IRO Journals
Journal of Innovative Image Processing

A Soft-Voting Ensemble with Grad-CAM Explainability for Diabetic Retinopathy Grading

Open Access
Published: 09 October, 2026
Pages: 1441-1469
  • Padmapriya M R. , Padmapriya M R.
    Research Scholar, Presidency School of Artificial intelligence and Advanced Computing, Presidency University, Bangalore, India
    Research Scholar, Presidency School of Artificial intelligence and Advanced Computing, Presidency University, Bangalore, India
  • Vijayakumar Adaickalam Vijayakumar Adaickalam
    Professor, Presidency School of Artificial intelligence and Advanced Computing, Presidency University, Bangalore, India
    Professor, Presidency School of Artificial intelligence and Advanced Computing, Presidency University, Bangalore, India
view PDF
view PDF

How to Cite

M R., Padmapriya, and Vijayakumar Adaickalam. 2026. “A Soft-Voting Ensemble With Grad-CAM Explainability for Diabetic Retinopathy Grading”. Journal of Innovative Image Processing 8 (4): 1441-69. https://doi.org/10.36548/jiip.2026.4.006.
ACM ACS APA ABNT Chicago Harvard IEEE MLA Turabian Vancouver AMA

Download Citation

Endnote/Zotero/Mendeley (RIS) BibTeX

Keywords

Diabetic Retinopathy
Deep Ensemble Models
Explainable AI
Gradient-weighted Class Activation Mapping (Grad-CAM)
Feedback-Guided Refinement
Medical Imaging

Abstract

The demand for automated Diabetic Retinopathy (DR) assessments has increased due to gap between available specialists and the demand for DR assessments. To address this, an ensemble of deep learning models has been developed which is able to provide a clinically relevant DR classifications across various retina image datasets with consistency and reliability. The ensemble comprises four different convolutional models: EfficientNet-B0; EfficientNet-B3; EfficientNet-B4 and DenseNet-201. These models have been combined soft-voting probability fusion method, which leverages their individual strengths. To mitigate significant class imbalance in DR grading, the models were trained using weighted cross-entropy loss and stratified sampling to maximize sensitivity to minority classes. Grad-CAM (Gradient-weighted Class Activation Mapping) is applied to explain model predictions, while a feedback-guided refinement mechanism, rather than a formal reinforcement learning algorithm, incorporates human feedback for continuous —to enable ongoing model improvement. Human review of the model’s explanations led to adjustments that resulted in more accurate predictions of subtle lesions and eliminated the model’s drift from attention focus human feedback has improved the interpretability of the proposed framework in the process of successive refinement, achieving higher clinical transparency and enabling better decision-maker. The presented framework was developed using an integrated dataset of 6,029 retinal fundus images and tested with a dataset of 1,113 images. On this combined test set, the proposed ensemble model achieved an accuracy of 98.63% and demonstrated consistent performance across all DR grades. To determine the model's generalizability, further validation was conducted using the public APTOS 2019 dataset, achieving an external accuracy of 89.00% and a macro-F1 score of 81.19%. Overall, the findings from this proposed highlight the capability of human-informed ensemble learning to develop scalable and accessible DR screening resources for both clinical facilities and low-resource environments.

