Angle-closure assessment in anterior segment OCT images via deep learning

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate anterior chamber angle (ACA) characterization is crucial for diagnosing angle-closure disease.
• Current gold standard, gonioscopy, is invasive, uncomfortable, and can yield inaccurate results due to eye deformation.
• A non-contact, objective method for ACA grading is needed to improve diagnostic accuracy and patient comfort.

Purpose of the Study:

• To develop and validate a novel classification schema for grading ACAs into open, appositional, and synechial types using Anterior Segment Optical Coherence Tomography (AS-OCT).
• To provide clinicians with a tool for better understanding angle-closure disease progression and guiding treatment decisions.
• To establish an automated, non-contact method for ACA assessment, overcoming the limitations of gonioscopy.

Main Methods:

• Utilized an image alignment technique to create AS-OCT image sequences.
• Developed an automated method for localizing the ACA region by segmenting the iris.
• Employed a Multi-Sequence Deep Network (MSDN) incorporating a Convolutional Neural Network (CNN) and a ConvLSTM-TC module for feature extraction and spatial analysis.
• Introduced a novel time-weighted cross-entropy loss (TC) for optimizing the classification model.

Main Results:

• The proposed MSDN architecture effectively extracted spatial and temporal features from AS-OCT images.
• The ConvLSTM-TC module and TC loss function improved the classification performance for ACA grading.
• The method demonstrated superior applicability, effectiveness, and accuracy compared to existing state-of-the-art techniques.
• Evaluated on 66 eyes (1584 sequences, 16,896 images), achieving high diagnostic performance.

Conclusions:

• The developed AS-OCT-based automated classification system provides a promising, non-contact alternative for ACA assessment.
• This AI-driven approach enhances the accuracy and efficiency of diagnosing and managing angle-closure disease.
• The findings suggest a significant advancement in ophthalmic imaging analysis for glaucoma screening and patient care.