International Journal of Scientific Research and Engineering Development

Retinal diseases such as diabetic retinopathy (DR), glaucoma, and age-related macular degeneration (AMD) are leading causes of preventable blindness worldwide, yet population-scale screening remains constrained by the limited availability of trained ophthalmologists, particularly in primary-care and rural settings. Single-agent deep learning classifiers, while effective at image-level grading, conflate image-quality assessment, lesion detection, severity classification, referral triage, and patient communication inside one undifferentiated pipeline, making them brittle to poor-quality captures and opaque to the clinicians who must act on their output. This paper proposes a multi-agent architecture — the Retinal Screening and Referral Agent (RSRA) — that decomposes early screening across six cooperating agents: an Orchestrator, an Image Quality and Preprocessing Agent, a Lesion Detection Agent, a Disease Classification and Risk Prediction Agent, a deterministic Referral Decision Agent, and an LLM Reasoning and Report-Generation Agent supported by a three-tier memory system. Agents communicate through the Model Context Protocol (MCP), a standardized client-server tool-calling layer, rather than bespoke point-to-point integrations. We present the complete system architecture, agent interaction sequence, clinical decision workflow, communication protocol, memory architecture, data flow, and deployment architecture, together with the coordination algorithms governing agent selection, scheduling, conflict resolution, and consensus. A worked case study demonstrates the pipeline end-to-end on a simulated moderate-DR fundus image. Full benchmark-scale quantitative evaluation is scoped as an immediate next phase and is described here as a concrete, reproducible protocol rather than presented as completed results.

Jahnavi Somaraju, Poolasetti Vamshi, "A Multi-Agent Architecture for AI-Based Early Screening, Referral, and Prediction of Retinal Diseases" International Journal of Scientific Research and Engineering Development, V9(4): Page(404-417) May-June 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.