EUVIMEDEUVIMED entdecken
Crossref

Multi-Map Attention Ensemble Network (MMAEN) for Automated Keratoconus Detection using Corneal Topography Images: A Deep Learning Approach with Cross-Map Attention Mechanism

Shalini R. Bakal, Satish R. Sankaye, Nagsen S. Bansod

International Journal of Drug Delivery Technology · 30.04.2026 · 16 (23s)

DOI: 10.25258/ijddt.16.23s.18 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Drug Delivery Technology, online erschienen am 30.04.2026. Verfasst von Shalini R. Bakal und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Keratoconus is a corneal progressive disorder, and if it is not detected earlier, can also lead to visual impairment. There are numerous traditional diagnosis techniques of corneal topography, which helps interpret the KC at clinical level. Therefore, if we are doing early-stage keratoconus detection, there may be a little difference in accuracies. In this study, we have proposed a novel model which includes multi-map attention ensemble network (MMAEN). Here we are using multiple corneal topographic maps and cross-map attention mechanism which helps for automated detection of keratoconus. MMAEN includes seven parallel EfficientNet-B0 feature extractors, where they process corneal maps 224x224x3 pixels. These features are integrated using cross-map attention mechanism, where the features of these maps also have optimal weighting. In this architecture, 5.27 million parameters are present, and we have used uncertainty quantification, and we have used Monte Carlo dropout method for uncertainty quantification. The evaluation of this model includes experimental configuration, that is original corneal topography images and pre-processed images, which includes ImageJ software. The pre-processed image configuration has achieved 97% test accuracy with 94% keratoconus sensitivity and 100% normal specificity. The original image configuration has achieved 94% accuracy and 88% sensitivity, which achieved 3% accuracy improvement with false negative, with 50% decrease in the false negative. In these two configurations, AUC score is very high for original it is 0.996 and for pre-processed it is 0.992. As well, we have calculated Cohen's kappa values 0.880 and 0.940 respectively, and that is why the reliability of model is achieved. MMAEN is an innovative multi-map attention framework, that is used for the automated keratoconus detection which gives superior performance. The pre-processing on the images eventually reduces false negatives, which gives early disease detection at the clinical level. This advance model also includes uncertainty quantification, which gives confidence estimates while predictions, also improves the clinical utility

Zitieren

APA

Bakal, S. R., Sankaye, S. R., & Bansod, N. S. (2026). Multi-Map Attention Ensemble Network (MMAEN) for Automated Keratoconus Detection using Corneal Topography Images: A Deep Learning Approach with Cross-Map Attention Mechanism. International Journal of Drug Delivery Technology, 16(23s). https://doi.org/10.25258/ijddt.16.23s.18

Vancouver

Bakal SR, Sankaye SR, Bansod NS. Multi-Map Attention Ensemble Network (MMAEN) for Automated Keratoconus Detection using Corneal Topography Images: A Deep Learning Approach with Cross-Map Attention Mechanism. International Journal of Drug Delivery Technology. 2026;16(23s). doi:10.25258/ijddt.16.23s.18

BibTeX
@article{Bakal2026,
  author = {Bakal, Shalini R. and Sankaye, Satish R. and Bansod, Nagsen S.},
  title = {Multi-Map Attention Ensemble Network (MMAEN) for Automated Keratoconus Detection using Corneal Topography Images: A Deep Learning Approach with Cross-Map Attention Mechanism},
  journal = {International Journal of Drug Delivery Technology},
  year = {2026},
  volume = {16},
  number = {23s},
  doi = {10.25258/ijddt.16.23s.18}
}
Kostenlos starten →