Diabetic retinopathy (DR) is a leading cause of preventable blindness worldwide, necessitating accurate and reliable automated severity grading systems for early diagnosis and timely intervention. However, existing deep learning methods often exhibit limited lesion localization, poor cross-dataset generalization, inadequate modeling of the ordinal progression of disease severity, and limited prediction reliability. To address these challenges, this study proposes a lesion-guided domain-adaptive framework that integrates an EfficientNetV2-S backbone for retinal feature extraction, lesion-guided attention, adversarial domain adaptation, Earth Mover's Distance-based ordinal learning, Monte Carlo (MC) Dropout for uncertainty estimation, and Grad-CAM for model interpretability. Experiments were conducted on a harmonized dataset created by combining the APTOS2019 and MESSIDOR datasets, with the original labels consolidated into three severity classes: No DR, Mild DR, and Severe DR. The proposed framework achieved an accuracy of 90.26%, precision of 90.10%, recall of 90.26%, F1-score of 90.14%, and a Quadratic Weighted Kappa (QWK) score of 89.28%. Uncertainty-aware evaluation using MC Dropout achieved 89.89% accuracy, 88.99% QWK, and a mean predictive entropy of 0.2571. Comparative evaluations, backbone comparisons, and ablation studies further demonstrated the effectiveness of the proposed components. These results indicate that the proposed framework provides accurate, robust, interpretable, and reliable DR severity assessment across heterogeneous retinal image datasets.
Amol C. Adamuthe, Shilpa Bhagat, Vijay H. Kalmani· Cureus Journal of Computer S...· 0 citations
The rapid growth of e-commerce websites has caused the number of customer reviews to rapidly increase, making the analysis of such reviews more challenging than ever before. Existing approaches to summarizing the content of customer reviews primarily focus on lexical similarity while paying little attention to product aspect preservation, sentiment alignment, and redundant information filtering. In this work, we propose a Sentiment-Aware Redundancy-Minimized Maximum Marginal Relevance (SA-RMMR) framework for creating informative extractive summaries that can preserve product aspects, customer sentiments, and informational diversity. Our framework relies on SBERT-based semantic representation of text, TF-IDF and part-of-speech-based aspect extraction, VADER-based sentiment alignment, and a redundancy-aware maximum marginal relevance ranking. In order to make our framework robust, we introduce semantic centrality estimation, adaptive retrieval confidence, reward mixing, and balance adjustment. The quality of the created summaries is estimated with conventional lexical metrics and a new metric called Composite Quality Index (CQI) that jointly evaluates semantic relevance, aspect coverage, sentiment alignment, and redundancies. We performed experiments on several Amazon product reviews datasets and have shown that our framework was the best among all methods considered in the experiment, obtaining the highest Aspect Coverage (0.835) and Sentiment Alignment (0.719) scores while having competitive semantic similarity (BERTScore-F1 = 0.825). Our framework significantly outperformed classical extractive approaches and transformer-based models, including BART, PEGASUS, and FLAN-T5, in overall quality of summaries produced by them. Ablation studies showed that semantic representation of sentences, aspect awareness, adaptive retrieval, and redundancy elimination all have positive impact on performance of the framework. The proposed SA-RMMR framework represents an efficient and interpretable approach to customer review summarization that takes into account not only semantic relevance but also product aspects, sentiment preservation, and redundant information filtering. Additionally, the proposed Composite Quality Index allows evaluation of the quality of summaries from multiple perspectives, not limited to conventional lexical overlap, which makes it a good choice for opinion mining tasks in e-commerce.
Vijay H. Kalmani, Amol C. Adamuthe, P. Bagane· Discover Artificial Intellig...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
