Abstract Purpose Automated classification of retinal diseases from optical coherence tomography (OCT) images plays a significant role in supporting early diagnosis and clinical decision‐making. In this study, a deep learning (DL) framework is developed by integrating a neutrosophic similarity score (NSS) with the EfficientNetB3 to improve image representation and classification accuracy. Methods Initially, OCT images are transformed into the neutrosophic domain, where NSS is combined with contrast limited adaptive histogram equalization (CLAHE) and applied to enhance informative structures by reducing noise, ambiguity and indeterminacy. Then the enhanced images are directly utilized for training an EfficientNetB3 model. A two‐phase training strategy is adopted, that consist of initial training with frozen base layers and then followed by fine‐tuning of deeper layers. The performance of the proposed approach is evaluated on a retinal OCT dataset and compared with that of conventional DL models trained on the original images. Results Experimental results show that the proposed NSS‐integrated EfficientNetB3 model achieved an overall accuracy of 95.25%, along with precision, recall, and F1‐score of 95.33%, 95.25% and 95.26%, respectively. The model demonstrates improved performance when compared to models trained on the original OCT images. Conclusions The results indicate that the integration of NSS with DL improves discriminative feature learning and classification performance. The proposed study highlights the effectiveness of neutrosophic‐based enhancement techniques for advancing automated retinal disease classification.
M et al. (Fri,) studied this question.