Artificial intelligence for morphology-based function prediction in neovascular age-related macular degeneration

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Spatially-resolved mapping of rod- and cone-function is crucial for monitoring macular diseases and assessing treatment outcomes.
• Current methods like mesopic and dark-adapted two-color fundus-controlled perimetry (FCP) are time-consuming.
• Neovascular age-related macular degeneration (nAMD) significantly impacts retinal function.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting mesopic and dark-adapted (DA) retinal sensitivity in nAMD patients.
• To assess the accuracy of ML-based retinal sensitivity prediction using multimodal imaging data.
• To explore the potential of this AI-driven approach as a surrogate endpoint in clinical trials.

Main Methods:

• Acquired extensive psychophysical testing and volumetric multimodal retinal imaging data (FCP, OCT, confocal scanning laser ophthalmoscopy).
• Employed a patient-wise leave-one-out cross-validation strategy for model evaluation.
• Utilized an artificial intelligence (AI) analysis strategy termed "inferred sensitivity".

Main Results:

• Achieved prediction accuracies (mean absolute error, MAE) of 3.94 dB for mesopic, 4.93 dB for DA cyan, and 4.02 dB for DA red FCP.
• Incorporating patient-specific sensitivity data improved MAE to 2.8 dB (mesopic), 3.71 dB (DA cyan), and 2.85 dB (DA red).
• Outer nuclear layer thickness was identified as the most significant predictive feature.

Conclusions:

• The AI-based "inferred sensitivity" approach accurately predicts mesopic and DA retinal sensitivity in nAMD.
• This method enables estimation of differential effects of retinal structural abnormalities on cone- and rod-function.
• Inferred sensitivity holds promise as a quasi-functional surrogate endpoint for future nAMD clinical trials.