Corneal Learning for Endothelial Assessment and Review using AI for Endothelial Count (CLEAR-EC) Worldwide Challenge

CLEAR-EC Challenge Overview:

This challenge is in partnership with Grand Challenge. Please see this site for official registration and submission instructions: https://clear-ec.grand-challenge.org/

Aim:

The aim of the competition is to develop an AI algorithm that will standardize and optimize the evaluation of corneal endothelial cells for corneal transplantation. Output values of the algorithm consist of the three key metrics:

• Cell density (# of living cells / Area)
• Coefficient of variation (Standard deviation of cell area / Mean of cell area)
• Hexagonality (# of cells with 6 sides / Total number of cells)

The resulting model will be piloted in eye banks to enable scalable, precise, and continuous quality monitoring, ultimately supporting higher success rates and reduced rejection in corneal transplantation.

Challenge Set-up: 

Phase I of the challenge provides an initial evaluation of all submitted models using a set of 100 test images. Phase II expands the test set to 1,000 images. Each team may submit up to three model runs in Phase I and up to two runs in Phase II. Performance will be assessed using cell density, coefficient of variation, and hexagonality, combined into a single score through a weighted mean percent error (MPE).

Dataset: 

The dataset consists of 9,000 training images and 1000 test images of specular microscopy images collected from eye banks in the US (New York, Georgia, Kentucky, and Delaware). Images will be provided in BMP format and are approximately 1.5-2MB each. The training dataset will be available through Zenodo at this link: https://zenodo.org/records/21270595.

Code Availability: 

The organizers have provided a public baseline code for all participants. This is an open reference pipeline (image loading → cell segmentation → metric calculation) so that participants can see how raw images become CD/CV/HEX predictions and how to format their outputs. The baseline establishes a reference level that participating methods are expected to surpass.

This code can be accessed at: https://github.com/CLEAR-EC/CLEAR-EC

Getting Started: 

1. Register for the challenge on Grand Challenge
2. Download the training data from Zenodo.
3. Study the baseline from Github. 
4. Build and containerize your algorithm using the example algorithm template from the challenge phase pack.
5. Submit your algorithm on Grand Challenge before the Phase I deadline.

Timeline:

Phase I: July 15 - September 14

Phase I Evaluation and Review: September 15 - September 30

Phase II: October 1 - November 14

Phase II Evaluation and Review: November 15 - November 30

Significance: 

In 2023, U.S. eye banks reported 137,697 total tissue recoveries from 69,637 patients, with graft assessments performed manually by highly trained technicians. Establishing an automated, standardized model for evaluating corneal grafts prior to transplantation may greatly enhance efficiency, reduce reliance on technician expertise, and lower costs while ensuring high-quality, uniform assessments. Artificial intelligence offers a transformative opportunity to evaluate thousands of grafts within minutes—an effort that currently requires weeks to months—by analyzing the entire tissue rather than isolated regions. Comprehensive, AI-generated metrics on tissue quality, endothelial cell density, morphology, and overall graft health will significantly improve transplant decision-making and ultimately patient outcomes.

Meet the Organizers: