A Mobile-Based Decision Support Tool for Preliminary Eye Anomaly Detection in Community Health Practice

It has been predicted that about 2.2 billion people globally have visual impairment, where half of these cases remain preventable or treatable in resource-constrained regions. Existing mobile health(mHealth) tools are fragmented. They focus on isolated diagnostic functions and require constant internet connectivity. This study presents an integrated, offline-capable mobile decision support tool that combines Snellen visual acuity, Ishihara color, and symptom-based AI analysis into a single application. The proposed tools used the Flutter framework and TensorFlow Lite and utilize a Convolutional Neural Network for on-device inference. These ensure data privacy and functionality in remote areas. A curated dataset of 1,200 clinical records from three different hospitals was used for model development with a pilot evaluation with 50 participants was conducted to validate performance. The study's diagnostic results show a high accuracy of 94%, with a precision of 95.6%, recall of 91.7%, and an F1-score of 93.6%. The Benchmark against existing solutions (Peek Acuity and IDx-DR) illustrates the proposed system's unique multi-modal capabilities and low-cost accessibility. The proposed study application for the community health workers is to bridge the gap between preliminary screening and clinical referral. These offer a scalable solution for eye care in low-resource settings.