Code-Free Glaucoma Classification Research with Google Cloud Vertex AI

In a new proof-of-concept study, the performance of the Google Cloud Vertex AI AutoML platform was evaluated to distinguish referable glaucoma cases using retinal fundus photographs.

Objective and Scope of the Research

Traditional deep learning models typically require programming expertise and specialized neural network architecture knowledge. This research examined how a commercial and no-code artificial intelligence platform can facilitate this process.

Dataset Used and Distribution

The study utilized a publicly available balanced retinal fundus image dataset. A total of 9,540 color fundus photographs, with an equal number from both classes, were used.

The images were divided into training, validation, and test sets, allowing for the development and testing of the model.

Model Development Process

Researchers developed the model using the platform's graphical user interface without writing any code or manually designing a neural network architecture.

During model training, a maximum compute budget of eight hours was chosen, and the process was completed with a cloud-based prediction endpoint.

Obtained Performance Results

The developed AutoML model generally achieved a mean precision of 0.974. At the default confidence threshold, precision and recall values were recorded at 92.5 percent.

The model correctly classified 94 percent of non-referable images and 91 percent of referable ones.

Future Studies and Requirements

The findings support that commercial artificial intelligence tools can serve as accessible instruments in ophthalmic image classification research.

However, because the model was evaluated using a single derived dataset, it must be validated with independent populations and different systems before clinical application.