Assessing Glaucoma Progression Using Machine Learning Trained on Longitudinal Visual Field and Clinical Data.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Traditional rule-based methods for glaucoma progression detection using visual fields (VFs) alone present limitations and trade-offs.
• Accurate and timely detection of glaucoma progression is crucial for effective patient management.

Purpose of the Study:

• To assess the performance of a convolutional long short-term memory (LSTM) neural network in determining glaucoma progression.
• To compare the diagnostic ability of a model trained on both VF and clinical data versus one trained solely on VF data.

Main Methods:

• Retrospective analysis of a longitudinal dataset merging VF and clinical data from 11,242 eyes.
• Development and testing of two machine learning models: one using only VF data, and another using both VF and clinical data.
• Comparison of model performance using area under the receiver operating characteristic curve (AUC) and mean accuracies.

Main Results:

• The convolutional LSTM network achieved 91% to 93% accuracy in detecting glaucoma progression based on four consecutive VF results.
• The model trained on combined VF and clinical data demonstrated superior diagnostic ability (AUC, 0.89-0.93) compared to the VF-only model (AUC, 0.79-0.82).

Conclusions:

• A convolutional LSTM architecture effectively captures spatiotemporal trends in VF data for glaucoma progression assessment.
• Integrating clinical data alongside VF results enhances the model's diagnostic performance, mirroring clinical practice.