A Novel Slit Lamp‐Based System for 3D Ultrasound of the Eye

Abstract

Objectives

This study aims to develop and evaluate a novel slit‐lamp‐based system to acquire 3‐dimensional (3D) ultrasound volumes using conventional ultrasound probes.

Methods

A slit‐lamp adapter was 3D printed and connected with a linear actuator and graphical user interface (GUI) system to control movement and create volumes of 2‐dimensional scans. The ultrasound probe contacted the patient's cornea to perform the scan. The GUI software was designed to acquire and process b‐scan images, which were exported to 3D Slicer and manually annotated. The annotated images were used to render 3D renderings of ophthalmic anatomy using MATLAB and 3D Slicer.

Results

The optimal resolution, quality, scan type, and patient comfort were determined through simulations. Higher scan numbers and light contact pressure resulted in sufficient image clarity. Out of 250 completed scans, four 3D ultrasound models were created, depicting diagnoses such as tractional retinal detachment, choroidal hemorrhage, and myopia. These models were compared with traditional imaging methods.

Conclusions

The study successfully created 3D ultrasound scans using a novel slit‐lamp‐based system and manual annotation. The accuracy of the models was comparable to other imaging forms and demonstrated greater details. This study represents a significant step toward an accessible point‐of‐care ultrasound system in ophthalmology.

Abbreviations

CAD

computer‐aided design

CMUT

capacitive micromachined ultrasound transducer

CT

computed tomography

GUI

graphical user interface

IFU

instructions for use

MRI

magnetic resonance imaging

UBM

ultrasound biomicroscopy

Ultrasound imaging has consistently played a vital role in medical diagnostics, offering the advantage of non‐invasive and instant visualization across a range of clinical applications. In ophthalmology, it is particularly useful for examining the eye's internal structures when the view is obstructed by media opacities (eg, dense cataracts or a vitreous hemorrhage). Although ultrasound has been a staple in eye care for many years, its technological advancements have been somewhat stagnant until recent years. The advent of 3‐dimensional ultrasound (3DUS) in the 1990s marked a significant leap forward, enabling more detailed examinations of the eye's anatomy. This advancement has found applications in various medical disciplines such as cardiology, orthopedics, urology, and obstetrics, highlighted by a multitude of research studies. ¹ , ² , ³

The initial implementation of 3DUS involved a bespoke rotational mechanism to spin an ultrasound probe around the surface of the eye in contact with the cornea, creating a 3‐dimensional (3D) representation from multiple 2‐dimensional (2D) B‐scan images, specifically of the eye's posterior section. ⁴ This pioneering effort was unique in its use of rotational movement to manipulate the ultrasound probe. Subsequent research shifted toward linear movements for probe actuation. For instance, Cusumano et al (1998) utilized a linear actuator for scanning, focusing on different ocular conditions such as a choroidal hemorrhage, ciliary body melanoma, and ciliary body detachment. ⁵ This early research paved the way for further investigations into diseases affecting the eye's anterior and posterior segments and the optic nerve.

For anterior segment analysis, 3DUS was effectively employed using a high frequency (38–50 MHz) ultrasound biomicroscopy (UBM) probe to delineate the anatomy of the ciliary body, ⁶ the back surface of the iris, ⁷ and the ciliary body's modifications during eye focusing. ⁸ More recent studies by Helms et al and Minaz et al have applied linear scanning techniques with a high‐frequency 50‐MHz UBM probe to study the anterior segment in various eye conditions, providing critical insights for diagnosing and treating anterior eye disorders. ⁹ , ¹⁰ In the realm of retinal imaging, the sole commercially available 3DUS device (3D i‐scan by Ophthalmic Technologies Inc., Toronto, Ontario) has been instrumental in positioning radioactive plaques for treating choroidal melanomas ¹¹ and measuring the volume of these tumors. ¹² It has also been useful in examining patients with uveitis where the retina was not visible. ¹³ An innovative 20‐MHz 5‐ring annular probe introduced the capability for synthetic focusing across different eye segments, enabling 3D modeling from linear scans with a single probe. ¹⁴ Additionally, 3DUS has been valuable in measuring the optic nerve's diameter behind the eye ¹⁴ and in studying cerebral edema in climbers at high altitudes. ¹⁵ Even probes not specifically designed for ophthalmic use have been applied to assess the eyelid, cornea, iris, entire eye, and blood flow in the temporal artery, ¹⁶ showcasing the versatility of 3DUS in offering detailed volumetric and structural insights into both intraocular and extraocular areas.

