Early diagnosis of retinal pathologies is critical for preventing irreversible blindness, particularly in rare conditions with limited labeled medical data. Traditional diagnostics employ a single imaging modality, limiting the identification of heterogeneous anomalies in the retina. DMRD-Net, a diagnostic system is presented that integrates spectral-domain optical coherence tomography with fundus photographs, utilizing two parallel branches of a neural network, that is EfficientNet-B0 encoders and few-shot episodic meta -learning module based on Prototypical Networks, that merge their outputs to enhance the precision of diagnosis. Supervised learning methodologies are employed to identify common retinal diseases, followed by the application of meta -learning technique, referred to as Prototypical Networks, to aggregate a limited set of data for the study of rare diseases. To support clinical confidence and improve transparency, explainable artificial intelligence is utilized to facilitate decision-making by models. It facilitated the evaluation of performance on both common and rare retinal disorders. The system achieved over 96% episodic accuracy in diagnosing rare conditions, including Macular Hole, Retinitis Pigmentosa, and Stargardt Disease, in Central Serous Chorioretinopathy. The overall classification accuracy for common diseases was 96.5%. Overall, DMRD-Net is a unified, data-efficient, and interpretable multimodal diagnostic system that works well for both common and rare retinal disorders.

Abstract Many domains require scalable algorithms that help to determine diagnoses efficiently and often within predefined time limits. Anytime diagnosis is able to determine solutions in such a way and thus is especially useful in real-time scenarios such as production scheduling, robot control, and communication networks management where diagnosis and corresponding reconfiguration capabilities play a major role. Anytime diagnosis in many cases comes along with a trade-off between diagnosis quality and the efficiency of diagnostic reasoning. In this paper we introduce and analyze FlexDiag which is an anytime direct diagnosis approach. We evaluate the algorithm with regard to performance and diagnosis quality using a configuration benchmark from the domain of feature models and an industrial configuration knowledge base from the automotive domain. Results show that FlexDiag helps to significantly increase the performance of direct diagnosis search with corresponding quality tradeoffs in terms of minimality and accuracy.