Automatic geographic atrophy segmentation using optical attenuation in OCT scans with deep learning.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Geographic atrophy (GA) is a leading cause of vision loss in age-related macular degeneration (AMD).
• Accurate identification and quantification of GA are crucial for monitoring disease progression and evaluating treatment efficacy.
• Current methods for GA assessment can be labor-intensive and subjective.

Purpose of the Study:

• To develop and validate a deep learning algorithm for automated identification, segmentation, and quantification of GA.
• To compare the performance of a deep learning model trained on optical attenuation coefficient (OAC) images versus sub-retinal pigment epithelium (subRPE) OCT images for GA detection.

Main Methods:

• Developed a deep learning algorithm utilizing U-Net architecture for GA analysis from optical coherence tomography (OCT) datasets.
• Calculated OACs from OCT scans to generate composite en face OAC images for GA lesion identification.
• Trained and evaluated two deep learning models using OAC images and subRPE OCT images, assessing performance with DICE similarity coefficients (DSCs) and comparing with manual segmentations.

Main Results:

• Both deep learning models achieved 100% sensitivity and specificity in identifying GA on a subject level.
• The model trained with OAC images demonstrated significantly higher DSCs (0.940 ± 0.032) compared to the subRPE OCT model (0.889 ± 0.056).
• The OAC-trained model showed stronger correlation (r=0.995) and smaller mean bias (0.011 mm) with manual segmentations than the subRPE OCT model (r=0.959, mean bias=0.117 mm).

Conclusions:

• The proposed deep learning model effectively and accurately identifies, segments, and quantifies GA using OCT scans.
• Utilizing composite OAC images in deep learning models enhances the precision of GA detection and measurement.
• This automated approach offers a promising tool for objective and efficient GA assessment in clinical practice.