Despite the introduction of 3D ophthalmic ultrasound nearly 3 decades ago and the development of 1 commercial device, the technology has not been widely adopted in clinical practice due to a lack of clinical relevance in comparison to other imaging modalities. Furthermore, there has not been any significant advancements in creating an ultrasound probe system that replicates the ease of use of optical imaging or integrates with other ophthalmic tests like keratometry, optical coherence tomography, or fundus photography.

Considering the limited availability of 3DUS in ophthalmology and the commercial failure of previous devices, our study aims to investigate the practicality and efficacy of integrating 3DUS with slit lamp examinations using readily available probes. Currently, pathologic myopia, thyroid eye disease, optic neuritis, multiple sclerosis, and idiopathic intracranial hypertension are a few examples of eye conditions that are commonly evaluated and reconstructed into 3D models through computed tomography (CT) and magnetic resonance imaging (MRI) imaging. ¹⁷ However, these imaging modalities are impractical for routine care due to not being done in office, are more expensive for patients, and results are not available in an expedient time frame. Similar results from CT and MRI are also readily visible on ultrasound imaging. To address the deficiency in point‐of‐care imaging modalities, we have developed a bespoke horizontal actuation mechanism designed to accommodate any ultrasound probe. This device can be attached to a slit lamp and used with both ophthalmic and non‐ophthalmic probes to generate 3D ultrasound images, particularly focusing on the posterior pole of the eye.

Methods

Subjects

This is a cohort study. The study involved patient's ophthalmic examinations in a retina clinic, including patients with tractional retinal detachment, choroidal hemorrhage, and myopia. The main parameters measured were the 3D rendering image quality and detection of features expected in different pathological conditions.

Development of Slit Lamp Adapter

We developed a motorized actuator to vertically, horizontally, and rotationally move ultrasound probes while images were acquired from subject's eyes (Figure 1A). We utilized multiple motorized actuators to linearly drive the ultrasound probe in the x–y plane and rotate the ultrasound probe along its long axis. This adapter was modeled in a 3D computer‐aided design (CAD) program (Autodesk Fusion v.2.0.18441) and the corresponding structure was printed with a 3D printer to fit it onto a Haag Streit slit lamp (Figure 1B).

Figure 1.

The ultrasound actuation system manipulates a prob in a radial, vertical and horizontal path (A) and is positioned on a slit lamp microscope (B). Software controls different actuation parameters while acquiring videos of consecutive B scan image volumes (C) while a subject remains steady in a chin rest (D).

The system consists of 2 main components: an Arduino Uno motor microcontroller and software with GUI front end and python scripts for fine‐tuning actuators and image processing. These 2 components communicate through a serial port and the UART protocol. We developed the GUI using Python Qt (Figure 1C) to control imaging parameters of the motorized slit lamp adapter and efficiently obtain data. This GUI allows fine‐tuning of parameters such as scan speed, degrees of rotation, gaze direction, and axis. The GUI also initiates screen‐acquisition (QtMultimedia module) to save a video of the ultrasound image while the actuator moves the probe along a defined path. At the beginning of each scan, the GUI sends a string containing imaging parameters, including scan speed, degrees of rotation, scanning type, and gaze direction, to the microcontroller. Upon receiving these parameters, the microcontroller initiates motor movements to prepare for scanning. Once preparation is complete, the microcontroller sends a message back to the GUI software, ensuring that video recording starts precisely when scanning begins. When the scan is finished, the microcontroller sends another message to the GUI, indicating that recording has concluded.

The GUI then processes the recorded frames and stores them in the user‐specified format, either as individual pictures or a video file. Additionally, a GIF file is generated to provide a brief preview of the recorded content. This approach enables reproducible and precise control and synchronization of the scanning process, allowing for accurate acquisition of video recordings that capture the desired scanning behavior.

