Comprehensive Performance Comparison of EfficientNet-B0 and MobileNetV3 for Multiclass Fundus Image Classification

Tiffany Azhar Izzuddin; Rosa Andrie Asmara; Cahya Rahmad; Dodit Suprianto

doi:10.25299/itjrd.2026.25049

Authors

Tiffany Azhar Izzuddin Department of Information Technology, Politeknik Negeri Malang
Rosa Andrie Asmara Department of Information Technology, Politeknik Negeri Malang
Cahya Rahmad Department of Information Technology, Politeknik Negeri Malang
Dodit Suprianto Department of Electrical Engineering, Politeknik Negeri Malang

DOI:

https://doi.org/10.25299/itjrd.2026.25049

Keywords:

Medical Image Classification, Fundus Image, Deep Learning, EfficientNet-B0, MobileNetV3, Comparative Analysis, Mobile Health

Abstract

Diabetic retinopathy, glaucoma, and cataracts are leading causes of preventable blindness worldwide, creating an urgent need for efficient and accessible screening tools. While deep learning offers promising solutions, a critical gap exists in the direct comparison of state-of-the-art, lightweight architectures for the simultaneous detection of multiple pathologies from a single fundus image. This study presents a comprehensive performance evaluation of two efficient convolutional neural networks, EfficientNet-B0 and MobileNetV3-Large, for four-class fundus image classification (normal, cataract, glaucoma, diabetic retinopathy). Using a dataset of 4,217 images and an identical transfer learning protocol, we demonstrate that EfficientNet-B0 achieves superior diagnostic accuracy (87% vs. 82%) and exhibits markedly higher sensitivity in detecting glaucoma (74% vs. 57%), a critical metric for early intervention. EfficientNet-B0 also showed near-perfect precision for diabetic retinopathy (99% vs. 82%), minimizing false positives. Conversely, MobileNetV3 proved significantly faster, with an inference time approximately four times quicker (0.0085s vs. 0.032s per image), highlighting a key accuracy-speed trade-off. The superior EfficientNet-B0 model was successfully deployed into a functional mobile application, validating its practical potential for point-of-care screening. This work provides crucial empirical evidence to guide model selection, demonstrating that EfficientNet-B0 is better suited for accuracy-critical diagnostics, while MobileNetV3 is optimal for high-throughput scenarios. This balance is essential for developing practical AI-assisted tools in ophthalmic diagnosis.

Downloads

Download data is not yet available.

References

[1] H. Hashemian et al., “Application of Artificial Intelligence in Ophthalmology: An Updated Comprehensive Review,” J Ophthalmic Vis Res, vol. 19, no. 3, pp. 354–367, 2024, doi: 10.18502/jovr.v19i3.15893.

[2] A. F. Zegeye, Y. Z. Temachu, and C. K. Mekonnen, “Prevalence and factors associated with Diabetes retinopathy among type 2 diabetic patients at Northwest Amhara Comprehensive Specialized Hospitals, Northwest Ethiopia 2021,” BMC Ophthalmol, vol. 23, no. 1, Jan. 2023, doi: 10.1186/s12886-022-02746-8.

[3] J. M. Kang and A. P. Tanna, “Glaucoma,” Medical Clinics of North America, vol. 105, no. 3, pp. 493–510, May 2021, doi: 10.1016/j.mcna.2021.01.004.

[4] A. Halim et al., “Prevalence and Associated Factors of Diabetic Retinopathy in People with Type 2 Diabetes Attending Community Based Diabetic Retinopathy Screening in Greater Bandung, Indonesia,” IJRetina, vol. 5, no. 1, p. 1, Feb. 2022, doi: 10.35479/ijretina.2022.vol005.iss001.172.

[5] N. Angriani, M. N. R. Akib, F. E. Mulyadi, S. Irmandha, S. Hadi, and R. F. Syamsu, “Glaucoma Risk Factors,” GMJ, vol. 4, no. 1, pp. 1–8, Apr. 2022, doi: 10.33096/gmj.v4i1.94.

[6] J. Sahlsten et al., “Deep Learning Fundus Image Analysis for Diabetic Retinopathy and Macular Edema Grading,” Sci Rep, vol. 9, no. 1, Jul. 2019, doi: 10.1038/s41598-019-47181-w.

[7] A. Shao et al., “Generative artificial intelligence for fundus fluorescein angiography interpretation and human expert evaluation,” npj Digit. Med., vol. 8, no. 1, p. 396, Jul. 2025, doi: 10.1038/s41746-025-01759-z.

[8] R. Gargeya and T. Leng, “Automated Identification of Diabetic Retinopathy Using Deep Learning,” Ophthalmology, vol. 124, no. 7, pp. 962–969, Jul. 2017, doi: 10.1016/j.ophtha.2017.02.008.

