# NIH Reporter #1R34EY038270-01A1

Deep Learning for low Vision Enhancement (DELVE)

**Recipient:** Not Specified

**Award Amount:** $542,346.00
**Ceiling:** $542,346.00

**Awarded:** September 10, 2026

**Identifier:** 1R34EY038270-01A1

This award funds the DELVE project, which aims to use deep learning and natural language processing on electronic health records to improve low vision rehabilitation services. The project will establish a multicenter network, harmonize data, validate prediction models, and develop tools to enhance patient care for visually impaired adults.

### Description

Low vision rehabilitation (LVR) can improve functional ability and quality of life for people with visual impairment, but referral rates for LVR are inconsistent and can be extraordinarily low. Current guidelines for LVR referral are too broad and there is no system to identify candidates for LVR to gain value from the services. Providers are unable to predict which patients are likely to benefit from LVR, given the variety of LVR interventional options and diverse characteristics of patients with low vision. Thus, there is a critical need to understand the provision of LVR services and develop algorithms that can identify patients who may be candidates for LVR in an automated manner to facilitate appropriate, timely referrals. To achieve that long-term goal, the project proposes to use recent advances in artificial intelligence and natural language processing that have enabled analysis of data in electronic health records (EHR), including clinical findings embedded in unstructured free-text notes, containing the essence of medical documentation with rich information. This R34 planning grant will establish the Deep Learning for low Vision Enhancement (DELVE) multicenter network of LVR providers at seven clinical practice sites who will coordinate and harmonize their EHR data for future research involving deep learning methods to enhance LVR practices. The project aims to create the DELVE multicenter EHR datasets, develop tools for data management and analysis, and finalize future clinical study protocols, including a manual of operations for data harmonization across sites and deep learning model validation. The initial focus is to validate and refine deep learning prediction models at two ophthalmologic academic centers to determine visual prognosis (predicting whether vision will improve to better than 20/40 within 1 year). Subsequently, the project will develop and evaluate deep learning models to predict factors related to LVR provider recommendations and patients’ acquisition of visual aids for television viewing or near reading. Findings from two ophthalmologic academic centers will serve as feasibility data to refine methods for future multisite studies. The project also aims to develop novel systems and multicenter DELVE databases to utilize clinical information and criteria generated from deep learning model data. Future work beyond this R34 will validate the algorithms for applicability to other optometric LVR practices. The long-term initiative is to leverage deep learning techniques to identify best practices and gaps in LVR services, drive new intervention development, and significantly enhance patient care for visually-impaired adults.

[View original record](https://reporter.nih.gov/search/0C83gI3MXki4-tEhFD4UKQ/project-details/11455191)