Multiple adaptors were designed to fit different ultrasound probes into the actuator assembly. We evaluated the Accutome B‐scan Plus, from which most of the figures were reconstructed; the Ellex Eye Cubed V3 B‐scan probe; and the Butterfly iQ+ probe, a capacitive micromachined ultrasound transducer (CMUT) commonly provided to medical students for training and used in intensive care and trauma settings were also evaluated.

Image Acquisition and Volume Reconstruction

This research was approved by the institutional review board of the University of California, Irvine and carried out following the principles of the Declaration of Helsinki. Written informed consent was obtained from all participants prior to the commencement of testing. Trained individuals performed ultrasound imaging using the novel device. Subjects were given proparacaine eye drops and were positioned in the slit lamp chin rest. GenTeal Gel (Alcon Laboratories, Inc., Fort Worth, TX, USA) eye lubricant was placed on the ultrasound probe to reduce friction between the probe and patient's eye. The ultrasound probe was placed in contact with the center of the subject's cornea and the automated actuation protocol (horizontal or rotational) was initiated while a B‐scan ultrasound video was acquired. B‐scan imaging volume took 20 seconds to acquire 800 scans. Video frames were isolated as individual JPEG files and stored for 3D volume reconstruction.

We used 3D Slicer (v. 5.6.0), an open source and widely used image analysis platform, to reconstruct the ultrasound scans into 3D models. ¹⁸ The 3D reconstruction occurred in 1 of 2 ways, depending on if the scan was a horizontal or rotational scan.

For horizontal scans, we uploaded individual video frame images into 3D Slicer and utilized the segment editor module to manually annotate each of the frames of the scan. The initial iteration of rotational scans involved a similar process but required a custom python script to rotate the video frames along an origin plane that intersected the curvature of the eye. Individual frames were then manually annotated to delineate the retina, choroid, sclera, optic nerve and extraocular muscles. After the manual annotation, the 3D model was constructed, displayed, and analyzed (Figures 5, 6, 7).

Figure 5.

3D reconstruction of 2 pathological myopia subjects (A and B). These figures exhibited side and oblique posterior views to show the staphyloma (B) and myopic features of the reconstruction. One plane of the subject's OCT HD21 line was also displayed to show relative accuracy of the model (C and D).

Figure 6.

Subject suffering from tractional retinal detachment, with the total retinal detachment with fibrotic membranes observed during surgery demonstrated by arrows (A). The patient had dense nuclear sclerosis, preventing a slit lamp view examination. The 3D reconstructions highlight how the model mimicked the same view during surgery (B–D). An arrow in (A) and (B) distinguishes the same anatomical features.

Figure 7.

3D reconstruction of the eye of a subject with choroidal hemorrhage. Reconstructions based on source ultrasound grayscale image data after masking out unannotated areas (A and B) as well as the reconstruction based on annotation alone (C) recapitulated findings from the aniridic pseudophakic ultra widefield fundus image (D).

The current and second iteration of rotational scans also utilize 3D Slicer to manually annotate the various anatomy in each frame but was then exported into MATLAB to create a 3D reconstruction (Figures 4, 6, and 7A,B). In MATLAB, the scans were reconstructed into 3D volumetric data, focusing solely on the portions of each image that were previously labeled using 3D Slicer. Subsequently, the vol3d function in Matlab was employed to visualize the resulting 3D data.

Figure 4.

Reconstructions based on interpolation from rotationally acquired B‐scan images. A, Reconstructions of right eye in rotational scans using 51, 102, 203, 405, and 809 frames (left to right) are shown. All 809 frames were manually annotated to delineate retina, choroid and optic nerve. Unannotated areas were masked to create 3D models based on source grayscale ultrasound data. B, Synthetic vertically oriented B scan image as outlined by dotted lines in A simulate cross sectional images as if the probe were actuated horizontally.

