Model-to-Data Approach for Deep Learning in Optical Coherence Tomography Intraretinal Fluid Segmentation.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Deep learning (DL) is rapidly advancing in medicine, particularly ophthalmology.
• Concerns about data privacy, security, and sharing are significant barriers to DL implementation.
• A model-to-data approach, transferring the algorithm instead of data, offers a potential solution.

Purpose of the Study:

• To evaluate the feasibility of a model-to-data deep learning approach in ophthalmology.
• To assess the model's ability to recognize intraretinal fluid (IRF) on optical coherence tomography (OCT) B-scans.

Main Methods:

• A single-center, cross-sectional study involving patients with exudative age-related macular degeneration.
• Training a DL model using a model-to-data approach on 400 OCT B-scans to detect IRF.
• Comparing the DL model's performance against manual human grading using Dice coefficients and intersection over union (IoU) scores.

Main Results:

• The DL model achieved a learning curve Dice coefficient greater than 80%.
• No statistically significant difference was found between the model and human graders in Dice coefficients or IoU scores (P > .05).
• The model-to-data approach successfully circumvented data sharing, security, and privacy issues.

Conclusions:

• A model-to-data DL approach is applicable in ophthalmology, demonstrating effectiveness in detecting IRF.
• This proof-of-concept study highlights the potential for secure AI implementation in eye care.
• Further large-scale, multicenter studies are warranted to explore the clinical relevance of this paradigm.