Purpose: This study aims to improve the diagnostic effectiveness of rare retinal diseases by introducing a novel classification approach that not only enhances accuracy but also supports the intelligent management of the diagnostic process through AI-based decision systems. Feed for the study: Retinitis pigmentosa (RP), cone-rod dystrophy (CORD), and Usher syndrome are inherited retinal disorders with low prevalence but significant clinical impact. Their early symptoms are subtle and often missed, posing serious challenges for timely diagnosis. The shortage of trained specialists and limited availability of medical imaging data further complicate diagnostic workflows and uncertainty management in clinical settings. Methodology: The proposed method combines the outputs of several deep learning models—EfficientNet, InceptionV3, and Residual Attention Vision Transformers (RS-A-ViT)—and applies intuitionistic fuzzy sets to model uncertainty and refine classification results. This fusion-based approach enables better handling of ambiguous or borderline cases and improves classification robustness despite limited datasets, which is critical for effective diagnostic workflow management. Findings: The results demonstrate a notable improvement in diagnostic performance, with classification accuracy increasing by up to 5.9 percentage points for the RS-A-ViT model. The approach proved especially beneficial in cases with overlapping visual features, effectively reducing uncertainty and increasing reliability in multi-class classification of RP, CORD, Usher syndrome, and normal cases—thus supporting more controlled and informed diagnostic decision-making. Practical implications: Beyond increasing diagnostic accuracy, the proposed method facilitates intelligent management of diagnostic workflows in ophthalmology. By providing automated triage, real-time decision support, and interpretability based on uncertainty modeling, it can alleviate the workload on specialists and enable earlier and more reliable detection of rare retinal diseases, even in resource-limited clinical environments.