Radial polarisation patterns identify macular damage: a machine learning approach.

Area of Science:

• Ophthalmology and visual science.
• Medical imaging and diagnostics.
• Machine learning in healthcare.

Background:

• Macular damage detection and monitoring are crucial for preserving vision.
• Current methods for assessing macular integrity have limitations.
• Polarisation-modulated patterns offer a novel approach for visual assessment.

Purpose of the Study:

• To evaluate the effectiveness of polarisation-modulated patterns in identifying macular damage and foveolar involvement.
• To compare the predictive power of polarisation-modulated patterns against traditional visual function measures.
• To develop a machine learning model for predicting macular damage.

Main Methods:

• A cross-sectional study of 520 eyes with varying macular conditions.
• Assessment of macular damage and foveolar integrity using optical coherence tomography.
• Application of Naïve Bayes supervised machine learning with feature selection (Fast Correlation-Based Filter) and 5-fold cross-validation.

Main Results:

• Radially structured polarisation-modulated patterns and age were identified as key predictors of macular damage and foveolar involvement.
• Traditional logMAR visual acuity measures were found to be redundant.
• The Naïve Bayes model achieved an area under the receiver operating characteristic curve exceeding 0.7, indicating good predictive performance.

Conclusions:

• Radially structured polarisation-modulated geometric patterns are superior to optotypes and logMAR acuity measures for predicting macular damage.
• This novel method shows promise for improved detection and monitoring of macular diseases.
• Machine learning enhances the predictive capabilities of polarisation-modulated patterns in ophthalmology.