JMIR Publications

ABSTRACT

Background:

Low vision substantially impairs reading ability, educational participation, and access to information among children and adolescents. Existing assistive technologies primarily focus on visual magnification and often provide limited support for higher-level cognitive processing during reading tasks.

Objective:

This study aimed to develop and evaluate a virtual reality (VR)-based multimodal intelligent reading aid integrating image enhancement, optical character recognition (OCR), text-to-speech (TTS), and large language model (LLM)-assisted semantic summarization for low-vision adolescents.

Methods:

This study consisted of 2 sequential components. First, a prospective ophthalmic safety evaluation was conducted in 22 healthy adults to assess the short-term ocular safety and tolerability of the VR device. Ophthalmic examinations included tear film break-up time (TBUT), phoria, accommodative function, intraocular pressure (IOP), best-corrected visual acuity (BCVA), and simulator sickness scores before and after device exposure. Second, a within-subject controlled functional evaluation was conducted in 36 low-vision adolescents aged 9–17 years. Participants completed standardized reading tasks under 2 conditions: unaided reading and VR-assisted reading. Reading completion time was evaluated as the primary outcome. Secondary outcomes included user satisfaction and system usability assessed using the System Usability Scale (SUS). Paired-sample t tests were used for statistical analyses.

Results:

In the ophthalmic safety cohort, no significant changes were observed in TBUT, distance phoria, near phoria, accommodative convergence/accommodation ratio, accommodative response, IOP, BCVA, or simulator sickness scores following VR exposure (all P>.05). No device-related adverse events or clinically significant discomfort were reported. In the functional evaluation cohort, the VR-based reading aid significantly improved reading performance compared with unaided reading. Mean reading completion time decreased from 52.11 (SD 8.83) seconds to 26.53 (SD 3.74) seconds (P<.001), representing an approximately 49% reduction in reading time. User satisfaction was significantly higher under the VR-assisted condition than under the unaided condition (85.56 [SD 4.23] vs 36.53 [SD 11.53]; P<.001). The mean SUS score was 71.32 (SD 3.71), exceeding the established benchmark score of 68 and indicating good overall usability and user acceptance.

Conclusions:

The proposed VR-based multimodal intelligent reading aid demonstrated favorable short-term ocular safety, good usability, and significant improvements in reading efficiency and user satisfaction among low-vision adolescents. By integrating immersive visualization, OCR-based text recognition, auditory feedback, and AI-assisted semantic processing into a unified digital health platform, the system may represent a promising approach for enhancing reading accessibility, educational participation, and independent learning in visually impaired populations. Clinical Trial: ClinicalTrials.gov 2024LSPJ164