Evaluating the Diagnostic Accuracy of a Portable, Motorized, and Remotely Controlled Slit Lamp Imaging Adaptor Prototype for Head-Mounted Displays

Abstract

Purpose: The purpose of this study was to validate the performance of a portable and remotely controlled slit lamp imaging adaptor.

Methods: Twenty patients with anterior eye segment conditions participated in a randomized masked clinical trial. Imaging was performed using a Haag-Streit AG, BX 900 slit lamp biomicroscope and a new slit lamp prototype. Three ophthalmologists independently reviewed masked and randomized 2D images from both instruments and conducted physical eye examinations based on these images. Inter- and intra-grader reliability were assessed using kappa statistics, and sensitivity and specificity were determined with reference to the clinical eye examinations performed during the patients’ visits.

Results: The sensitivity and specificity of the evaluations with the prototype were 47.8% and 64.1%. Similarly, the evaluations from the conventional system obtained a sensitivity and specificity of 49.5% and 66.2%. The differences in the sensitivity and specificity between imaging modalities were not statistically significant (P > 0.05). The intra-grader reliability showed moderate to substantial agreement between the systems (κ = 0.522–0.708). The inter-grader reliability also indicated moderate agreement for the evaluations with the conventional system (κ = 0.552) and the prototype (κ = 0.474).

Conclusions: This study presents a new prototype that exhibits diagnostic accuracy on par with conventional slit lamps and moderate reliability. Further studies with larger sample sizes are required to characterize the prototype's performance. However, its remote functionality and accessibility suggest the potential to extend eye care.

Translational Relevance: The development of portable and remotely controlled eye imaging systems will enhance teleophthalmology services and broaden access to eye care at the primary care level.

Introduction

More than two billion people worldwide live with vision problems. This number is expected to double in the next 3 decades. However, the timely recognition and treatment of eye conditions could prevent the development of half of the vision pathology. For instance, in the United States, according to the Centers for Disease Control and Prevention (CDC), approximately 12 million people 40 years old and older have vision impairment. Census data shows that in every 5 people in the United States over 40 years old, 3 have vision problems.¹ Nevertheless, out of the 93 million adults in the United States prone to develop severe vision impairment and blindness, only half of them visit an eye care specialist during the year. In the United States alone, vision impairment is considered 1 of the top 10 causes of work disability. Vision is important for improving an individual's quality of life and productivity. However, vision care is limited by the cost and accessibility to eye specialists and awareness of eye conditions. For instance, in 2018, a survey targeting 50 to 80 years old adults in the United States asked the main reasons for not having an eye examination in 3 or more years. Of the 2013 respondents, 42% stated that they had not experienced any vision problems, 36% answered that it was difficult for them to get the examination, and approximately 48% complained about the cost of care and not having medical insurance that covered vision care.² 

Therefore, to increase screening, diagnosis and awareness of vision problems is essential to develop ease-of-use, portable, and cost-effective screening technologies for eye conditions. These technologies will also facilitate the integration of eye care at a primary healthcare level and increase the adoption of telemedicine visits for evaluating eye conditions in areas with limited access to specialized care. The conventional devices used in ophthalmology are expensive, they must be operated by trained personnel, and they are limited to a clinical setting. Like their clinical counterparts, portable devices available to evaluate eye conditions require the assistance of a technician.^3–10 On the other hand, to assess the general physio-anatomic functions of the eyes, doctors need to perform a standard comprehensive eye examination. A comprehensive eye examination incorporates visual field and visual acuity testing, tonometry, pupillometry, slit lamp biomicroscope, and fundus imaging.¹¹ However, each test requires a specific device or method for examination. Therefore, there is a need to develop portable and standalone devices that can carry multiple eye diagnostic tests and do not require skilled operators. 

Extended reality (XR) headsets in ophthalmology can create a single device to perform multiple screening tests for eye conditions. There are already several self-guided applications for XR headsets to evaluate visual field, color sensitivity, contrast, and dark adaptation, among others.^12–14 However, these devices do not have cameras to image the eye for a clinical examination, and imaging is essential for ocular assessments. Hence, there is a need to develop a camera adaptor for XR headsets to conduct a basic eye evaluation in a single portable device. 

