms, and 9.6-ms scan intervals in a P22 OIR rat. (A-C) Inner retina en face OCTA projections. Areas marked by white boxes are enlarged in (D-F), respectively. Blue arrows mark the locations of neovascularization; green boxes indicate area where motion artifacts occur. The longer scan intervals (4.8-ms and 9.6-ms) provided slightly better delineation and connectivity of peripheral vasculature than the 2.4-ms interval. However, motion artifacts were more prominent at longer intervals. Notably, the artifacts at 9.6-ms (F) appeared as distracting bright bands that washed out vascular detail, whereas the artifacts at 2.4-ms (D) were less detrimental, often leaving the general shape of the neovascularization still discernible.

4. Discussion

This study successfully demonstrated an efficient bidirectional interleaved scanning pattern for UWF-OCTA that eliminates flyback and allows for the adaptive adjustment of t_inter without altering the total image acquisition time or the spatial sampling density. This capability directly addresses a limitation of conventional unidirectional scanning patterns, where t_inter is typically tied with t_B-scan and laser operational speed, forcing undesirable compromises among flow sensitivity, sampling parameters, and total scan duration.

By eliminating flyback time, the bidirectional scan ensures that more of the laser’s operational duty cycle is used for productive data acquisition, effectively reducing the acquisition time compared to unidirectional raster scanning. This efficiency is particularly vital for UWF imaging, which demands high data throughput. When combined with the ultra-high speed of modern swept-source lasers (such as the 800 kHz and 1 MHz A-line rate lasers used in our prototypes), this efficiency enabled comprehensive UWF-OCTA volumes to be acquired in just a few seconds (as demonstrated in this work, approximately 2.4 seconds for ROP infants, 3 seconds for macaques, and 4.3 seconds for rats). Such rapid acquisition times are crucial for minimizing motion artifacts across the extended scan area and improving patient tolerance and cooperation, particularly in pediatric populations or other subjects who may have difficulty maintaining stable fixation.

A principal advantage of this scanning methodology is the decoupling of t_inter from sampling density, enabling the optimization of scan intervals for specific anatomical targets in a clinically practical manner. In conventional unidirectional scanning designed for continuous acquisition, increasing t_inter to detect slow flow necessitates a proportional increase in t_B-scan, forcing spatial oversampling. Our bidirectional interleaved approach resolves this by adjusting t_inter via the number of interleaved positions (N), independent of sampling density. This flexibility is essential because optimal visualization depends heavily on the target vessels’ flow velocities. For instance, in our ROP study, extending t_inter to 8 ms enhanced sensitivity to the slow flow of neovascular tufts, a finding consistent with OCTA decorrelation theory [17,18,32–34].

However, longer intervals with their increased sensitivity could also render the acquisition vulnerable to motion, which can corrupt the flow signal across the entire interleaved strip of N B-scans. While a shorter scan interval is not immune to motion artifacts, the penalty to angiogram quality is greatly reduced (as suggested by the results from the OIR rat). The detrimental impact of longer intervals was particularly noticeable in choroidal imaging, where factors such as higher pulsatile flow and speckle noise might contribute to greater signal variability. Notably, our results suggest that for demanding in vivo retinal imaging, where motion artifacts often dominate overall image quality, the penalties of longer intervals generally outweigh their marginal benefits for flow sensitivity. In a clinical context, these severe motion artifacts can directly impede accurate OCTA evaluations by obscuring the true boundaries of neovascularization, introducing false-positive flow signals, and severely compromising the reliability of automated quantitative measurements such as vessel density or non-perfusion zone metrics. Nevertheless, the ability to efficiently access longer t_inter without inflating the A-line density beyond the necessary spatial resolution holds significant potential beyond our current tested cases. For instance, longer intervals could be highly advantageous in applications with less physiological motion, such as in vivo cerebral OCTA imaging. Ultimately, the choice of t_inter is based on priority and application. In a clinical setting, a shorter t_inter might offer the reliability needed for routine screening, while a longer t_inter can be considered as a supplemental acquisition if slow-flow detail is paramount. Our adaptive method provides a practical framework for navigating these trade-offs.

