Multiparametric OCTA Biomarkers for Classifying Mild Diabetic Retinopathy: A Cross-Sectional Evaluation.

Twenty-seven OCTA parameters demonstrated significant group-level differences between mild DR and healthy controls.

• Parameters were extracted from superficial and deep vascular plexuses across foveal and parafoveal regions.
• Quantitative parameters described perfusion, vessel length, caliber, tortuosity, branching, and fragmentation.
• Correlation filtering and PCA-guided selection were applied prior to model evaluation.

Single-parameter SVM models demonstrated moderate discriminative performance for classifying mild DR versus healthy controls.

• Mean AUC range for single-parameter models was 0.47–0.88.
• Performance was assessed using repeated random partitioning into independent training and validation sets across 50 splits.
• Discrimination was quantified by the area under the ROC curve (AUC).

Multiparameter OCTA feature combinations yielded consistently higher discrimination than single-parameter models.

• A subset of multiparameter combinations achieved average AUCs in the approximately 0.95 range.
• All feature combinations were evaluated using support vector machine (SVM) models.
• Performance was assessed using 50 repeated random partitions into independent training and validation sets.

The most stable contributors to high-performing models were not the strongest individual discriminators but rather features weakly correlated with the broader parameter set.

• Stable contributors were identified as features recurrently integrated across high-AUC combinations.
• These features reflected complementary vascular information across layers and regions.
• Feature stability and complementarity, rather than univariate strength alone, underlie robust model performance.

Structured integration of complementary OCTA features can enhance early DR discrimination without relying on opaque end-to-end models.

• The approach used PCA-guided selection combined with SVM classification.
• The study evaluated all feature combinations systematically across superficial and deep vascular plexuses.
• The authors emphasize this framework informs future quantitative OCTA analyses in a transparent, interpretable manner.