Impact of acquisition area on deep-learning-based glaucoma detection in different plexuses in OCTA

Area of Science:

• Ophthalmology
• Medical Imaging
• Neuroscience

Background:

• Glaucoma is a neurodegenerative disease leading to irreversible blindness.
• Early diagnosis and treatment are crucial for slowing glaucoma progression.
• Optical coherence tomography angiography (OCTA) offers non-invasive insights into retinal microcirculation, a key indicator of glaucoma.

Purpose of the Study:

• To evaluate the impact of OCTA acquisition area and capillary plexuses on machine learning-based glaucoma detection.
• To compare the performance of vessel density (VD) and deep learning approaches in distinguishing glaucoma patients from healthy controls.

Main Methods:

• Analysis of OCTA scans from different areas: 3x3mm macula, 6.44x6.4mm macula, and 6x6mm optic nerve head (ONH).
• Machine learning models were trained to differentiate glaucoma patients from healthy controls using vascular data.
• Comparison of deep learning performance against vessel density (VD) and traditional OCT biomarkers.

Main Results:

• The 6x6mm ONH scan area demonstrated the highest performance across all capillary plexuses.
• Deep learning approaches significantly outperformed VD as a biomarker for glaucoma detection.
• OCTA-based deep learning also showed superior performance compared to conventional OCT biomarkers.

Conclusions:

• OCTA scans of the optic nerve head (ONH) are valuable for glaucoma assessment.
• Deep learning utilizing OCTA data, especially from the ONH, offers a powerful tool for glaucoma diagnosis and monitoring.
• OCTA may serve as a valuable adjunct to OCT in clinical glaucoma studies.