Predicting Wet Age-Related Macular Degeneration Recurrence Prediction Using CNN Models With Multimodal Retinal Imaging
Digital Health Research · 2025
DOI: 10.61499/dhr.2025.3.e2 ↗Kurz gefasst
Fachartikel in der Zeitschrift Digital Health Research, 2025. Verfasst von Won Tae Yoon und 5 weiteren Autor:innen. Die Arbeit nennt 23 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: National Research Foundation of Korea.
Abstract (Originalsprache)
Background
Wet age-related macular degeneration (wet AMD) is a vision-threatening condition that typically develops with age.The standard treatment involves intravitreal injections, but suboptimal timing of these injections can result in severe outcomes, including irreversible vision loss.While numerous studies have focused on the detection of wet AMD, the prediction of its recurrence remains relatively underexplored.Existing approaches to recurrence prediction primarily rely on a single modality-optical coherence tomography (OCT)-and have achieved only limited prognostic accuracy.Therefore, there is a pressing need to comprehensively investigate recurrence prediction in wet AMD to improve prognostic performance.
Methods
Compared to the existing studies, we collected 3 different types of images from the patients with wet AMD; OCT vertical, OCT horizontal, and fundus.We first trained and evaluated various convolutional neural network-based models using single-modality data which are pre-and post-treatment versions of each type of the collected images.We also evaluated a dual-modality scenario using the 2 optimal performance modalities, which are fundus post-treatment and OCT horizontal pre-treatment.Finally, we examined a multimodality case which used the entire 6 modalities in our study for wet AMD recurrence prediction.Performance was assessed using area under the receiver operating characteristic curve (AUC) with several classification performance metrics.
Results
Among single-modality approaches, the OCT horizontal pre-treatment model achieved an AUC of 0.617 ± 0.045, while the fundus post-treatment model reached an accuracy of 0.612 ± 0.008.The dual-modality model combining fundus post-treatment and OCT horizontal pre-treatment images attained an AUC of 0.622 ± 0.037, whereas the multimodality model incorporating all imaging sources yielded an AUC of 0.564 ± 0.026.
Conclusion
Based on our experimental results with 3 different imaging modalities, we found that the fundus-only model achieved the optimal accuracy, and the multimodal combination of OCT and fundus yielded the highest AUC, indicating that considering multimodal imaging data would be critical in wet AMD recurrence prediction.
Förderung
- National Research Foundation of Korea
Zitieren
APA
Yoon, W. T., Kim, H., Moon, J., Ngo, D., Kim, J. H., & Cho, B. H. (2025). Predicting Wet Age-Related Macular Degeneration Recurrence Prediction Using CNN Models With Multimodal Retinal Imaging. Digital Health Research. https://doi.org/10.61499/dhr.2025.3.e2
Vancouver
Yoon WT, Kim H, Moon J, Ngo D, Kim JH, Cho BH. Predicting Wet Age-Related Macular Degeneration Recurrence Prediction Using CNN Models With Multimodal Retinal Imaging. Digital Health Research. 2025. doi:10.61499/dhr.2025.3.e2
BibTeX
@article{Yoon2025,
author = {Yoon, Won Tae and Kim, Hun-gyeom and Moon, Junhyung and Ngo, Dat and Kim, Jae Hui and Cho, Baek Hwan},
title = {Predicting Wet Age-Related Macular Degeneration Recurrence Prediction Using CNN Models With Multimodal Retinal Imaging},
journal = {Digital Health Research},
year = {2025},
doi = {10.61499/dhr.2025.3.e2}
}
EUVIMED