Evaluating OCT Device-Reported Image Quality Score: Towards a Task-Specific Quality Gate for Deep Learning-based Outer-Retina and Choroid Boundary Segmentation

Device-reported image quality (IQ) scores are commonly used as a proxy for downstream deep-learning segmentation reliability, but this assumption is rarely tested directly. We provide empirical evidence across 5,000+ OCT B-scans that Heidelberg Engineering’s device-reported Q-score is not a reliable predictor of deep-learning segmentation performance for outer-retina and choroid boundaries — the correlation between IQ score and segmentation quality is negligible. We introduce a task-specific quality metric adapting Earth Mover’s Distance to one-dimensional boundary data (the area between predicted and ground-truth boundary vectors) as a more reliable quality gate for clinical deployment.

In medRxiv, 2026. doi:10.64898/2026.05.17.26353399. Under review at Translational Vision Science & Technology (TVST).