Retinal Nerve Fiber Layer Features Identified by Unsupervised Machine Learning on Optical Coherence Tomography Scans...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computational Biology

Background:

• Glaucoma diagnosis and progression monitoring rely on structural and functional measures.
• Current methods may not capture subtle, early structural changes indicative of glaucoma.
• Wide-angle swept-source optical coherence tomography (SS-OCT) provides high-resolution retinal imaging.

Purpose of the Study:

• To apply computational techniques to wide-angle SS-OCT images.
• To identify novel, glaucoma-related structural features.
• To improve glaucoma detection and prediction of glaucomatous progression.

Main Methods:

• Collected wide-angle SS-OCT, SD-OCT, standard automated perimetry (SAP), and frequency doubling technology (FDT) data over 2 years.
• Extracted retinal nerve fiber layer (RNFL) thickness maps from SS-OCT images.
• Utilized principal component analysis (PCA) for unsupervised machine learning to identify novel RNFL structural features.

Main Results:

• RNFL PCA features showed significant associations with visual field mean deviation (MD) and circumpapillary RNFL thickness (cpRNFLt).
• RNFL PCA demonstrated superior diagnostic accuracy for glaucoma detection (AUC 0.95) compared to cpRNFLt (0.90), SAP MD (0.86), and FDT MD (0.83).
• RNFL PCA also achieved significantly higher accuracy in predicting glaucomatous progression.

Conclusions:

• A computational approach using SS-OCT can identify novel structural biomarkers for glaucoma.
• These identified features enhance the accuracy of glaucoma detection and progression prediction.
• This method offers a promising advancement in managing glaucoma patients.