Results

To determine the optimal resolution, quality, and scan type for 3D reconstructions, a set of simulations was conducted using 2D images reconstructed from both horizontal and rotational line scans at 25, 50, 100, and 200 scan increments (Figure 2). For both horizontal and rotational reconstruction methods, the process began with generating a 3D cubic space where width and height matched the individual scan image dimensions. In horizontal reconstructions, the cubic space length was determined by the scanner's maximum scan distance, with scanned images distributed uniformly along this length. To address the gaps between scanned images, a 1‐dimensional linear interpolation algorithm was applied to aligned pixels using the equation y = y₀ + [(x − x₀) (y₁ − y₀)]/(x₁ − x₀), where y₀ and y₁ represent adjacent original pixel values, x₀ and x₁ denote their spatial positions, and x and y correspond to the interpolating pixel's spatial position and value. The rotational reconstruction followed similar principles but utilized a 3D volumetric space where length represented a complete 360° scan, with images positioned according to inter‐scan angles. The final step involved mapping this volumetric space to a 3D representation of the original rotationally collected data, employing reconstruction techniques adapted from CT scanning methodology as outlined in Lee Feldkamp's 1984 publication “Practical cone‐beam algorithm.” ¹⁹

Figure 2.

Simulation of 2D images reconstructed from horizontal and rotational line scans with interpolation using 25, 50, 100, and 200 line scans on the cross‐section where the blue lines show how the scan is segmented. This 2D simulation demonstrates reconstructions based on a higher number of input line scans leading to more image details.

The simulation results indicated that image clarity and resolution significantly improved with an increasing number of scans, demonstrating the system's capability to generate high‐resolution reconstructions using a limited number of 3D scans (blue lines in Figure 2A) and interpolation between adjacent lines. For rotational line scans of a 2D image and reconstruction with interpolation, the resolution of peripheral areas of the image was lower than the central portions of the image (Figure 2).

To optimize the quality of the scans without compromising patient comfort, we illustrated the impact of scan technique and probe pressure on the ocular imaging area in 3 different scenarios (Figure 3). Figure 3 illustrates the differences between horizontal and rotational ultrasound scanning paths across the eye. In Figure 3A, the horizontal scan path resulted in low‐quality cross‐sectional images at either end due to loss of contact and imaging variation, highlighting the limitations of maintaining scan orthogonality on a curved surface. The rotational scan path showed improved image consistency, as each cross‐sectional scan remained perpendicular to the eye's axis (Figure 3B).

Figure 3.

Horizontal scans showed that hard contact yielded a larger scanned volume but was intolerable to subjects, while light contact was tolerable but had a reduced field of view. Rotational scanning provided good patient comfort and a circular scan of the posterior pole. A, 3D reconstruction of horizontal scans showed slightly larger scanned area with hard contact, with light contact limits marked by red lines. B, 3D reconstruction of the eye from rotational scanning resulted in a perfectly circular image.

Comparing light and hard contact during horizontal scanning showed that although hard contact produced a larger scanned volume, it was intolerable for patients, whereas light contact was tolerable but resulted in a smaller field of view (Figure 3C). Model reconstructions from light and hard contact were compared and the differences were shown in Figure 3D, where red lines indicated the extent of light contact cross sectional scan limits. Light contact rotational scanning, however, offered acceptable patient comfort and generated a circular scan of the posterior pole (Figure 3C). 3D reconstruction of the eye using rotational scanning yielded a circular volume reconstruction.

We evaluated the impact of different numbers of scans in a rotational volume on quality of 3D reconstructions of a healthy eye (Figure 4). Rotational scans processed to produce interpolated views provided sufficient data to reconstruct the cross‐sectional view comparable to a single scan of similar orientation. Note that the cross‐sectional views in Figure 4B are offset from the long axis of the eye and not bisecting the nerve.

Using our slit lamp‐based ultrasound actuation system and data acquired from over 250 ultrasound volumetric scans, we created 3D reconstructions of patients of interest (Figures 5, 6, 7) by manually segmenting retina, choroid, optic nerve and extraocular muscles. Reconstructions were generated using 180 B‐scan frames of a light contact rotational actuation with the probe rotating 180°.

