EUVIMEDEUVIMED entdecken
Crossref

Multistage Preprocessing and Ensemble Classification for Diabetic Retinopathy Detection

Ahmed Hefdhi Hussein Hussein, Mudhar A. Al-Obaidi, Sadik Kamel Gharghan, Shatha A. Alsalman, Mohammed Ayad Saad

Journal of Innovative Image Processing · 06.07.2026 · 8 (3) · 827-843

DOI: 10.36548/jiip.2026.3.005 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Innovative Image Processing, online erschienen am 06.07.2026. Verfasst von Ahmed Hefdhi Hussein Hussein und 4 weiteren Autor:innen. Die Arbeit nennt 17 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Diabetic retinopathy (DR) is a leading cause of preventable blindness among people with diabetes and demands improved automated screening. The principal challenge addressed here is the variability of fundus-image quality, which limits the reliability of automated DR detection. A multistage image enhancement and classification framework is proposed. First, the input image is preprocessed by dark-border cropping, circular retinal region extraction, Ben Graham's weighted-Gaussian contrast enhancement, and Contrast Limited Adaptive Histogram Equalization (CLAHE) on the CIELAB L-channel, followed by resizing to 640×640 pixels. Second, a fine-tuned ResNet50V2 network with appended Global-Average-Pooling, dropout, and dense layers produces a 2048-dimensional feature vector. Third, a two-stage ensemble of RBF-kernel SVMs is used: (i) five one-vs-rest classifiers covering all DR classes and (ii) three additional SVMs targeting the intermediate grades (Mild, Moderate, Severe), combined by a four-rule confidence-scored decision system. All SVM hyperparameters are tuned by 5-fold stratified cross-validation on a training subset, and features are standardised (z-score) before classification. On the Kaggle DR Detection dataset, using an approximate 81% for training 9% for validation and 10% for testing. The model achieves an overall test accuracy of 83% and an F_1-score of 0.91 for the No DR class. The proposed hybrid deep learning and SVM approach performs competitively for grading diabetic retinopathy, and also offers transparency in terms of evaluation.

Zitieren

APA

Hussein, A. H. H., Al-Obaidi, M. A., Gharghan, S. K., Alsalman, S. A., & Saad, M. A. (2026). Multistage Preprocessing and Ensemble Classification for Diabetic Retinopathy Detection. Journal of Innovative Image Processing, 8(3), 827-843. https://doi.org/10.36548/jiip.2026.3.005

Vancouver

Hussein AHH, Al-Obaidi MA, Gharghan SK, Alsalman SA, Saad MA. Multistage Preprocessing and Ensemble Classification for Diabetic Retinopathy Detection. Journal of Innovative Image Processing. 2026;8(3):827-843. doi:10.36548/jiip.2026.3.005

BibTeX
@article{Hussein2026,
  author = {Hussein, Ahmed Hefdhi Hussein and Al-Obaidi, Mudhar A. and Gharghan, Sadik Kamel and Alsalman, Shatha A. and Saad, Mohammed Ayad},
  title = {Multistage Preprocessing and Ensemble Classification for Diabetic Retinopathy Detection},
  journal = {Journal of Innovative Image Processing},
  year = {2026},
  volume = {8},
  number = {3},
  pages = {827--843},
  doi = {10.36548/jiip.2026.3.005}
}
Kostenlos starten →