Artificial Intelligence

for RObust Glaucoma

Screening Challenge Lite

AI for Medical Imaging Course

7 Sep 2022

​

Coen de Vente

Lite

The original AIROGS challenge

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

1

Organizers

AIROGS Lite is a stripped down version of AIROGS with less data and no ungradability task, specifically developed for this course.

What is glaucoma?

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

2

Brain

What is glaucoma?

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

3

Optic nerve head

What is glaucoma?

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

4

What is glaucoma?

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

5

What is glaucoma?

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

6

What is glaucoma?

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

7

What is glaucoma?

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

8

Screening motivation

Population studies

Over 50% of glaucoma cases undiagnosed^1,2

About 10% will go blind in both eyes³

About 24% will go blind in one eye³

Globally up to 111.8 million in 2040⁴

Main cause: detected too late^*

All undiagnosed cases had visited an optometrist or ophthalmologist in preceding year (Australia)⁵

​

1. Friedman DS et al. Arch Ophathalmol. 2004;122:532-538; 2. Quigley Ha. Br J Ophthalmol. 1996;80:389-393; 3. Mokhles P et al., J Glaucoma. 2016-25(7):623-8. 3. Tham YC et al. Ophthalmology. 2014;121(11):2081-90. 5. Wong EY et al. Ophthalmology. 2004;111:1508-1514. *Apart from compliance issues, availability of therapy, etc.

​

​

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

9

Screening motivation

Too many people visually impaired by glaucoma

Our societies as a whole don’t do a very good job at detecting glaucoma

Eye care professionals miss many cases

Once detected, the disease may be well controlled (medication, laser, surgery)

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

10

AI screening motivation

Screening can fulfill need for early detection

​

Well visible in color fundus images

​

AI solutions can enable cost-effective glaucoma screening

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

11

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

12

Aim

Development an AI solution for glaucoma screening by classifying CFPs as “referable glaucoma” or “no referable glaucoma”.

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

13

Data

Subset of Rotterdam EyePACS AIROGS dataset

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

14

Test set (hidden labels)

*The use of additional CFP development data (including other data from the original AIROGS challenge and weights pretrained on fundus image data) is prohibited. The use of such data or pretraining is also not allowed in any stage of the algorithm (so also not in a preprocessing or postprocessing step). The use of other data and pretraining with other data, such as natural images such as those from ImageNet or other medical images is allowed.

Announced later

Evaluation Pipeline

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

15

Rank participants

Glaucoma screening performance

Output of test set (as CSV file)

Metrics

Participant methods overview

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

16

Winning method

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

17

Wang, H., Li, S., & Fang, Y. (2022). A workflow for computer-aided diagnosis of glaucoma. Retrieved 17 May 2022, from

https://github.com/wanghe-thu-pumc/Airogs_workflow/blob/main/Airogs_fine_my.pdf

Winning method

More details: https://ieeexplore.ieee.org/document/9854585

​

Other winners: See https://airogs.grand-challenge.org/evaluation/final-test-phase/leaderboard/

Note 1: AIROGS participants had other data and an additional task.

Note 2: Some participants manually annotated the dataset with vessels/optic disk for a preprocessing step to increase performance. You are allowed to do so, as well.

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

18

Wang, H., Sun, H., Fang, Y., Li, S., Feng, M., & Wang, R. (2022, March). A Workflow for Computer-Aided Diagnosis of Glaucoma. In 2022 IEEE International Symposium on Biomedical Imaging Challenges (ISBIC) (pp. 1-4). IEEE.

Referable glaucoma

output

Summary

Glaucoma is a major cause for blindness and is often detected too late.

​

Aim is to mimic a glaucoma screening setting by classifying glaucoma from retinal images.

​

Test labels will not be available, only the images. You will need to upload your predictions to our challenge website.

​

The approach of choice is up to you.

​

Method justification, experimentation and reporting are more important than a high score!�

Links

Development data: https://doi.org/10.5281/zenodo.7056954

Challenge website: https://airogs-lite.grand-challenge.org/

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

19

Contact

E-mail: c.w.devente@uva.nl

​