The clinical slit lamp biomicroscope is one of the most used instruments in ophthalmology and a central component of a comprehensive eye examination. During this assessment, eye care professionals evaluate the anterior eye segment structures, which include the cornea, iris, anterior chamber, eye lens, and posterior chamber. Moreover, there are adaptors for the slit lamp to evaluate retina structures and measure intraocular pressure and iridocorneal angle.^15,16 With a slit lamp examination, doctors can detect the presence of several ocular pathologies, including glaucoma and cataracts, two of the leading causes of blindness and vision impairment worldwide. Because of the versatility of the slit lamp biomicroscope, the new imaging adaptor prototype for XR headsets provides the basic functions of slit lamp examination. This adaptor contains slit/diffuse illumination and a variable focus and magnification camera. In addition, the system's positioning and imaging acquisition are remotely controlled from an external computer to facilitate the evaluation of eye conditions in telemedicine visits. 

A previous study conducted by our research group evaluated the usability of the new prototype. In this study, users with experience in eye imaging techniques, such as imaging technicians, optometrists, and ophthalmology fellows, and inexperienced users were assigned to capture a slit cross-sectional view using the new prototype. Researchers compared the task completion time and the image quality between the users. The results showed that both user groups acquired good eye images with similar quality and in about the same time. The next step in this project is to evaluate the diagnostic sensitivity and specificity of the new prototype. To the best of our knowledge, this is the first study that evaluates the reliability, accuracy, and clinical capability of a portable, motorized, and remotely controlled slit lamp imaging prototype for head-mounted displays. A portable and remotely controlled slit imaging system will facilitate and increase screening of ocular conditions, and can revolutionize eye care, particularly in remote and underserved areas. 

Methods

Study Design

This study is designed as a randomized, masked, clinical study to assess the reliability and validity of the images acquired with a portable and remotely controlled anterior eye segment imaging adaptor for head-mounted displays. The University of Miami Institutional Review Board reviewed and approved the study protocol. All ethical guidelines for research involving humans established in the Declaration of Helsinki and the Health Insurance Portability and Accountability Act (HIPAA) were followed. Patients were recruited during their scheduled clinical visit at the Bascom Palmer Eye Institute (BPEI) in Miami, Florida. Patients were included in the study if they had a slit lamp image of the right eye taken with the conventional device on the same day they were imaged with the new prototype. The inclusion criteria were not based on the presence or absence of an eye condition. Participants were eligible to be included in the study if they were over 18 years old and were willing and able to sign the informed consent form. Twenty patients agreed to participate in the study, and only their right eyes were evaluated. 

Following the study protocol, the researcher acquired a 2D diffused illuminated eye image, a 2D image of the slit beam centered in the eye pupil, and a 2D video of the complete imaging session for each patient with the new prototype. In addition, the 2D conventional clinical slit lamp images taken during their medical visit were obtained to compare the image quality and diagnosis accuracy. The model of the clinical slit lamp available was a Haag-Streit AG, BX 900 (Haag-Streit Co., Köniz, Switzerland). The clinical slit lamp biomicroscope was operated by an ophthalmology technician. The 2D images and videos taken with the new prototype and the reference images were randomized and added to a digital eye physical examination form under a secured network. The form had a total of 40 sections, 2 sections per patient. One section corresponded to the images acquired with the clinical slit lamp, and another with images and videos acquired with the new prototype. Sections were randomly added to the form. Three ophthalmologists from BPEI completed a physical eye examination from the videos and images presented in the form. Their responses were compared to assess the inter-grader and intra-rater reliability between the two imaging systems. The researcher utilized the clinical eye assessment completed by the attending doctor during the patients’ visit as the ground truth evaluations to analyze the sensitivity and specificity values. 

On the other hand, after testing, patients received a questionnaire to evaluate the performance of the prototype and their level of comfort when using the system. Patient input is significant because they are the main beneficiaries of developing a portable imaging system. 

