A-Eye Web Platform

Automated 3D MRI Segmentation of Human Eye Structures

Upload a human T1-weighted orbital MRI and receive fully-automated segmentation masks for 9 structures of the eye, together with quantitative biomarkers — axial length and volumetry — delivered as a ready-to-use .zip archive.

Start Segmentation
FAQ Cite this work Read the paper
90 MRI cases processed
9 eye structures segmented
1,245 subjects in training cohort

What does A-Eye segment?

The pipeline is based on nnU-Net and was trained on a large-scale, multi-centre cohort of healthy adult subjects. It provides segmentation masks and quantitative biomarkers for the following structures:

Lens Lens The crystalline lens
Globe Globe The full eyeball volume
Optic Nerve Optic Nerve The optic nerve trunk
Intraconal Fat Intraconal Fat Fat inside the muscle cone
Extraconal Fat Extraconal Fat Fat outside the muscle cone
Lateral Rectus Lateral Rectus Lateral rectus muscle
Medial Rectus Medial Rectus Medial rectus muscle
Inferior Rectus Inferior Rectus Inferior rectus muscle
Superior Rectus Superior Rectus Superior rectus muscle

Pipeline Overview

A-Eye uses nnU-Net, a self-configuring deep-learning framework for biomedical image segmentation. The model was validated on 1,245 healthy adult subjects and achieves state-of-the-art performance across all nine orbital structures.

nnU-Net pipeline overview

Accepted input formats

  • .nii — NIfTI image
  • .nii.gz — compressed NIfTI
  • .zip / .7z — archive of NIfTI or DICOM files
  • .dcm — DICOM (inside an archive)

T1-weighted sequences only. Up to 10 cases per submission.

What you get back

  • Per-eye masks — left, right, merged (.nii.gz)
  • Cropped per-eye masks used for the biomarkers
  • Inference details and the logs of your run
  • Axial length visualisations (.png)
  • Biomarkers summary table (biomarkers.csv)

All results packaged in a single .zip download, together with the license agreement. Each case folder also holds the image the masks are defined on, so you can check the overlay straight away — keep your own copy of what you uploaded.

Cite this work

If A-Eye contributes to your research, please cite the associated publication and dataset:

Jaime Barranco, Adrian Luyken, Hamza Kebiri, Philipp Stachs, Pedro M. Gordaliza, Oscar Esteban, Yasser Aleman, Raphael Sznitman, Oliver Stachs, Sönke Langner, Benedetta Franceschiello, Meritxell Bach Cuadra. A-eye: Automated 3D MRI Segmentation and Morphometric Feature Extraction for Eye and Orbit Atlas Construction. PLOS ONE. https://doi.org/10.1371/journal.pone.0352257
Barranco J, Luyken A, Stachs P, Esteban O, Aleman-Gomez Y, Stachs O, et al. MR-Eye atlas: a large-scale atlas of the eye based on T1-weighted MR imaging [dataset]. Zenodo; 2024. https://doi.org/10.5281/zenodo.13325369
Read the paper

Resources & downloads

All code, eye atlases, and pretrained model weights behind A-eye are openly available.

GitHub Source code of the A-Eye pipeline and web platform and the Docker image with nnU-Net and the pretrained model weights already installed, for plug-and-play segmentation. Zenodo Eye atlases — male (594), female (616) and combined (1,210) subjects — with their majority-voting segmentations, the manual segmentation of the combined atlas, the pretrained model weights, and a reference to the Docker image.