[9] S. H. Abdullah, R. Magdalena, and R. Y. N. Fu’adah, “KLASIFIKASI DIABETIC RETINOPATHY BERBASIS PENGOLAHAN CITRA FUNDUS DAN DEEP LEARNING,” JESCE, vol. 5, no. 2, pp. 84–90, Feb. 2022, doi: 10.31289/jesce.v5i2.5659.

[10] K. Pin, J. Ho Chang, and Y. Nam, “Comparative Study of Transfer Learning Models for Retinal Disease Diagnosis from Fundus Images,” Computers, Materials & Continua, vol. 70, no. 3, pp. 5821–5834, 2022, doi: 10.32604/cmc.2022.021943.

[11] M. Sarosa, A. Badriyah, R. A. Asmara, M. K. Wardani, D. F. A. Riza, and Y. M. Azis, “Performance Analysis of MobileNET on Grape Leaf Disease Detection,” in 2024 International Conference on Advanced Information Scientific Development (ICAISD), Jawa Barat, Indonesia: IEEE, Nov. 2024, pp. 64–68. doi: 10.1109/ICAISD63055.2024.10894920.

[12] M. Tan and Q. V. Le, “EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks,” Sep. 11, 2020, arXiv: arXiv:1905.11946. doi: 10.48550/arXiv.1905.11946.

[13] A. Howard et al., “Searching for MobileNetV3,” Nov. 20, 2019, arXiv: arXiv:1905.02244. doi: 10.48550/arXiv.1905.02244.

[14] A. Géron, “Hands-on Machine Learning with Scikit-Learn, Keras, and TensorFlow,” 2019.

[15] Md. E. Rayed, S. M. S. Islam, S. I. Niha, J. R. Jim, M. M. Kabir, and M. F. Mridha, “Deep learning for medical image segmentation: State-of-the-art advancements and challenges,” Informatics in Medicine Unlocked, vol. 47, p. 101504, 2024, doi: 10.1016/j.imu.2024.101504.

[16] I. Sutskever, J. Martens, and G. Dahl, “On the importance of initialization and momentum in deep learning,” 2013.

[17] L. N. Smith, “Cyclical Learning Rates for Training Neural Networks,” 2015, arXiv. doi: 10.48550/ARXIV.1506.01186.

[18] D. Masters and C. Luschi, “Revisiting Small Batch Training for Deep Neural Networks,” 2018, arXiv. doi: 10.48550/ARXIV.1804.07612.

[19] S. Al-Fahdawi et al., “Fundus-DeepNet: Multi-label deep learning classification system for enhanced detection of multiple ocular diseases through data fusion of fundus images,” Information Fusion, vol. 102, p. 102059, Feb. 2024, doi: 10.1016/j.inffus.2023.102059.

[20] J. C. Obi, “A comparative study of several classification metrics and their performances on data,” World J. Adv. Eng. Technol. Sci., vol. 8, no. 1, pp. 308–314, Feb. 2023, doi: 10.30574/wjaets.2023.8.1.0054.

[21] M. V. Mathiazhagan and A. K. Chellappan, “An efficient optimized deep learning model for diabetic retinopathy classification,” JBRHA, vol. 39, no. 2, p. 3293, Mar. 2025, doi: 10.54517/jbrha3293.

[22] D. P. Kingma and J. Ba, “Adam: A Method for Stochastic Optimization,” 2014, arXiv. doi: 10.48550/ARXIV.1412.6980.

[23] L. Liu et al., “On the Variance of the Adaptive Learning Rate and Beyond,” 2019, arXiv. doi: 10.48550/ARXIV.1908.03265.

[24] Y. Pattanaik et al., “Exploring Optimization Dynamics: Hybrid Approaches Combining Adaptive and Traditional Techniques for Deep Learning Models,” in 2025 4th International Conference on Distributed Computing and Electrical Circuits and Electronics (ICDCECE), Ballari, India: IEEE, Apr. 2025, pp. 1–6. doi: 10.1109/ICDCECE65353.2025.11035942.

[25] S. Liu, W. Wang, L. Deng, and H. Xu, “Cnn-trans model: A parallel dual-branch network for fundus image classification,” Biomedical Signal Processing and Control, vol. 96, p. 106621, Oct. 2024, doi: 10.1016/j.bspc.2024.106621.

[26] A. Chowdhury and S. Mahfuz, “Diabetic Retinopathy Detection from Fundus Images Using Deep Convolutional Neural Networks,” in 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Lisbon, Portugal: IEEE, Dec. 2024, pp. 6362–6369. doi: 10.1109/BIBM62325.2024.10822813.

[27] S. Gupta and A. Khadka, “Diabetic Retinopathy Detection through Multiclass Classification of Fundus Image Using Convolutional Neural Network,” Pokhara Eng. Coll. J., vol. 2, no. 1, pp. 47–56, Mar. 2025, doi: 10.3126/pecj.v2i1.76831.