Imaging System

The anterior segment imaging adaptor for head-mounted displays utilized in the study is presented in Figure 1. Figure 2 shows how the prototype is positioned on the patient's head. Like the conventional slit lamp, the imaging adaptor provides diffused and slit illumination. The generated slit has a 12 mm fixed length and a variable width (1 mm – 5 mm) that can be manually adjusted. The light source has a warm white color of approximately 3000 to 3300 kelvins (K). In addition, the new prototype utilizes a 64 MP Arducam Hawkeye camera (Arducam, China, Nanjing). This component provides a variable focus from 80 mm to infinity and allows up to 10 times digital magnification. This camera model simplifies the prototype because it does not require additional optical components, such as a telescope system, to acquire a magnified eye image. The camera and the slit light source are fixed at an angle of 25 degrees. However, the slit beam can rotate at its fixed position to sweep the slit beam across the eye. The prototype clips to the HoloLens 2 head-mounted display (Microsoft, Redmond, WA, USA). Three servo 9 g motors are used to control the system's horizontal and lateral position and the sweeping of the slit beam. A Raspberry Pi 4B (Raspberry Pi Foundation, UK, Cambridge) controls all the system's primary operations, including image acquisition, changing illumination settings, and adjusting system position. A custom-made app developed in Python Tkinter serves as the system controller software. The Raspberry Pi unit allows remote access to the system from another computer through the VNC app and supports the Hawkeye camera processing unit. 

Figure 1.

 

Slit lamp imaging prototype mounted on a HoloLens 2.

Figure 2.

 

User wearing the prototype.

Imaging Protocol

Recruited patients had imaging orders for the conventional clinical slit lamp to be completed by a specialized technician in the Imaging Center at the BPEI Miami location. Patients were imaged with the new prototype when their imaging sessions were completed or after the doctors’ evaluations. During testing, the researcher first positioned the HoloLens 2 glasses on the patient's head with the display flipped upward to set the attached slit lamp imaging adaptor at eye level, as seen in Figure 2. Next, using the master control app, the researcher started the camera preview and adjusted the camera focus if needed. Afterward, the light source was turned on, and the research operator used the motor controllers to adjust the position of the slit beam to center it in the eye pupil. The camera's magnification was increased to take a closer look at the ocular structures. Next, the slit was placed in the center of the pupil, and the slit image was taken with the developed camera app. Then, the autonomous slit beam sweeping option was activated to move the slit across the eye in small and continuous steps. Next, the slit illumination was changed to diffuse light to acquire a general view of the eye and external adnexa. The researcher obtained another image with diffuse illumination and applied the continue sweeping option to improve the view of the different eye regions. 

Questionnaire for Patients

After testing, researchers handed a questionnaire to the patients to evaluate the prototype. The questionnaire was developed in Google Forms. The questions follow the Likert scale (1) strongly disagree, (2) partially disagree, (3) neutral, (4) partially agree, and (5) strongly agree. The questions in this form assessed the patients’ level of comfort and safety when using the prototype. It also included questions that evaluated if the weight of the device and the illumination system used were acceptable to the patient. This questionnaire is intended to evaluate the patient's experience with the current prototype to further improve the next system's design. 

Image Grading

The 2D images and videos acquired with the wearable and remotely controlled slit lamp system and those acquired with the conventional slit lamp were randomly added to an eye evaluation form. This assessment was also developed using the Google Forms platform. The form followed the format of the physical eye examination available in the electronic clinical records. The questionnaire contained 40 sections in total. The eye evaluation form was previously described in another paper presented by this research group.¹⁷ Each section alternated between the images from the conventional slit lamp and the images from the new portable slit lamp prototype of different patients. Each question in the section evaluated one of the following ocular structures: external lids and lashes, conjunctiva and sclera, cornea, anterior chamber, iris, and lens. For each eye structure, common pathologies were available in checkboxes to select. The questions also included the option to add other conditions not available in the checkbox choices. The last question at the end of each section asked the reviewers to grade the images’ quality based on the following 5 points scale.⁸ The criteria followed are (1) inadequate for any diagnostic purpose; (2) unable to exclude all emergent findings; (3) only able to exclude emergent findings; (4) not ideal but still able to exclude subtle findings; and (5) ideal quality. Three ophthalmologists from Bascom Palmer Eye Institute received the eye evaluation form to complete an eye examination from the acquired images and videos. The doctors selected to review the images did not examine the patient with the slit lamp during their clinical appointment. 

