Task-Specific Quality Gating for Retinal Optical Coherence Tomography B-Scans: Learned Representations Over Scalar Metrics in Choroid Segmentation
Authors
Affiliations (1)
Affiliations (1)
- University of Pittsburgh
Abstract
Automated segmentation of Optical Coherence Tomography (OCT) images is a critical component of structural biomarker extraction for retinal diagnostics. Deep learning models achieve state-of-the-art performance on controlled datasets, yet exhibit unpredictable failures on real-world data. Current quality gates rely on device-reported scan quality scores, which have been shown to be unreliable predictors of segmentation performance. We define scan quality in a task-specific sense that is, whether a given B-scan will yield a reliable segmentation from a particular trained model. Under this definition, we perform a systematic evaluation of No-Reference Image Quality Assessment (NR-IQA) metrics, general-purpose and domain-specific pretrained representations as alternative quality gates. To this end, we use choroid segmentation as the prototype task, with a dataset of 6,076 OCT B-scans from 80 subjects. These quality gate candidates are evaluated at three levels: scalar metrics (BRISQUE, NIQE, PIQE, SNR, PSNR), supervised linear probing, and unsupervised partitioning (K-Means) of the feature vectors and the learned representations. All scalar NR-IQA metrics proved inadequate (|r| < 0.20). General-purpose ImageNet-based pretrained representations (EfficientNet-b0, ResNet-50, ViT-B/16) improve upon NR-IQA, achieving ROC-AUC up to 0.77, indicating that learned representations are better suited to task-specific quality gating than hand-crafted scalar statistics. Retinal foundation models (FMs) further close the gap: RETFound (OCT-specific FM) achieves ROC-AUC {approx} 0.81. Unsupervised K-Means partitioning indicates that general-purpose ImageNet-pretrained embeddings, despite carrying a linearly decodable quality signal, do not reliably organize scans by quality geometrically, whereas the retinal FMs produce quality-aligned clusters that exceed a patient-level permutation null, suggesting that domain-specific pretraining provides additional, complementary benefit on top of general-purpose learned representations.