Figure 5 presents 3D reconstructions of 2 myopic patients. The reconstructions displayed the distinct, rounded anatomy characteristic of myopic eyes, with Figure 5B also showing a clear staphyloma. We contrast the 3D models produced by 3D ultrasound with 9 mm HD OCT scans as shown in Figure 5, C and D.

Figure 6 shows ultrasound and surgical views of a diabetic tractional retinal in a patient who had a dense nuclear sclerosis obscuring any view of the fundus. After manually annotating retina, choroid and optic nerve to mask unannotated areas we developed reconstructions using source grayscale ultrasound image data (Figure 6, A and B) as well as the 3D model of annotations alone (Figure 6C). The patient underwent cataract extraction and vitrectomy, and the 3D reconstruction based on annotations (Figure 6C) accurately depicted preretinal fibrotic membranes observed in the surgical video (arrows in Figure 6, C and D).

Figure 7 displays a 3D reconstruction of a patient with a choroidal hemorrhage. Comparison with the aniridic pseudophakic fundus image revealed a clear correlation in the size and location of the choroidal hemorrhage.

Discussion

In this work, we demonstrated the first slit lamp‐based actuation system for acquiring 3DUS volumes using different scan paths. The engineered slit lamp ultrasound adapter and software (Figure 2) exhibited versatility in accommodating various ultrasound probes and facilitating probe manipulation across the eye in multiple directions. This adaptability should enhances the system's applicability across diverse point‐of‐care settings.

While our study focuses on proof‐of‐concept development and validation of a slit lamp‐based actuator for acquiring 3D ultrasound volumes, we acknowledge that this system is currently optimized for cooperative, seated patients and does not support supine imaging. However, kinetic assessments are feasible with this system, as long as the patient can change their gaze. In fact, the slit lamp stabilizes both the patient's head and the probe position, enabling precise and reproducible imaging during controlled eye movements. This facilitates evaluation of dynamic changes while the probe remains in stable contact with the ocular surface. Regarding clinical implementation, it is important to note that the ultrasound probe used with our actuator is not integrated into the actuator itself, and the proposed commercial design would be compatible with existing ultrasound probes already in clinical use. Sterilization and high‐level disinfection would therefore follow the probe manufacturer's existing instructions for use (IFU). Disposable probe covers can also be employed to further reduce contamination risk. The actuator housing holds the probe in a manner that allows full exposure of the distal tip, so only the probe—not the actuator—requires disinfection between patients. While formal sterilization protocols would need to be established in any commercial setting, we do not anticipate this as a limiting factor in clinical translation.

This system was intentionally designed to accommodate a wide range of existing ultrasound probes, including entry‐level and legacy devices, to ensure broad accessibility and compatibility across different clinical settings. For this proof‐of‐concept, we utilized both an entry‐level probe (Accutome B‐scan Plus) and an older generation Ellex Eye Cubed V3 probe. While we acknowledge that higher‐end, multi‐ring annular array probes—such as the Ellex Eye Prime or Lumibird Absolu—offer superior axial resolution and focusing capabilities, these devices often operate at lower frame rates. As a result, acquiring high‐quality 3D volumes may require longer scan times and greater patient stability. Future studies will explore performance using newer annular array technologies to evaluate the trade‐offs between image quality, acquisition time, and ease of use in a clinical setting.

An important consideration in ophthalmic ultrasound is the choice between handheld and slit lamp‐based actuation. Handheld systems offer maximal flexibility in probe positioning and are advantageous for patients who are unable to sit at a slit lamp, such as those in a hospital bed or with limited mobility. However, they rely heavily on operator stability and can introduce variability in scan alignment and pressure, which may result in motion artifacts or inconsistent image quality. In contrast, the slit lamp‐based system offers enhanced reproducibility by stabilizing the patient's head and eliminating motion artifacts introduced by operator hand movement. While axial imaging through the lens is sometimes used, imaging at the peripheral limbus is preferred to avoid lens‐induced artifacts. Both handheld and slit lamp‐based systems permit limbal imaging; however, with the slit lamp, it requires directing the patient's gaze appropriately to align the probe. This approach maintains image quality while allowing peripheral imaging critical for avoiding distortion from dense phakic lenses or intraocular lenses. Future implementations of the system may benefit from including gaze‐guidance or positional feedback to optimize probe placement at the limbus.