Statistical Analysis

The eye assessment performed during the patient's clinical visit was used as the ground truth to evaluate and compare the sensitivity and specificity of the eye imaging prototype and the conventional slit lamp. Researchers collected the reviewers’ responses and classified them into six categories: true negatives (TNs); true positives (TPs); false negatives (FNs); false positives (FPs); unable to tell (UT); and other conditions (OCs). TN and TP represented the responses that correctly identified the absence and presence of a condition. FN and FP were assigned to the evaluations that failed to identify a condition's absence and presence of a pathology. In comparison, the UT category included the conditions identified by the reviewers as unable to evaluate from 2D images and videos. The OC category was assigned to the responses where a condition was misidentified. The images’ overall sensitivity and specificity percentage were calculated for two imaging systems, as well as the positive predicted value (PPV) and the negative predictive value (NNV) for both imaging systems were evaluated. These parameters reflect the proportion of positives and negatives that are considered TPs and TNs.¹⁸ The sensitivity and specificity percentage for each ocular structure were also evaluated. A Z-test for two sample proportions was performed to determine if the sensitivity and specificity percentage differences between the two modalities were statistically significant. A P value of 0.05 was established to determine the significance. 

The quality of the images obtained with the prototype and the conventional slit lamp was evaluated by the reviewers based on their ability to provide a diagnosis from the images. The reviewers’ average score of the two modalities was calculated. The image quality scores between the two modalities were compared using a Wilcox-ranked sum test. A P value of 0.05 was assumed to establish significance. 

A Cohen's kappa test was performed to evaluate intra-grader reliability between the two imaging systems. This test measures the reliability between two raters or tests that are evaluating the same phenomenon. Cohen's kappa value (κ) ranges from −1 to +1, where kappa values less than 0 indicate no agreement. Kappa values between 0.01 and 0.20 indicate none to slight agreement, 0.21 to 0.40 are fair, 0.41 to 0.60 are moderate, 0.61 to 0.80 are substantial, and 0.81 to 1.00 is almost perfect agreement.¹⁹ 

On the other hand, a Fleiss’ kappa test was performed to evaluate inter-grader reliability and agreement among the three graders. Unlike Cohen's test, the Fleiss kappa measures the degree of agreement in classification expected by chance by three or more rates. Similarly, the Fleiss kappa (κ) ranges from −1 to +1, where a kappa of −1 indicates that the observers or reviewers disagreed. A kappa of 0 indicates that the agreement was not better than chance. Finally, a kappa closer to 1 represents an agreement increasingly better than chance or that the reviewers agreed on every question.²⁰ The classification scale of the Cohen's kappa also applies to the Fleiss kappa values. 

Results

This study evaluated the diagnostic accuracy and reliability of the portable and remotely controlled slit lamp imaging adaptor for head-mounted displays. The images obtained with the prototype included a diffused illuminated eye image and a slit illuminated image. Examples of the images acquired with the conventional slit lamp and the prototype are presented in Figures 3 to 18. 

Figure 3.

 

Diffuse illuminated eye image from conventional system for patient A.

Figure 4.

 

Diffuse illuminated eye image from prototype for patient A.

Figure 5.

 

Slit illuminated eye image from conventional system for patient A.

Figure 6.

 

Slit illuminated eye image from prototype for patient A.

Figure 7.

 

Diffuse illuminated eye image from conventional system for patient B.

Figure 8.

 

Diffuse illuminated eye image from prototype for patient B.

Figure 9.

 

Slit illuminated eye image from conventional system for patient B.

Figure 10.

 

Slit illuminated eye image from prototype for patient B.

Figure 11.

 

Diffuse illuminated eye image from conventional system for patient C.

Figure 12.

 

Diffuse illuminated eye image from prototype for patient C.

Figure 13.

 

Slit illuminated eye image from conventional system for patient C.

Figure 14.

 

Slit illuminated eye image from prototype for patient C.

Figure 15.

 

Diffuse illuminated eye image from conventional system for patient D.

Figure 16.

 

Diffuse illuminated eye image from prototype for patient D.

Figure 17.

 

Slit illuminated eye image from conventional system for patient D.

Figure 18.

 

Slit illuminated eye image from prototype for patient D.

