The Impact of Image Input Size on the Efficiency and Suitability of Deep Learning Models in Diabetic Retinopathy

• Thiago Araújo UFRGS
• Beatriz Schaan UFRGS / HCPA
• Carla M. D. S. Freitas UFRGS
• Philippe Navaux UFRGS

DOI: https://doi.org/10.5753/eramiars.2025.16713

Resumo

The high computational cost and environmental impact of deep learning (DL) models hinder their adoption for diabetic retinopathy (DR) classification, underscoring the need for models that strike a balance between performance and efficiency. This study evaluates how image input size affects the effectiveness of convolutional neural networks (CNNs) in classifying DR. Using a dataset of 2,579 Brazilian fundus images from the Federal University of São Paulo (UNIFESP), we assessed the performance of the EfficientNetV2B0 and MobileNet models across five input image sizes. We measured the area under the ROC curve (AUC), energy consumption, and training time. For the MobileNet model, an input resolution of 300×300 achieved an AUC of 0.91, resulting in a 69% reduction in energy consumption and a 74% reduction in training time. EfficientNetV2B0 achieved an AUC of 0.93 at the same resolution, with reductions in energy consumption and training time of 66% and 70%, respectively. These results highlight the importance of optimizing input size as a simple yet effective way to improve the efficiency and sustainability of DL diagnostic models while maintaining high predictive accuracy.

Referências

Asiri, A. A., Shaf, A., Ali, T., Aamir, M., Irfan, M., and Alqahtani, S. (2024). Enhancing brain tumor diagnosis: an optimized cnn hyperparameter model for improved accuracy and reliability. PeerJ Computer Science, 10:e1878.

Bahrami, A., Rakhshaninejad, M., Ghousi, R., and Atashi, A. (2025). Enhancing machine learning performance in cardiac surgery icu: Hyperparameter optimization with metaheuristic algorithm. PloS one, 20(2):e0311250.

Bolón-Canedo, V., Morán-Fernández, L., Cancela, B., and Alonso-Betanzos, A. (2024). A review of green artificial intelligence: Towards a more sustainable future. Neurocomputing, 599:128096.

Julian, A. and Devipriya, R. (2024). Exploring hyperparameter tuning strategies for optimizing model performance. In 2024 International Conference on Trends in Quantum Computing and Emerging Business Technologies, pages 1–4. IEEE.

Viviani, L. A. and Ranganathan, P. (2024). Evaluating the suitability of lstm models for edge computing. In 2024 Cyber Awareness and Research Symposium (CARS), pages 1–7. IEEE.

Wojciuk, M., Swiderska-Chadaj, Z., Siwek, K., and Gertych, A. (2024). Improving classification accuracy of fine-tuned cnn models: Impact of hyperparameter optimization. Heliyon, 10(5).