Deep-ZOMA: A Deep Learning-Based Approach for Automated Morphometric Analysis of Zebrafish Larvae Ocular Structures

Purpose: Zebrafish (Danio rerio) ocular measurements are widely used in ocular disease research and drug discovery. Traditional methods rely on manual annotations, causing challenges in precision, efficiency, and workload. Here, we introduce a deep learning-based zebrafish ocular morphometric analysis (Deep-ZOMA) tool for quantitative measurement of zebrafish larvae ocular structures. Methods: A dual-center dataset of 1820 bright-field images (1010 internal; 810 external) was annotated for key ocular regions. A UNet++ segmentation network was trained with augmentation and hybrid loss. Sixteen morphometric parameters were computed using customized quantification algorithms. Performance was evaluated on internal and external datasets using Dice coefficient, IoU, Student's t-test, correlation, and agreement analyses. Utility was further tested on an slc4a7 knockdown/rescue dataset and compared with three ophthalmology graduate students. Results: Deep-ZOMA achieved mean Dice coefficients of 0.96 on the internal set and 0.95 on the external set for primary ocular regions, with IoU values >0.90. Automated measurements showed strong correlations and excellent agreement with expert measurements. In the slc4a7 model, Deep-ZOMA accurately identified microphthalmia and rescue phenotypes consistent with expert annotations. Measurement and data-entry times were >20-fold faster than those of human observers, with comparable or better accuracy. Conclusions: Deep-ZOMA provides a reliable and efficient solution for high-throughput zebrafish ocular morphometry, supporting applications in ocular genetics, drug screening, and phenotypic studies. Translational Relevance: This deep learning-based system enables accurate, reproducible zebrafish ocular morphometry, accelerating translational research by linking genetic or pharmacological perturbations with quantifiable ocular outcomes relevant to human eye diseases.