Virtual Reality-based Perceptual Learning Therapy for Visually Impaired Population

Visual impairments significantly impact individuals' activities of daily living (ADL) and quality of life (QoL). While perceptual learning (PL) has shown promise in vision rehabilitation, traditional PL usually involves a protracted training process, and its training effects may sometimes lack ecological validity. This study protocol outlines a two-arm, single-blind, randomized controlled trial to evaluate the efficacy of an innovative virtual reality (VR)-based PL therapy integrated with machine learning algorithms. Twenty participants with visual field deficits will be recruited and randomized in a 1:1 ratio into either the VR training group or a no-intervention control group. The intervention consists of six visits (1-7 days apart), each involving two training tasks (20 - 30 minutes per task). The VR training task is an Ecological Multiple Object Tracking task and a Dual task. Training stimuli are personalized to target each participant's transition zone (1~20 dB) based on the result of clinical visual field test. A Deep Reinforcement Learning (DRL) framework, specifically the Proximal Policy Optimization (PPO) algorithm, is utilized as an adaptive controller to dynamically modulate task difficulty and maintain participants in an optimal cognitive "flow state". The primary outcome is visual field area improvement (measured by Humphrey Field Analyzer). Secondary outcomes include pre- and post-intervention assessments of visual search, balance ability, fear of falling, ADL, and QoL. It is hypothesized that participants in the VR training group will obtain improvements in primary and secondary outcomes.