EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Diffusion-based skin disease data augmentation with fine-grained detail preservation and interpolation for data diversity

Mujung Kim iD, Jisang Yoo, Soonchul Kwon iD, Byung Jun Kim, Changsik John Pak, Chong Hyun Won, Suk-Ho Moon, Woo Jin Song, Han Gyu Cha, Kyung Hee Park iD

PLOS One · 03.10.2025 · 20 (10) · e0331404

DOI: 10.1371/journal.pone.0331404 ↗

Kurz gefasst

Fachartikel in der Zeitschrift PLOS One, online erschienen am 03.10.2025. Verfasst von Mujung Kim und 9 weiteren Autor:innen. Beteiligte Einrichtung: Department of Electronic Engineering, Kwangwoon University, Seoul, Republic of Korea. Hauptthemen laut MEDLINE: Skin Diseases, Image Processing, Computer-Assisted. Die Arbeit wurde bisher 7-mal zitiert (Crossref) und nennt 83 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Ministry of Science and ICT, South Korea, Kwangwoon University.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

We propose a data augmentation technique utilizing a Diffusion-based generative deep learning model to address the issue of data scarcity in skin disease diagnosis research. Specifically, we enhanced the Stable Diffusion model, a Latent Diffusion Model (LDM), to generate high-quality synthetic images. To mitigate detail loss in existing Diffusion models, we incorporated lesion area masks and improved the encoder and decoder structures of the LDM. Multi-level embeddings were applied using a CLIP encoder-based image encoder to capture detailed representations, ranging from textures to overall shapes. Additionally, we employed pre-trained segmentation and inpainting models to generate normal skin regions and used interpolation techniques to synthesize synthetic images with gradually varying visual characteristics, while having limitations for clinical use, this approach contributes to enhanced data diversity and can be used as reference material. To validate our method, we conducted classification experiments on seven skin diseases using datasets combining synthetic and real images. The results showed improvements in classification performance, demonstrating the effectiveness of the proposed technique in addressing medical data scarcity and enhancing diagnostic accuracy.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Skin DiseasesImage Processing, Computer-AssistedHumansDeep LearningSkinAlgorithms

Autor:innen und Institutionen

  • Department of Electronic Engineering, Kwangwoon University, Seoul, Republic of Korea.
  • Graduate School of Smart Convergence, Kwangwoon University, Seoul, Republic of Korea.
  • Department of Reconstructive and Plastic Surgery, Seoul National University Hospital, Seoul, Republic of Korea.
  • Department of Plastic and Reconstructive Surgery, Asan Medical Center, Seoul, Republic of Korea.
  • Department of Dermatology, Asan Medical Center, Seoul, Republic of Korea.
  • Department of Plastic and Reconstructive Surgery, Seoul St. Mary's Hospital, Seoul, Republic of Korea.

ORCID-Profile: Mujung Kim, Soonchul Kwon, Kyung Hee Park

Förderung

  • Ministry of Science and ICT, South Korea
  • Kwangwoon University

Zitieren

APA

Kim, M., Yoo, J., Kwon, S., Kim, B. J., Pak, C. J., Won, C. H., Moon, S. H., Song, W. J., Cha, H. G., & Park, K. H. (2025). Diffusion-based skin disease data augmentation with fine-grained detail preservation and interpolation for data diversity. PLOS One, 20(10), e0331404. https://doi.org/10.1371/journal.pone.0331404

Vancouver

Kim M, Yoo J, Kwon S, Kim BJ, Pak CJ, Won CH, et al. Diffusion-based skin disease data augmentation with fine-grained detail preservation and interpolation for data diversity. PLOS One. 2025;20(10):e0331404. doi:10.1371/journal.pone.0331404

BibTeX
@article{Kim2025,
  author = {Kim, Mujung and Yoo, Jisang and Kwon, Soonchul and Kim, Byung Jun and Pak, Changsik John and Won, Chong Hyun and Moon, Suk-Ho and Song, Woo Jin and Cha, Han Gyu and Park, Kyung Hee},
  title = {Diffusion-based skin disease data augmentation with fine-grained detail preservation and interpolation for data diversity},
  journal = {PLOS One},
  year = {2025},
  volume = {20},
  number = {10},
  pages = {e0331404},
  doi = {10.1371/journal.pone.0331404}
}
Kostenlos starten →