Assessment of a Segmentation-Free Deep Learning Algorithm for Diagnosing Glaucoma From Optical Coherence Tomography...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Conventional retinal nerve fiber layer (RNFL) segmentation in spectral-domain optical coherence tomography (SD-OCT) scans can lead to errors in diagnosing glaucoma.
• Accurate detection of glaucomatous damage is crucial for preserving vision.

Purpose of the Study:

• To develop and evaluate a segmentation-free deep learning (DL) algorithm for assessing glaucomatous damage using entire SD-OCT circle B-scan images.
• To compare the performance of the DL algorithm against conventional RNFL thickness parameters.

Main Methods:

• A convolutional neural network was trained on SD-OCT circle B-scan images from eyes with and without glaucoma.
• The dataset included 20,806 SD-OCT images from 1154 eyes (635 individuals).
• Performance was evaluated using the area under the receiver operating characteristic curve (AUC) and sensitivity at specific specificities.

Main Results:

• The DL algorithm demonstrated a significantly higher AUC than global RNFL thickness (0.96 vs 0.87) and individual sectors for discriminating between glaucoma and controls (P < .001).
• At 95% specificity, the DL algorithm achieved 81% sensitivity, compared to 67% for global RNFL thickness.
• The DL algorithm showed superior performance across all stages of glaucoma, particularly in preperimetric and mild perimetric disease.

Conclusions:

• A segmentation-free DL algorithm offers improved diagnostic accuracy for glaucomatous damage on SD-OCT compared to conventional RNFL thickness parameters.
• This AI-driven approach shows particular promise for detecting early-stage glaucoma.
• Further research should explore integrating this DL method with clinical data for enhanced diagnostic decision-making.