First, the sensitivity and specificity of the evaluations from the two imaging modalities were compared among each grader. As shown in Table 1, the average sensitivity percentages for the clinical slit lamp and the prototype were 49.5% and 47.8%. At the same time, the average specificity percentages were 66.2% and 64.1%. The sensitivity and specificity percentages obtained with the new prototype increased by 15% in comparison with the results obtained with an older version of the prototype.¹⁷ The average accuracy percentage for the conventional system was 58.9% and for the prototype it was 56.9%. A Z-test for two sample proportions was conducted to determine if the difference in sensitivities between the two modalities was statistically significant. As presented in Table 2, the Z-test showed no statistically significant differences between the two imaging modalities (P > 0.05). The PPV and NPV for the clinical system were 68.9% and 25.6%. Whereas for the prototype, the average PPV was 65.3%, and the NPV was 25.9%. In addition, the Z-test results show no statistical differences in the calculated sensitivity values within graders for each of the evaluated ocular structures (Table 3). A list of conditions identified with both imaging modalities is presented in Table 4. 

Table 1.

 

Overall Sensitivity and Specificity Values for Each Grader

 

Table 2.

 

Two Sample Z-Test Alpha Coefficients for all Evaluations

 

Table 3.

 

Two Sample Z-Tests Alpha Values for Sensitivity Percentage Comparison Across Ocular Structures

 

Table 4.

 

Eye Conditions Identified From the Images by Ocular Structures

 

On the other hand, the intra-grader reliability was calculated using Cohen's kappa test (Table 5). A total of 140 evaluations were compared, which included 20 assessments for each of the 7 eye structures evaluated, between the 2 imaging systems for each grader. This test assessed whether the graders agreed on the clinical assessment between the two imaging modalities. Table 5 shows moderate to substantial agreement between the 2 systems evaluations for each grader, κ1 = 0.522, κ2 = 0.708, and κ3 = 0.605, P < 0.001. 

Table 5.

 

Intra-Grader Reliability: Cohen's Kappa

 

Moreover, Table 5 shows the percentage of TP and TN evaluations agreed upon by each grader. Grader 1 agreed in 17 TP out of the 35 TP evaluations and 34 TN out of the 48 TN evaluations identified in the crosstabulation between the assessments from the conventional slit lamp and the prototype. Grader 2 agreed with 24 TP evaluations and 50 TN evaluations from the 32 TP and 55 TN identified. Whereas grader 3 agreed with 29 TP and 45 TN evaluations from the 35 TP and 52 TN evaluations placed overall. 

The Fleiss's kappa test was performed to analyze the inter-grader reliability, results are presented in Table 6. The kappa values show moderate agreement among the grader's evaluations. When all the evaluations were compared, the kappa value was 0.513, with a 95% confidence interval (CI) between 0.479 and 0.547. Moderate agreement was also found when comparing the ophthalmologist evaluations of the different imaging modalities. The kappa value for the conventional slit lamp was 0.552, with a 95% CI between 0.503 and 0.601. The prototype evaluations obtained a kappa of 0.474 with a 95% CI between 0.426 and 5.22. The kappa values for the rating categories show the highest agreement among the TP, TN, and FP evaluations for the overall assessment, the conventional slit lamp, and the prototype. On the other hand, Fleiss's kappa values show poor to moderate agreement on the OC and UT rating categories, ranging from 0.09 to 0.387 (Table 7). 

Table 6.

 

Intergrader Agreement: Fleiss's Kappa Results

 

Table 7.

 

Individuals Fleiss's Kappa for Rating Categories.

 

The last question in the eye evaluation form asked the ophthalmologist to grade the image quality based on a 5-point scale. The average score for the images from the conventional system was 2.63, and the average score from the prototype was 2.57. Figure 19 shows the boxplots of the image quality score of each grader. The quality score assigned by each grader to the images from the conventional slit lamp and the prototype were compared using the Wilcoxon signed-rank test. The P test values calculated were 0.109, 0.102, and 0.649 for the first, second, and third graders. No significant differences were identified in the image quality of the systems evaluated. 

Figure 19.

 

Boxplots displaying image quality scores for each grader.

After the patients were tested with the prototype, the researcher gave them a survey to evaluate their level of comfort when using the device. The questionnaire consistency was evaluated using Cronbach's alpha. The alpha value obtained was 0.909, which indicates a high level of internal consistency in the seven Likert scale questions. Six of the survey questions in the software were graded above 4, except for the fifth question that showed the average score was 3.9. The average question score was 4.5. Table 8 presents the questions with the calculated average scores. 

Table 8.

 

Averaged Patient's Questionnaire Responses

 

Discussion

This study assessed the diagnostic accuracy of a portable and remotely controlled slit lamp imaging adaptor for head-mounted displays. 

