A deep learning approach to investigate the filtration bleb functionality after glaucoma surgery: a preliminary study

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Filtration blebs (FBs) are crucial for glaucoma surgery success.
• Distinguishing functioning from failed FBs is vital for patient management.
• Objective assessment of FB status can be challenging.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for classifying filtration bleb (FB) functionality.
• To differentiate between functioning and failed FBs using slit-lamp images.

Main Methods:

• A retrospective, multicenter study utilized 119 post-trabeculectomy FB images.
• A deep learning classification algorithm was developed using ResNet architectures.
• Transfer and ensemble learning techniques were applied to optimize the model.

Main Results:

• The DL model achieved 74% accuracy, 74% sensitivity, and 87% specificity.
• Area under the ROC curve was 0.8, and area under the precision-recall curve was 0.74.
• The model demonstrated statistically significant performance (P=0.00307, Kappa=0.58).

Conclusions:

• The DL model effectively discriminates functioning from failed filtration blebs.
• This AI approach can assist clinicians in managing patients post-glaucoma surgery.
• The model offers a valuable tool for objective FB assessment.