References

  1. Aiche, Ishaq, Youcef Brik, Bilal Attallah, Oussama Bouguerra, Abdelaziz Rabehi, Mustapha Habib, Doaa Sami Khafaga, and El-Sayed M. El-Kenawy. ”RDE-DR: Robust Deep Ensemble CNNs for Automated Diabetic Retinopathy Detection from Fundus Images.” Scientific Reports 16, no. 1 (2026): 15226.
  2. Tashrif, Md Tanjum An, Dipanjali Kundu, Mst Moriom Akter Bithee, Anichur Rahman, Fahmid Al Farid, Hezerul Abdul Karim, and Abu Saleh Musa Miah. ”Privacy-aware Diabetic Retinopathy Grading and Visual Lesion-Focused Interpretability Through Mixture-of-Experts Federated Deep Learning with Explainable AI.” Scientific Reports (2026).
  3. Menaka, S. R., Suresh Muthusamy, Prabhjot Kaur Sidhu, Abhinandan Routray, G. Uma Maheswari, and Nebojsa Bacanin. ”A Novel Diabetic Retinopathy Detection from Fundus Images Using Hybrid Quantum Convolutional Neural Network Models.” Scientific Reports (2026).
  4. Rajesh, Anand E., Oliver Q. Davidson, Cecilia S. Lee, and Aaron Y. Lee. ”Artificial Intelligence and Diabetic Retinopathy: AI Framework, Prospective Studies, Head-to-Head Validation, and Cost-Effectiveness.” Diabetes care 46, no. 10 (2023): 1728-1739.
  5. Saxena, Mudit, Pratap Narra, Mayank Saxena, and Rakhi Saxena. ”Deep Learning Ensemble Framework for Multiclass Diabetic Retinopathy Classification.” TELKOMNIKA (Telecommunication Computing Electronics and Control) 22, no. 3 (2024): 665-672.
  6. Shoaib, Mohamed R., Heba M. Emara, Jun Zhao, Walid El-Shafai, Naglaa F. Soliman, Ahmed S. Mubarak, Osama A. Omer, Fathi E. Abd El-Samie, and Hamada Esmaiel. ”Deep Learning Innovations in Diagnosing Diabetic Retinopathy: The Potential of Transfer Learning and the DiaCNN Model.” Computers in Biology and Medicine 169 (2024): 107834.
  7. Sushith, Mishmala, A. Sathiya, V. Kalaipoonguzhali, and V. Sathya. ”A Hybrid Deep Learning Framework for Early Detection of Diabetic Retinopathy Using Retinal Fundus Images.” Scientific Reports 15, no. 1 (2025): 15166.
  8. Ramesh, Radhakrishnan, and Selvarajan Sathiamoorthy. ”A Deep Learning Grading Classification of Diabetic Retinopathy on Retinal Fundus Images with Bio-Inspired Optimization.” Engineering, Technology & Applied Science Research 13, no. 4 (2023): 11248-11252.
  9. Anitha, K., P. Shanmuga Prabha, K. Sashi Rekha, M. Vigilson Prem, and J. Jegan Amarnath. ”Detecting Diabetic Retinopathy Using a Hybrid Ensemble XL Machine Model with Dual Weighted-Kernel ELM and Improved Mayfly Optimization.” Expert Systems with Applications 253 (2024): 124221.
  10. Joseph, Sanil, Jerrome Selvaraj, Iswarya Mani, Thandavarayan Kumaragurupari, Xianwen Shang, Poonam Mudgil, Thulasiraj Ravilla, and Mingguang He. ”Diagnostic Accuracy of Artificial Intelligence-Based Automated Diabetic Retinopathy Screening in Real-World Settings: A Systematic Review and Meta-Analysis.” American journal of ophthalmology 263 (2024): 214-230.
  11. Kong, Mingui, and Su Jeong Song. ”Artificial Intelligence Applications in Diabetic Retinopathy: What We Have Now and What to Expect in the Future.” Endocrinology and Metabolism 39, no. 3 (2024): 416-424.
  12. Chopra, Muskaan, Lorenz Sparrenberg, Armin Berger, Sarthak Khanna, Jan H. Terheyden, and Rafet Sifa. ”From Retinal Pixels to Patients: Evolution of Deep Learning Research in Diabetic Retinopathy Screening.” In 2025 IEEE International Conference on Big Data (BigData), IEEE, 2025, 7045-7054.
  13. Lim, Wei Xiang, ZhiYuan Chen, and Amr Ahmed. ”The Adoption of Deep Learning Interpretability Techniques on Diabetic Retinopathy Analysis: A Review.” Medical & biological engineering & computing 60, no. 3 (2022): 633-642.
  14. Herrero-Tudela, Maria, Roberto Romero-Oraa, Roberto Hornero, Gonzalo C. Gutierrez Tobal, Maria I. Lopez, and Maria Garcia. ”An Explainable Deep-Learning Model Reveals Clinical Clues in Diabetic Retinopathy Through SHAP.” Biomedical Signal Processing and Control 102 (2025): 107328.
