Utilization of Image-Based Deep Learning in Multimodal Glaucoma Detection Neural Network from a Primary Patient...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma diagnosis relies on integrating various data sources.
• Current AI models often use single data modalities, potentially limiting accuracy.
• Developing a robust AI for glaucoma requires a comprehensive approach.

Purpose of the Study:

• To create a multimodal neural network for glaucoma detection.
• To train the model on time-matched fundus photographs, OCT scans, and Humphrey visual field (HVF) data.
• To evaluate the model's performance on a clinical dataset.

Main Methods:

• Developed a two-component multimodal neural network.
• Used convolutional neural networks for feature extraction from each modality.
• Integrated features using a multilayer perceptron for final classification.

Main Results:

• Single-modality models performed poorly on the clinical dataset.
• Multimodal integration significantly improved performance metrics.
• Area under the curve (AUC) increased from 0.57 to 0.86, and specificity from 0.77 to 0.85.

Conclusions:

• Multimodal AI models offer superior glaucoma detection compared to single-modality approaches.
• Integrating structural and functional data enhances AI model accuracy.
• This study demonstrates a production-level AI solution for glaucoma diagnosis using multimodal data.