Quantification of Key Retinal Features in Early and Late Age-Related Macular Degeneration Using Deep Learning.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Age-related macular degeneration (AMD) is a leading cause of vision loss.
• Accurate segmentation of AMD features is crucial for diagnosis and monitoring.
• Current manual segmentation methods are time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To develop and validate a deep learning model for segmenting 13 key features of neovascular and atrophic AMD.
• To compare the model's segmentation performance against experienced human graders.

Main Methods:

• A deep neural network was trained using 307 optical coherence tomography (OCT) volumes with 2712 manually delineated B-scans.
• The model performed voxel-level segmentation of 13 common AMD abnormalities.
• Performance was evaluated on 112 B-scans using Dice score, intraclass correlation coefficient, and free-response ROC analysis.

Main Results:

• The model achieved a mean Dice score of 0.63 ± 0.15 on 11 of 13 features, comparable to observers (0.61 ± 0.17).
• The mean intraclass correlation coefficient for the model was 0.66 ± 0.22, also similar to observers (0.62 ± 0.21).
• Free-response ROC analysis indicated similar or superior sensitivity per false positive compared to human observers.

Conclusions:

• The deep learning model's automatic segmentation quality is comparable to experienced graders, outperforming them in some instances.
• Quantified parameters from the model can be integrated into clinical routines.
• The model offers potential for advancing research on treatment response in AMD.