Skip to content

AutoMorph

Summary

AutoMorph is an open-source deep learning pipeline designed for the automated analysis and quantification of retinal vascular morphology from color fundus photographs (CFPs). It automates key steps in Oculomics research, including image pre-processing, quality grading, anatomical segmentation, and feature extraction. AutoMorph is widely used in clinical and genetic studies to link retinal vascular features with systemic and ophthalmic diseases.

Workflow and Modules

The AutoMorph pipeline consists of four major functional modules: 1. Image Pre-processing: Normalizes fundus photographs to reduce variation in illumination, resolution, and color across different cameras and datasets. 2. Quality Grading: Uses an EfficientNet-b4 architecture trained on the EyePACS-Q dataset to evaluate the usability of images, filtering out low-quality scans that could lead to erroneous vascular measurements. 3. Anatomical Segmentation: Automatically maps crucial retinal structures, including: - Binary vessel segmentation (creating masks of the vascular tree). - Artery/vein classification. - Optic disc and optic cup segmentation. 4. Vascular Morphology Measurement: Extracts quantitative metrics from the segmented maps, such as vessel caliber, tortuosity, fractal dimension, and branching complexity.

Applications

In the multi-omic study of retinal features by Julian et al. (2026), AutoMorph was utilized for two key steps: - Quality Control: Excluding fundus photographs of "reject" quality to select high-quality images for downstream analysis. - Vasculature Masking: Generating binary masks of the vascular tree from CFP images. These masks were used to weight the reconstruction loss of the retinal adversarial autoencoder (Ret-AAE) by 12.29x for vasculature pixels, ensuring that critical vessel structures were preserved in the low-dimensional embeddings.

Citations

  • Zhou, Y. et al. (2022). AutoMorph: automated retinal vascular morphology quantification via a deep learning pipeline. Translational Vision Science & Technology, 11(12), 12. DOI: 10.1167/tvst.11.12.12
  • Julian, T. H. et al. (2026). Multi-omic analysis of deep learning-derived phenotypes links ophthalmic imaging to cardiovascular and neurological traits. Nature Cardiovascular Research, 5, 541–554. DOI: 10.1038/s44161-026-00815-5. Source paper: s44161-026-00815-5.pdf