ROQUS: a retinal OCT quality and usability score

A deep learning-based retinal OCT B-scan quality metric, ROQUS, assesses overall quality and clinical usability. The method uses a ranking strategy, producing an unbounded score where higher values indicate better quality. The model was evaluated on in-house and public datasets with real and simulated acquisition issues. ROQUS achieved 0.85 ROC-AUC in identifying B-scans with acquisition issues, surpassing classical metrics, and effectively handling noise and brightness variations. The inter-human and human-ROQUS agreement level when ranking B-scan pairs of different qualities was similar. ROQUS enables objective quality assessment, improving the identification of poor-quality acquisitions, enhancing clinical research, and streamlining daily practice.