A deep learning framework has been developed to assist in the diagnosis and referral of retinal diseases using three-dimensional optical coherence tomography (OCT) scans. The framework, trained on 14,884 scans, achieves performance comparable to or exceeding that of expert clinicians in diagnosing and referring patients with sight-threatening retinal diseases. The system uses a two-stage approach: a segmentation network that creates device-independent tissue maps, followed by a classification network that analyzes these maps to provide diagnoses and referral suggestions. The segmentation network is trained on manually segmented OCT scans, while the classification network is trained on scans with confirmed diagnoses and optimal referral decisions. The framework demonstrates robust performance across different OCT devices, maintaining accuracy even when using tissue segmentations from a different type of device. It also shows high accuracy in diagnosing multiple retinal pathologies and achieving expert-level performance in referral decisions. The system was tested on two independent datasets of 997 and 116 patients, achieving an error rate of 5.5% on the first dataset and 3.4% on the second. The framework's results were compared to those of expert clinicians, showing that it performs as well as or better than human experts in most cases. The system's ability to generalize across different OCT devices and its potential for clinical application in various settings make it a promising tool for improving the efficiency and accuracy of retinal disease diagnosis and referral. The framework also has the potential to be used in other areas of medicine where medical imaging is involved, and its results could be used to train healthcare professionals to expert levels. The study highlights the potential of deep learning in clinical settings, demonstrating its ability to provide accurate and reliable diagnoses and referrals, even in complex and ambiguous cases. The framework's results suggest that deep learning can be a valuable tool in the diagnosis and management of retinal diseases, with the potential to improve patient outcomes and reduce the workload of healthcare professionals.A deep learning framework has been developed to assist in the diagnosis and referral of retinal diseases using three-dimensional optical coherence tomography (OCT) scans. The framework, trained on 14,884 scans, achieves performance comparable to or exceeding that of expert clinicians in diagnosing and referring patients with sight-threatening retinal diseases. The system uses a two-stage approach: a segmentation network that creates device-independent tissue maps, followed by a classification network that analyzes these maps to provide diagnoses and referral suggestions. The segmentation network is trained on manually segmented OCT scans, while the classification network is trained on scans with confirmed diagnoses and optimal referral decisions. The framework demonstrates robust performance across different OCT devices, maintaining accuracy even when using tissue segmentations from a different type of device. It also shows high accuracy in diagnosing multiple retinal pathologies and achieving expert-level performance in referral decisions. The system was tested on two independent datasets of 997 and 116 patients, achieving an error rate of 5.5% on the first dataset and 3.4% on the second. The framework's results were compared to those of expert clinicians, showing that it performs as well as or better than human experts in most cases. The system's ability to generalize across different OCT devices and its potential for clinical application in various settings make it a promising tool for improving the efficiency and accuracy of retinal disease diagnosis and referral. The framework also has the potential to be used in other areas of medicine where medical imaging is involved, and its results could be used to train healthcare professionals to expert levels. The study highlights the potential of deep learning in clinical settings, demonstrating its ability to provide accurate and reliable diagnoses and referrals, even in complex and ambiguous cases. The framework's results suggest that deep learning can be a valuable tool in the diagnosis and management of retinal diseases, with the potential to improve patient outcomes and reduce the workload of healthcare professionals.