The P values obtained after completing a Z-test of two sample proportions showed no statistically significant difference in the sensitivity and specificity percentages of the evaluations completed by each grader using the 2D images from the prototype and the slit lamp. The analysis of the graders’ evaluations by ocular structures gathered similar results. These outcomes showed that the eye evaluations completed with the images of the prototype have similar sensitivity and specificity as the assessments completed with images from the clinical slit lamp biomicroscope. These results demonstrate that images obtained using a novel portable, head-mounted, and remotely controlled slit lamp prototype exhibit similar performance to those acquired with the costly and table bound clinical slit lamp. 

The new prototype also significantly improved the sensitivity percentage values from the first design. In a previous paper, our research group presented the first prototype of a camera system with diffused illumination attached to a head-mounted display to image the anterior eye segment.¹⁷ The sensitivity and specificity of this prototype were evaluated and compared with the standard method to complete an eye evaluation used in telemedicine visits, via the computer web camera. Results showed that the wearable camera system (33.3% and 78.1%) had significantly higher sensitivity and specificity than the videos recorded with a web camera (14.6% and 56.2%).¹⁷ Part of the discussion in this paper included the need to improve the prototype design to increase diagnostic accuracy because even though it was shown that this device allowed a good examination of the eye than the standard system for telemedicine visits, the new prototype needed to provide accuracy comparable to the standard clinical system. Therefore, the new imaging adaptor presented in this paper included features such as a slit beam illumination system, a remotely controlled imaging process, a motorized system to position the system, and a camera with variable focus and magnification. The sensitivity and specificity percentage of the clinical evaluations increased with the current prototype design, highlighting the positive influence of design changes. 

Moreover, the sensitivity and specificity percentages obtained with this system are like those reported by other studies evaluating other portable anterior eye segment imaging systems. For instance, a research group from the Lion Eye Institute in Perth, Australia, compared digital images from a portable-handheld slit lamp camera with images from a Zeiss slit lamp camera and the clinical assessment.²¹ The average sensitivity percentage from the identified conditions with the digital camera was approximately 39%, and the average with the traditional slit lamp was approximately 35%. Another study assessed the diagnostic accuracy of a smartphone-based camera system and a Nidek VersaCam (Nidek, Fremont, CA, USA) when assessing cornea pathologies.²² The reported sensitivity percentages range for the smartphone and portable ophthalmic cameras were from 54% to 71% and 66% to 75%. It is noteworthy that sensitivity and specificity may increase if the provider can remotely control the device performing a live examination focusing on specific areas of interest. 

On the other hand, a percentage of positive conditions were misidentified by the graders and categorized as OC. Suppose these evaluations are included as TPs because the images allowed the doctors to observe an abnormal condition that required additional information to provide a correct diagnosis. In that case, the sensitivity increases to almost 57% for both imaging modalities. Furthermore, one factor that may have lowered tare sensitivity and specificity percentages of the evaluations from the images is the presence of certain ocular conditions, such as corneal opacities, that prevent the evaluation of the anterior chamber, iris, and lens. For instance, the subjects from Figures 20 and 21 had a positive evaluation for nuclear sclerosis, but the conditions on the cornea did not allow proper evaluation of the lens. In addition, the chart evaluations completed during the clinical visit by the attending doctor are thorough and more sensitive because it allows the doctor to interact with the patient, ask for the symptoms and complaints, and to look at the patient's medical records. These actions lead to a differential diagnosis approach. Conversely, the graders in this study only have the images to determine a diagnosis that may require additional information to differentiate between conditions. 

Figure 20.

 

Diffuse illuminated eye image from conventional system of patient E.

Figure 21.

 

Diffuse illuminated eye image from prototype of patient E.

On the other hand, the analysis of the intra-grader reliability showed moderate agreement between the conventional and prototype evaluations for each grader, with Cohen's kappa ranging from 0.522 to 0.708. The percentage agreement between the different categories is highest for the TPs and TNs. These results supported the Z-test comparing the sensitivity and specificity percentages between the imaging modalities for each grader, where no statistical significance was determined. 

