EUVIMEDEUVIMED entdecken
Crossref

A Robust Preprocessing Pipeline for Reliable NPDR Analysis from Retinal Fundus Images

Swarup Suresh Kulkarni, Sharanabasava Inamadar

International Journal of Drug Delivery Technology · 11.06.2026 · 16 (52s)

DOI: 10.25258/ijddt.16.52s.8 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Drug Delivery Technology, online erschienen am 11.06.2026. Verfasst von Swarup Suresh Kulkarni und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The earliest treatable stage of diabetic retinopathy is Non-Proliferative Diabetic Retinopathy (NPDR) where the consistency and quality of the retinal fundus images are critical for its detection. But in practice, fundus data often suffer from non-uniform illumination, poor contrast, noise and artifacts, which significantly reduce the visibility of retinal lesions and result in the reduction of accuracy of automatic diagnostic systems. Current literature is more concerned with classification schemes than it is with a standardized preprocessing scheme that is specifically designed to boost NPDR sensitive features. This paper presents a strong and standardized preprocessing pipeline that will help to enhance the quality of retinal fundus images and increase the visibility of early NPDR lesions. The given framework combines the region-of-interest extraction, green channel enhancement, contrast-limited adaptive histogram equalization (CLAHE), adaptive noise filtering, illumination normalization, and intensity standardization into one workflow. The pipeline is trained to retain delicate pathological features, including microaneurysms and exudates, unlike traditional preprocessing methods and minimizes inter-dataset variations. Experimental analysis performed on benchmark retinal images show a significant improvement of image quality and diagnostic preparation. The suggested pipeline showed significant improvements in contrast enhancement and structural similarity indicators and PSNR and SSIM values were improved relative to raw images. Moreover, downstream NPDR classification performance was found to have statistically significant improvement upon receiving preprocessed images. The Ablation analysis proves that every preprocessing step leads to an improved representation of lesions and stability of the model. The presented framework provides a dataset-free preprocessing standard, enhancing the robustness, generalization, and clinical reliability of AI-based NPDR screening systems, which can be scaled to support the deployment of AI-based ophthalmic diagnostics in practice

Zitieren

APA

Kulkarni, S. S., & Inamadar, S. (2026). A Robust Preprocessing Pipeline for Reliable NPDR Analysis from Retinal Fundus Images. International Journal of Drug Delivery Technology, 16(52s). https://doi.org/10.25258/ijddt.16.52s.8

Vancouver

Kulkarni SS, Inamadar S. A Robust Preprocessing Pipeline for Reliable NPDR Analysis from Retinal Fundus Images. International Journal of Drug Delivery Technology. 2026;16(52s). doi:10.25258/ijddt.16.52s.8

BibTeX
@article{Kulkarni2026,
  author = {Kulkarni, Swarup Suresh and Inamadar, Sharanabasava},
  title = {A Robust Preprocessing Pipeline for Reliable NPDR Analysis from Retinal Fundus Images},
  journal = {International Journal of Drug Delivery Technology},
  year = {2026},
  volume = {16},
  number = {52s},
  doi = {10.25258/ijddt.16.52s.8}
}
Kostenlos starten →