Development of an AI-Based Automated Morphological Analysis System for Glaucoma: Validation Using an Engineered Exosome Therapy Model

Purpose: Current glaucoma therapy evaluation is hindered by subjective, low-throughput manual counting. We developed an AI-based automated platform for high-dimensional quantification of retinal ganglion cell（RGC） morphometry. To validate its sensitivity, we utilized a novel immunomodulatory model—functionalized M2-microglia-derived exosomes. This study establishes a "Software as a Medical Device" (SaMD) workflow correlating therapeutic efficacy with quantifiable morphological recovery.
Methods: Validation Model Construction: We engineered M2-sEVs co-delivering siRNA targeting proinflammatory miRNA-9-5p and the cGAS-STING inhibitor, DDA. This dual-cargo system was administered to a glaucoma animal model to suppress neuroinflammation and promote RGC survival.
AI-Driven Automated Analysis: Retinal flat-mounts were immunostained for αRGCs. A U-Net deep learning pipeline automatically segmented RGCs from confocal images. Unlike manual counting, this platform executed comprehensive morphological profiling, generating reports on soma area, dendritic field coverage, length, and complexity indices (Sholl analysis, fractal dimension).
Results: We found that the engineered M2-sEV intervention could successfully reprogram the retinal microenvironment, preserving RGC viability. Crucially, the AI-based platform is expected to demonstrate superior sensitivity in detecting therapeutic outcomes compared to conventional cell counting. The automated analysis aims to reveal that while untreated eyes exhibit significant dendritic retraction and fragmentation, the AI-quantified metrics in the exosome-treated group will show preserved dendritic complexity and field area comparable to healthy baselines. Preliminary data suggests the software can rapidly process large datasets with high segmentation accuracy (Dice coefficient).
Conclusions: This study validates an AI-driven pipeline capable of detecting subtle morphological changes often missed by manual counting. By objectively quantifying the neuroprotective effects of engineered exosomes, we demonstrate the software’s potential as a high-throughput, standardized tool for evaluating glaucoma therapeutics and structural restoration.