TY  - JOUR
AU  - Saeed, Muhammad Usman
AU  - Dastgir, Aqsa
AU  - Ghani, Muhammad Ahmad Nawaz Ul
AU  - Manzoor, Arslan
PY  - 2025
DA  - 2025/10/25
TI  - RetinoNet: An Efficient MobileNetV3-Based Model for Diabetic Retinopathy Detection Using Multi-Scale Feature Fusion
JO  - Journal of Artificial Intelligence in Bioinformatics
T2  - Journal of Artificial Intelligence in Bioinformatics
JF  - Journal of Artificial Intelligence in Bioinformatics
VL  - 1
IS  - 2
SP  - 58
EP  - 68
DO  - 10.62762/JAIB.2025.322062
UR  - https://www.icck.org/article/abs/JAIB.2025.322062
KW  - diabetic retinopathy
KW  - feature fusion
KW  - bio-informatics
KW  - multi-scale
AB  - Diabetic retinopathy (DR) is a leading cause of blindness globally, requiring timely detection and classification to prevent vision loss. Deep learning techniques offer significant potential for automating DR detection by analyzing retinal fundus images with high precision. This paper proposes a RetinoNet model that consists of MobileNetV3, Convolutional Block Attention Module (CBAM), Atrous Spatial Pyramid Pooling (ASPP), and Feature Pyramid Network (FPN). MobileNetV3 provides a lightweight and efficient foundation for feature extraction, while CBAM emphasizes critical spatial and channel information, enabling the detection of subtle retinal abnormalities. ASPP captures multi-scale contextual information through atrous convolutions, improving the model's ability to identify lesions of varying sizes and shapes. FPN combines hierarchical features from multiple network levels, ensuring both fine-grained details and high-level semantics are leveraged for accurate classification. The model was trained on the APTOS dataset. Evaluation metrics such as accuracy, precision, recall, and F1 score demonstrate the effectiveness of the proposed model in achieving state-of-the-art performance for DR detection and classification across five severity levels. This approach addresses computational challenges and improves generalization, making it suitable for both clinical and remote healthcare applications.
SN  - 3068-7535
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Saeed2025RetinoNet,
  author = {Muhammad Usman Saeed and Aqsa Dastgir and Muhammad Ahmad Nawaz Ul Ghani and Arslan Manzoor},
  title = {RetinoNet: An Efficient MobileNetV3-Based Model for Diabetic Retinopathy Detection Using Multi-Scale Feature Fusion},
  journal = {Journal of Artificial Intelligence in Bioinformatics},
  year = {2025},
  volume = {1},
  number = {2},
  pages = {58-68},
  doi = {10.62762/JAIB.2025.322062},
  url = {https://www.icck.org/article/abs/JAIB.2025.322062},
  abstract = {Diabetic retinopathy (DR) is a leading cause of blindness globally, requiring timely detection and classification to prevent vision loss. Deep learning techniques offer significant potential for automating DR detection by analyzing retinal fundus images with high precision. This paper proposes a RetinoNet model that consists of MobileNetV3, Convolutional Block Attention Module (CBAM), Atrous Spatial Pyramid Pooling (ASPP), and Feature Pyramid Network (FPN). MobileNetV3 provides a lightweight and efficient foundation for feature extraction, while CBAM emphasizes critical spatial and channel information, enabling the detection of subtle retinal abnormalities. ASPP captures multi-scale contextual information through atrous convolutions, improving the model's ability to identify lesions of varying sizes and shapes. FPN combines hierarchical features from multiple network levels, ensuring both fine-grained details and high-level semantics are leveraged for accurate classification. The model was trained on the APTOS dataset. Evaluation metrics such as accuracy, precision, recall, and F1 score demonstrate the effectiveness of the proposed model in achieving state-of-the-art performance for DR detection and classification across five severity levels. This approach addresses computational challenges and improves generalization, making it suitable for both clinical and remote healthcare applications.},
  keywords = {diabetic retinopathy, feature fusion, bio-informatics, multi-scale},
  issn = {3068-7535},
  publisher = {Institute of Central Computation and Knowledge}
}