Unified image preprocessing and attention-enhanced deep learning framework for cross-device multi-stage retinopathy of prematurity classification
Background Retinopathy of Prematurity (ROP) is an important retinal disorder in premature infants and a major cause of preventable childhood blindness. Neonatal fundus images exhibit substantial variation across imaging systems, complicating automated analysis. Methods This study presents a unified multi-source framework for five-class ROP image classification combining preprocessing, attention-enhanced deep learning, and Grad-CAM visualization. An initial dataset of 1,151 images was divided into training, validation, and testing subsets, followed by independent offline class-wise augmentation. The pipeline combines adaptive gamma correction, PCAR-CLAHE enhancement, glare suppression, resizing, and normalization. DenseNet201 with CBAM was used for classification. Results On the 300-image augmented test subset, the model achieved an accuracy of 85%, sensitivity of 85%, specificity of 96.25%, and AUROC of 0.9587. Conclusion The framework provides a unified approach for heterogeneous neonatal fundus images with qualitative visual explanations. Findings support further investigation of cross-device generalization and prospective clinical validation.
Authors
- R Haritha
- P. A. Balakrishnan (ORCID: https://orcid.org/0000-0001-8139-9161)
- A. Anny Leema (ORCID: https://orcid.org/0000-0002-0704-2794)
- K. Sowmiya
- K. S. Prasanna Kumar
Institutions
- Vellore Institute of Technology University (IN)
Publication Details
- Journal
- Frontiers in Artificial Intelligence
- Published
- 2026-11-18
- DOI
- https://doi.org/10.3389/frai.2026.1956092
- Primary Topic
- Retinal Imaging and Analysis
- Type
- article
- Field-Weighted Citation Impact
- 0.00