Papilledema and Pseudopapilledema Recognition using Artificial Intelligence: A Code-Free Automated Machine Learning Model

Theme: Neuro-ophthalmology
What: Neuro-ophthalmology
Part of: Neuro-ophthalmology II / Neuro-ophtalmolgie II
When: 6/1/2024, 02:00 PM - 03:30 PM
Where: Room | Salle 714 B

Abstract

Purpose: Papilledema is characterized by bilateral optic nerve swelling due to increased intracranial pressure. It can sometimes be challenging to distinguish papilledema from pseudopapilledema, which is defined as an abnormal elevation of the optic nerve without swelling. Automated Machine Learning (AutoML) enables the creation of artificial intelligence algorithms without requiring programming knowledge. The goal of this project is to develop a deep learning algorithm using AutoML to differentiate between papilledema and pseudopapilledema.

Study design: Diagnostic accuarcy study

Methods: An ophthalmology trainee with no previous coding experience designed a deep learning (DL) model in Google Vertex AutoML Image Classification. We used a public dataset with 1368 optic head photos produced by the Department of Ophthalmology of Kim’s Eye Hospital in South Korea. Model training, validation and testing were done using this database. External validation was performed on 30 images collected from different websites.

Results: The AutoML model demonstrated excellent discriminating performance. The area under the precision-recall curve was 0.981. At the 0.5 confidence threshold cut-off, the overall performance metrics were as follows: precision (97.8%), sensitivity (97.8%), specificity (98.9%), and accuracy (99.0%). Looking at each subgroup specifically, precision varied from 93.5–100.0%, sensitivity varied from 96.6–100.0%, specificity varied from 98.3–100.0% and accuracy varied from 97.8–100.0%. Pseudopapilledema was the most accurately predicted subgroup (with an accuracy of 100.0%). Our model had similar performance metrics to published DL models handcrafter by AI experts, despite being the first study to look at pseudopapilledema. 

Conclusion: A machine learning model developed without a single line of code by an ophthalmology trainee could accurately identify and classify papilledema from pseudopapilledema and normal optic nerve images with an accuracy comparable or better than models developed by experts. 

Presenter(s)

Presenting Author: Samir Touma

Additional Author(s):

Tracy Aoun, University of Montreal

Fares Antaki, University of Montreal

Daniel Milad, University of Montreal

Renaud Duval, University of Montreal

---