EUVIMEDEUVIMED entdecken
Crossref

Optimization of deep learning architectures for differentiating cytomegalovirus infection in severe ulcerative colitis

A. R. Choe, G. Kim, B. Ju Ran, P. Yehyun, S. Ki-Nam, A. So-Hyun, J. Sung-Ae

Endoscopy · 01.04.2024 · 56 (S 02 S 02) · S97

DOI: 10.1055/s-0044-1782899 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Endoscopy, erschienen am 01.04.2024. Verfasst von A. R. Choe und 6 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Aims

Cytomegalovirus (CMV) reactivation is common in patients with severe ulcerative colitis (UC). Patients with UC exacerbated by reactivated CMV experience worse prognoses than those without CMV reactivation. However, CMV is primarily diagnosed through biopsies, and it can take quite a time to get results, making early diagnosis challenging. To address this, we have conducted research using deep learning to differentiate CMV from severe ulcerative colitis (UC), which has similar features, through endoscopic imaging, enabling early diagnosis of CMV.

Methods

This study leveraged endoscopic imaging to classify CMV and severe UC within a dataset comprising 86 cases from Ewha Womans University hospitals, deploying 7 convolutional neural networks. The training utilized a 4:1 ratio, enhanced by a 5-fold cross-validation, to counter the dataset's constraints. Networks underwent optimization for 10 epochs, a batch size of 10, sigmoid activation, and a learning rate of 1e-4. To improve the performance of the network, we employed augmentation techniques such as random rotations and flipping, and performed brightness standardization and resizing of resolution. Additionally, to distinguish lesions that are difficult to differentiate with RGB values, we converted the images to HSV values to separately process areas of reflected light and ulcerated regions.

Results

Among the total of 7 networks used in this study, Densenet121 showed the best performance with an accuracy of 0.8267 (Precision 0.7979). Densenet201 and MobileNetV2 also demonstrated the next best performances with accuracies of 0.7857 and 0.7143, respectively.The standardization and augmentation performed during the preprocessing process have been confirmed to improve the network's performance, and the network's robustness was verified through 5-fold cross-validation.

Conclusions

The research underscores the potential of deep learning models in differentiating CMV from severe UC through endoscopy images, indicating a promising direction for non-invasive diagnostics and timely treatment interventions. Publication History Article published online: 15 April 2024 © 2024. European Society of Gastrointestinal Endoscopy. All rights reserved. Georg Thieme Verlag KG Rüdigerstraße 14, 70469 Stuttgart, Germany

Zitieren

APA

Choe, A. R., Kim, G., Ran, B. J., Yehyun, P., Ki-Nam, S., So-Hyun, A., & Sung-Ae, J. (2024). Optimization of deep learning architectures for differentiating cytomegalovirus infection in severe ulcerative colitis. Endoscopy, 56(S 02 S 02), S97. https://doi.org/10.1055/s-0044-1782899

Vancouver

Choe AR, Kim G, Ran BJ, Yehyun P, Ki-Nam S, So-Hyun A, et al. Optimization of deep learning architectures for differentiating cytomegalovirus infection in severe ulcerative colitis. Endoscopy. 2024;56(S 02 S 02):S97. doi:10.1055/s-0044-1782899

BibTeX
@article{Choe2024,
  author = {Choe, A. R. and Kim, G. and Ran, B. Ju and Yehyun, P. and Ki-Nam, S. and So-Hyun, A. and Sung-Ae, J.},
  title = {Optimization of deep learning architectures for differentiating cytomegalovirus infection in severe ulcerative colitis},
  journal = {Endoscopy},
  year = {2024},
  volume = {56},
  number = {S 02 S 02},
  pages = {S97},
  doi = {10.1055/s-0044-1782899}
}
Kostenlos starten →