  15. Djoumessi, Kerol, Ziwei Huang, Laura Kühlewein, Annekatrin Rickmann, Natalia Simon, Lisa M. Koch, and Philipp Berens. ”An Inherently Interpretable AI Model Improves Screening Speed and Accuracy for Early Diabetic Retinopathy.” PLOS Digital Health 4, no. 5 (2025): e0000831.
  16. Yao, Jie, Joshua Lim, Gilbert Yong San Lim, Jasmine Chiat Ling Ong, Yuhe Ke, Ting Fang Tan, Tien-En Tan, Stela Vujosevic, and Daniel Shu Wei Ting. ”Novel Artificial Intelligence Algorithms for Diabetic Retinopathy and Diabetic Macular Edema.” Eye and Vision 11, no. 1 (2024): 23.
  17. Hu, Mingzhe, Jiahan Zhang, Luke Matkovic, Tian Liu, and Xiaofeng Yang. ”Reinforcement Learning in Medical Image Analysis: Concepts, Applications, Challenges, and Future Directions.” Journal of Applied Clinical Medical Physics 24, no. 2 (2023): e13898.
  18. Fountoukidou, Tatiana, and Raphael Sznitman. ”A Reinforcement Learning Approach for VQA Validation: An Application to Diabetic Macular Edema Grading.” Medical image analysis 87 (2023): 102822.
  19. Yu, Chao, Jiming Liu, Shamim Nemati, and Guosheng Yin. ”Reinforcement Learning in Healthcare: A Survey.” ACM Computing Surveys (CSUR) 55, no. 1 (2021): 1-36.
  20. Porwal, Prasanna, Samiksha Pachade, Ravi Kamble, Manesh Kokare, Girish Deshmukh, Vivek Sahasrabuddhe, and Fabrice Meriaudeau. ”Indian Diabetic Retinopathy Image Dataset (IDRiD): A Database for Diabetic Retinopathy Screening Research.” Data 3, no. 3 (2018): 25.
  21. Nagpal, Dimple, Surya Narayan Panda, Muthukumaran Malarvel, Priyadarshini A. Pattanaik, and Mohammad Zubair Khan. ”A Review of Diabetic Retinopathy: Datasets, Approaches, Evaluation Metrics and Future Trends.” Journal of King Saud University Computer and Information Sciences 34, no. 9 (2022): 7138-7152.
  22. Decencière, Etienne, Xiwei Zhang, Guy Cazuguel, Bruno Lay, Béatrice Cochener, Caroline Trone, Philippe Gain et al. ”Feedback on a Publicly Distributed Image Database: The Messidor Database.” Image Analysis & Stereology (2014): 231-234.
  23. Wang, Zijian, Yi Wang, Chun Ma, Xuan Bao, and Ya Li. ”Diabetic Retinopathy Classification Using a Multi-Attention Residual Refinement Architecture.” Scientific Reports 15, no. 1 (2025): 29266.
  24. Vijayalakshmi, S., J. Samuel Manoharan, B. Nivetha, and A. Sathiya. ”Multi-task Deep Learning Framework Combining CNN: Vision Transformers and PSO for Accurate Diabetic Retinopathy Diagnosis and Lesion Localization.” Scientific Reports 15, no. 1 (2025): 35076.
  25. Saadna, Yassmina, Saliha Mezzoudj, and Meriem Khelifa. ”Efficient Transformer Architectures for Diabetic Retinopathy Classification from Fundus Images: DR-MobileViT, DR-EfficientFormer, and DR-SwinTiny.” Informatica 49, no. 29 (2025): 1-16.
  26. Sekar, Priyadharshini, Ramasubramanian Bhoopalan, N. Nagaprasad, Tadesse Regassa Mamo, S. P. Dhanabal, and Ramaswamy Krishnaraj. ”D-TNet: A Hybrid Dense Net-Transformer Model for Robust Diabetic Retinopathy Detection.” Scientific Reports 15, no. 1 (2025): 39594.
  27. Sushith, Mishmala, Ajanthaa Lakkshmanan, M. Saravanan, and S. Castro. ”Attention Dual Transformer with Adaptive Temporal Convolutional for Diabetic Retinopathy Detection.” Scientific Reports 15, no. 1 (2025): 7694.
  28. Abbasi, Rashid, Farhan Amin, Amerah Alabrah, Gyu Sang Choi, Salabat Khan, Md Belal Bin Heyat, Muhammad Shahid Iqbal, and Huiling Chen. ”Diabetic Retinopathy Detection Using Adaptive Deep Convolutional Neural Networks on Fundus Images.” Scientific Reports 15, no. 1 (2025): 24647.
  29. Asia Pacific Tele-Ophthalmology Society, ”APTOS 2019 Blindness Detection,” Kaggle, 2019. [Online]. Available: https://www.kaggle.com/c/aptos2019-blindness-detection
  30. Valarmathi Srinivasan, Vijayabhanu Rajagopal, ”An Optimization Method for Tuning Hyper-Parameters of SGAN with Ensemble Classification Regression Model,” International Journal of Electrical and Electronics Engineering, vol. 10, no. 4, 2023, 24-36. Crossref, https://doi.org/10.14445/23488379/IJEEE-V10I4P103.