COMPARATIVE ANALYSIS OF DUAL-EYE CNN MODELS AND AN EXPLAINABLE MULTIMODAL CNN–VISION TRANSFORMER FOR OPHTHALMIC DISEASE DETECTION
International Journal of Drug Delivery Technology · 10.06.2026 · 16 (53s)
DOI: 10.25258/ijddt.16.53s.37 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Drug Delivery Technology, online erschienen am 10.06.2026. Verfasst von Umesh Lakhtariya und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Deep learning for the detection of ophthalmic disease has drawn much attention due to its promise for early diagnosis and automatic screening. Among existing approaches, dual-eye convolutional neural network (CNN) architectures have obtained promising performance by exploiting the bilateral ocular information from retinal images. However, the generalization capability and robustness of such models are still limited, especially across diverse datasets and disease categories. In this paper we report an extensive comparative study of dual-eye CNNbased models (DenseNet121, ResNet50) and traditional machine learning methods (Logistic Regression, Support Vector Machine, Random Forest, XGBoost) over several ophthalmic datasets. The experimental results show that the CNN-based models achieve higher accuracy and F1-score than the traditional methods, but there are still large performance gaps in recall and class-wise generalization, especially for complicated retinal conditions. To tackle these limitations, we propose an explainable multimodal hybrid framework that combines convolutional neural networks and Vision Transformers (ViT) with feature-level fusion. The proposed method aims to combine the ability of CNNs to extract spatial features with the ability of transformers to model global context, in addition to multimodal data including OCT, fundus images and clinical data. Moreover, explainable AI (XAI) techniques are expected to improve model interpretability and clinical trust. The results from our experimental analysis underline the importance of sophisticated hybrid architectures and strongly motivate the proposed framework. This work lays the groundwork for the development of more robust, interpretable and clinically applicable AI systems for ophthalmic disease detection.
Zitieren
APA
Lakhtariya, U., & Upadhyaya, A. N. (2026). COMPARATIVE ANALYSIS OF DUAL-EYE CNN MODELS AND AN EXPLAINABLE MULTIMODAL CNN–VISION TRANSFORMER FOR OPHTHALMIC DISEASE DETECTION. International Journal of Drug Delivery Technology, 16(53s). https://doi.org/10.25258/ijddt.16.53s.37
Vancouver
Lakhtariya U, Upadhyaya AN. COMPARATIVE ANALYSIS OF DUAL-EYE CNN MODELS AND AN EXPLAINABLE MULTIMODAL CNN–VISION TRANSFORMER FOR OPHTHALMIC DISEASE DETECTION. International Journal of Drug Delivery Technology. 2026;16(53s). doi:10.25258/ijddt.16.53s.37
BibTeX
@article{Lakhtariya2026,
author = {Lakhtariya, Umesh and Upadhyaya, Ajay N.},
title = {COMPARATIVE ANALYSIS OF DUAL-EYE CNN MODELS AND AN EXPLAINABLE MULTIMODAL CNN–VISION TRANSFORMER FOR OPHTHALMIC DISEASE DETECTION},
journal = {International Journal of Drug Delivery Technology},
year = {2026},
volume = {16},
number = {53s},
doi = {10.25258/ijddt.16.53s.37}
}
EUVIMED