Retinal Clinical Assessment With AI-derived Quantitative Information

AI-derived Retinal Quantification Versus Routine Clinical Interpretation in Ophthalmic Assessment: a Randomized Controlled Trial

This randomized controlled trial evaluates whether providing clinicians with AI-derived quantitative retinal information improves the quality and efficiency of retinal clinical assessment. Participating ophthalmologists and ophthalmology trainees will be randomly assigned to one of two groups. The intervention group will write clinical reports with access to automated quantitative measurements generated from fundus image analysis, including multiple retinal structural and vascular biomarkers. The control group will complete the same reporting tasks using only the original fundus images without AI-generated quantitative information.

All reports produced by both groups will be de-identified and independently evaluated by a separate panel of senior ophthalmologists who are blinded to group allocation. The expert evaluators will assess report accuracy, completeness, clarity, and overall clinical quality using predefined scoring criteria. The study aims to determine whether access to quantitative retinal biomarkers enhances clinicians' reporting performance and reduces reporting time during retinal assessment tasks.

Study Overview

• Status: Not yet recruiting
• Conditions: Epiretinal Membrane; AMD; Macular Hole; Pathological Myopia; Diabetic Retinopathy (DR); Retinal Vein Occlusion (RVO); no Obvious Abnormalities; Cup-to-disc Ratio Bigger Than 0.5
• Intervention / Treatment: Diagnostic test: AI-derived retinal quantitative information-assisted reporting

• Study Type: Observational
• Enrollment (Estimated): 29

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Child; Adult; Older Adult
• Accepts Healthy Volunteers: Yes

• Sampling Method: Non-Probability Sample

Study Population

The study population consists of practicing ophthalmologists and ophthalmology trainees who are responsible for interpreting fundus images and generating clinical reports. These clinicians will be randomly assigned to either the intervention group, which has access to AI-derived quantitative retinal information during report writing, or the control group, which performs report writing using only the original fundus images without AI assistance.

A separate panel of senior ophthalmologists, who are not involved in the reporting task, will serve as blinded expert evaluators. They will independently assess all completed reports based on predefined quality dimensions, including accuracy, completeness, clarity, and consistency of interpretation.

The retinal fundus images used in this study are de-identified clinical images representing a range of normal and abnormal retinal presentations. All images are of sufficient quality for interpretation and contain no patient-identifiable information

Description

Inclusion Criteria:

Clinician Participants (Report Writers)

1. Board-certified ophthalmologists or ophthalmology trainees (registrars or fellows) with clinical experience in interpreting fundus images.
2. Capable of independently completing retinal clinical reports based on fundus photography.
3. Willing and able to participate in the study tasks (report writing) under assigned study conditions.
4. Able to provide informed consent.

Expert Evaluators (Outcome Assessors)

1. Senior ophthalmologists with at least 5 years of post-certification clinical experience.
2. Not involved in the report-writing stage of the study.
3. Willing to evaluate de-identified reports across predefined quality dimensions.
4. Able to provide informed consent.

Fundus Images (Data Inputs)

1. Retinal fundus photographs of sufficient quality for clinical interpretation.
2. Images representing a range of common retinal findings (normal or abnormal).
3. Previously collected, de-identified images with no patient-identifiable information.

Exclusion Criteria:

Clinician Participants

1. Lack of experience in interpreting fundus images (e.g., interns, medical students).
2. Prior involvement in the development, training, or validation of the AI system being tested.
3. Inability to complete reporting tasks due to time constraints or technical limitations.
4. Any condition that may interfere with ability to perform study tasks (e.g., prolonged absence).

Expert Evaluators

1. Participation in the intervention or control reporting arms.
2. Prior exposure to or involvement in development of the AI system.
3. Any conflict of interest affecting impartiality of report quality evaluation.

Fundus Images

1. Poor-quality images with insufficient clarity for interpretation.
2. Images containing artifacts or cropping that prevent accurate segmentation or assessment.
3. Images with any remaining patient identifiers (excluded to maintain confidentiality).

Study Plan

How is the study designed?

Number of groups / cohorts

3

Cohorts and Interventions

Group / Cohort	Intervention / Treatment
AI-derived retinal quantification	Diagnostic test: AI-derived retinal quantitative information-assisted reporting; Clinicians assigned to the intervention arm will complete retinal clinical reports with access to an AI system that provides automated retinal feature quantification. The system generates multiple quantitative retinal biomarkers-including vessel characteristics, optic nerve head metrics, macular indices, and other region-specific structural measurements-derived from automated segmentation of each fundus image. During report writing, clinicians can view these AI-generated quantitative values alongside the image. The system does not provide diagnostic labels, impressions, or textual interpretations; it only supplies numerical measurements intended to support clinicians' assessment. All clinical judgments, narrative descriptions, and final conclusions in the report are made solely by the clinician.
Routine clinical interpretation
Outcome Assessor

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Expert-rated clinical report quality	All clinical reports generated by clinicians in both the AI-assisted and control groups will be anonymized and independently evaluated by a separate panel of senior ophthalmologists who are blinded to group allocation. The expert evaluators will score each report using predefined criteria assessing accuracy, completeness, clarity, consistency with the fundus image, and overall clinical quality. Scores will be recorded using a standardized multi-dimensional rating scale. The primary outcome is the mean overall quality score per report.	Assessed after completion of all reporting tasks (approximately 1-2 weeks per participant)

Collaborators and Investigators

• Sponsor: Beijing Tongren Hospital

Study record dates

Study Major Dates

• Study Start (Estimated): April 15, 2026
• Primary Completion (Estimated): May 15, 2026
• Study Completion (Estimated): May 15, 2026

Study Registration Dates

• First Submitted: December 5, 2025
• First Submitted That Met QC Criteria: December 5, 2025
• First Posted (Actual): December 18, 2025

Study Record Updates

• Last Update Posted (Actual): April 29, 2026
• Last Update Submitted That Met QC Criteria: April 28, 2026
• Last Verified: December 1, 2025

More Information

Terms related to this study

• Additional Relevant MeSH Terms: Endocrine System Diseases; Vascular Diseases; Cardiovascular Diseases; Diabetes Mellitus; Eye Diseases; Diabetic Angiopathies; Diabetes Complications; Refractive Errors; Embolism and Thrombosis; Retinal Diseases; Venous Thrombosis; Thrombosis; Myopia; Diabetic Retinopathy; Retinal Vein Occlusion; Retinal Perforations; Epiretinal Membrane; Myopia, Degenerative
• Other Study ID Numbers: TRECK2018-056-GZ(2022)-07