For upright scanning with the patient seated at the slit lamp, we recognize that the use of GenTeal gel—while convenient and widely available—has limitations due to its relatively low viscosity. In future applications, a more viscous coupling medium may help reduce slippage and maintain consistent probe contact during the scan. We did not observe any clinical evidence of corneal abrasion, but a formal evaluation of ocular surface safety and comfort is not anticipated to be any different from conventional standard operating procedures in direct‐contact ophthalmic lutrasound. Rapid scan times also may obviate the need for more viscous contact media. Notably, our ability to acquire a complete 3D scan volume in approximately 20 seconds is a key strength of this system and highlights the potential for rapid, point‐of‐care imaging with minimal patient burden. This short acquisition time is particularly advantageous compared to other 3D imaging modalities that require longer patient fixation or immobility.

We identified limitations of rotational volume imaging and optimized image parameters to yield models with higher resolution at the periphery of rotational volumes where physical distance between adjacent scans was maximal. One notable advantage of rotational imaging was found to be in the ability to optimize scan quality while minimizing patient discomfort. We discovered that rotational scans, particularly with light pressure, offered comprehensive coverage of the ocular imaging area without causing pain to patients. This highlighted the practical advantage of rotational scans over horizontal scans. Although high density volume scanning can more easily reproduce high resolution models of the eye, rotational scanning offers a wider field of image view with greater patient comfort. Rotational scanning is likely to be most valuable for rapidly screening patients for abnormalities prior to acquiring higher resolution scans focused on identified abnormalities. Additionally, as frame count increased, the details, information from the images, and smoothness increased in 3D models and simulated vertical scans based on interpolation of multiple rotationally oriented b‐scan images (Figure 4B).

A significant limitation with rotational scans is that they introduce challenges in aligning images for processing as the 2 hemispheres of a 3D volume did not automatically align. Further refinement of the rotational scanning technique by aligning the probe to the center of the patient's eye and enhancing image processing algorithms may mitigate these aberrations and improve reconstructed model quality. Volumetric feature mapping and registration of horizontally acquired volumes may allow for alignment and registration of rotationally acquired volumes by providing alignment landmarks. Manual segmentation of individual images is tedious; however, machine learning‐based segmentation tools could expedite objective segmentation. ²⁰

Manual segmentation was performed by a trained research scientist with experience in ultrasound image interpretation, but not by an expert ophthalmic ultrasonographer. In future studies, a limited number of scans from a scan volume will be precisely segmented in a masked fashion by 2 retina specialists to create a high‐quality annotated dataset for training AI models ²¹ to automate segmentation and reduce interpreter bias.

Our study demonstrated the capability of our approach to resolve structural abnormalities in eyes with various pathologies, such as myopia, choroidal hemorrhage, and tractional retinal detachment. However, further studies with larger and more diverse cohorts are warranted to validate the generalizability of our findings across different eye conditions and patient populations and identify scenarios for clinical utility. This platform offers an additional opportunity to provide longitudinal study of quantifying anatomical changes using the 3D modeling instead of standard percentage involvement or subjective assessment used in 2D scanning. Even without 3D modeling, the ability to standardize and automate image acquisition and use AI‐based segmentation or disease detection could reduce the need for trained ophthalmic ultrasonographers in non‐ophthalmic care settings. This could have greatest utility in emergency rooms where acute vision loss frequently presents and emergency medicine physicians perform ophthalmic ultrasound with non‐ophthalmic probes. The slit lamp mounted system can accommodate any type of ultrasound probe. ²² , ²³

In conclusion, our study presented a promising approach for generating high‐quality 3D reconstructions of ocular structures using ultrasound and a novel slit lamp‐based actuation system. We found that existing ultrasound hardware could be leveraged to produce informative 3D models of intraocular and extraocular anatomy. This methodology could offer potential benefits in visualizing macroscopic disease anatomy often unavailable at the point of care. While our system offered notable successes in scan quality and patient comfort, addressing the identified challenges and limitations would be crucial for further optimizing its utility and clinical applicability.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.