The inter-grader reliability was also assessed using the Fleiss’ kappa test. A kappa value 0.513 shows moderate agreement among graders when all evaluations are assessed. The evaluations for each imaging modality were also compared between graders. The evaluations from the conventional images and the evaluations from the prototype images also resulted in a moderate agreement among the graders. Moreover, Fleiss's kappa for the individual rating categories shows moderate to high agreement among the graders for the TPs, TNs, and FPs. A high percentage of positive and negative conditions were clearly identified by all the graders using only eye images. 

The image quality score assigned by the graders to the images from the two systems evaluated showed no significant differences for each grader. The average score calculated was approximately 3 in the image quality scale provided, indicating that the images can only exclude emergent findings. In a telemedicine setting, a system that allows the evaluation of emergent findings will increase the screening and detection of conditions before an urgent condition develops and permanent changes occur. 

The responses from the patients’ questionnaire show positive results. The patients felt safe and comfortable when using the device for testing. They reported that the light intensity of the system was comfortable as well. The weight of the prototype system was the lowest-ranking question, but this can be addressed in the subsequent design stage. However, overall, they were satisfied with the current prototype. 

The next design stage for the system includes reducing the footprint by using smaller motors and creating a more ergonomic prototype. Our group is also exploring creating a single portable and motorized slit lamp imaging system that can be controlled remotely instead of attaching the current system to a head-mounted display. The new prototype will also include features from the clinical slit lamp biomicroscope, such as rotating the slit beam around the camera system to change the angle of incidence of the slit beam and a better optical method to increase the camera magnification. These design changes will increase the sensitivity and specificity of the system and allow the examination of more ocular pathologies. In addition, with the creation of a motorized slit lamp, the imaging system can be automatized, and scripts can be developed with instructions on how to image the different conditions or ocular structures best, allowing the system to complete an examination without assistance. 

This study evaluated the diagnostic accuracy of a portable and remotely controlled slit lamp imaging adaptor for head-mounted displays. The study compared images from the imaging adaptor with the clinical slit lamp images and the physical eye examinations completed by the attending doctors. These images were graded by three different ophthalmologists and compared with the clinical examination. The eye examinations completed from the images of both imaging systems had similar sensitivity and specificity percentages, and there was a moderate agreement between graders in the evaluations. This imaging system is adapted to a head-mounted display with the end goal of developing a device that can carry multiple eye diagnostics examinations to facilitate telemedicine and increase access of vision are. Currently, there are several applications using virtual reality (VR) technology that test visual function, such as visual fields and visual acuity. However, imaging is a main component of an eye examination, and a necessary assessment in teleophthalmology. Therefore, an eye imaging component was added to the VR headset. In addition, this system is motorized, it can be controlled remotely, and little training is required to operate it, which contributes to lowering vision care cost, and increasing accessibility. 

A limiting factor that could have lowered the sensitivity of the images acquired with the prototype includes the fact that the images taken with the system were standard to all the patients and conditions, only a diffused and a centered slit image were obtained. Different eye conditions are evaluated with different slit lamp biomicroscope settings. Therefore, a future study could explore the accuracy of detecting a specific patient population with a determined condition, such as cataracts or corneal ulcers. Moreover, in this study, we are comparing the new imaging prototype with the clinical standard, but it is also necessary to compare the performance between the developed prototype and a portable or smartphone based slit lamp imaging system. On the other hand, the current prototype is still in a developmental phase, the image and illumination quality need to be optimized to facilitate the evaluation of more eye conditions. 

In conclusion, this project presented a portable slit lamp imaging system that is remotely controlled, and the images acquired allow to examine ocular conditions with similar sensitivity and specificity as the evaluations from images of the traditional slit lamp and other portable systems. The potential of this prototype is creating a portable and autonomous slit lamp imaging system to use in telemedicine or a primary care setting to increase screening of ophthalmic conditions. 

Acknowledgments

This work could not have been completed without the assistance of the team at the Imaging Center at Bascom Palmer Eye Institute, especially Miguel Guerrero and Alejandro Gutierrez. 

Supported by a National Eye Institute Center Core Grant to the University of Miami (P30 EY014801), and Research to Prevent Blindness (RPB). The funding organization had no role in the design or conduct of this research. 

IP is owned by the University of Miami and licensed to Heru. 

Disclosure: A. Diego, None; A.M. Anter, None; G.R. Gameiro, None; M. Matosas, None; G. Mijares, None; M.A. Shousha, Board of Directors for Heru (S, I)