This study has certain limitations. First, while our method decouples t_inter from the sampling density, t_B-scan itself remains constrained by the galvanometer scanner's mechanical inertia, particularly at the large angles required for UWF imaging. Furthermore, operating the scanner bidirectionally introduces inherent hardware calibration complexities demanding rigorous tuning of the fast-axis galvanometer to maintain spatial alignment of forward and backward scans if not relying on software registration. In our experience, the maximum stable operational speed for the UWF scan angles (105°–140°) is approximately 1 kHz (t_B-scan = 1 ms). While this suffices for near-Nyquist sampling in our current prototypes it presents a bottleneck for further increasing scanning speed in UWF systems or lowering t_inter. This physical limitation directly influenced our protocol: to test a sub-millisecond t_B-scan for rat’s choroidal imaging, we were compelled to reduce the FOV to maintain scanner stability, resulting in oversampling. Second, while we incorporated quantitative CNR to validate noise profiles in the healthy rats, the limited sample sizes for the human and non-human primate subjects, combined with the inherently lower overall image quality or higher motion in these demanding in vivo scenarios, render rigorous quantification impractical. Therefore, the assessment of image quality and flow detection was mainly qualitative. However, we believe the conclusion from our qualitative comparison remains representative of real-world scenarios in which images may be inspected or reviewed by clinicians under time constraints and rely on visual interpretability above all else. Furthermore, while we have demonstrated the principle of our adaptive scanning approach in four distinct applications, further validation across a broader range of ocular pathologies and diverse patient populations is warranted to fully establish its clinical utility.

5. Conclusion

We have implemented and validated a bidirectional interleaved scanning pattern for UWF-OCTA that eliminates flyback and features an adaptive scan interval. This capability allows the OCTA scan interval to be dynamically adjusted to suit specific imaging requirements, such as enhancing sensitivity to slow neovascular flow in diseases like ROP, or prioritizing high reliability against motion for visualizing structure with higher blood flow like the choroid, all without altering the total acquisition time or the image sampling density. This flexibility is highly advantageous across diverse applications, from detailed vascular research in animal models to demanding clinical imaging situations such as ROP screening in unsedated infants. By successfully isolating t_inter as an independent variable, this framework provides a vital, practical platform for guiding clinicians and industry manufacturers in optimizing UWF-OCTA protocols. Allowing users to balance flow sensitivity while minimizing motion artifacts within a consistently short acquisition timeframe represents a significant advancement towards optimizing UWF-OCTA broader clinical adoption and targeted pathological research.

Funding

National Institutes of Health https://ror.org/01cwqze88 ( R01 EY036429, R01 EY035410, R01 EY024544, R01 EY027833, R01 EY031394, R01 HD107494, R43EY036781, R01 EY035309, P51 OD011092, P30 EY010572, T32 EY023211, UL1TR002369); the Jennie P. Weeks Endowed Fund; the Malcolm M. Marquis, MD Endowed Fund for Innovation; Oregon Health & Science University https://ror.org/009avj582 ( an unrestricted departmental funding grant); Research to Prevent Blindness https://ror.org/04drjs621 ( Dr. H. James and Carole Free Catalyst Award, Stein Innovation Award); Edward N. & Della L. Thome Memorial Foundation Award; Bright Focus Foundation ( G2020168, M20230081).

Disclosures

Shuibin Ni: Siloam Vision (P); Tristan T. Hormel: Ifocus imaging (I); Yukun Guo: Optovue/Visionix (P), Genentech/Roche (P, R), Ifocus Imaging (P); J. Peter Campbell: Siloam Vision (O, P), Genentech (R); Yifan Jian: Siloam Vision (F, P); Yali Jia: Optovue/Visionx (P,R), Genetech (P, R, F), Ifocus Imaging (I, P), Boehringer Ingelheim (C), Kugler (R). These potential conflicts of interest have been reviewed and managed by OHSU. Other authors declare no conflicts of interest related to this article.

Data availability

Data underlying the results presented in this paper are not publicly available at this time but maybe obtained from the authors upon reasonable request.