기본 정보
연구 분야
프로젝트
논문
구성원
article|
인용수 0
·2025
Enhanced Diffusion Model with Multi-Level Embeddings for Medical Image Data Augmentation in Skin Disease
Mujung Kim, Jisang Yoo, Soonchul Kwon, Byung Jun Kim, Changsik John Pak, Chong Hyun Won, Suk‐Ho Moon, Woo Jin Song, Han Gyu, Kyung Hee Park
초록

Deep learning has enabled applications in medical diagnosis, education, and research. However, obtaining large-scale, high-quality data remains challenging due to privacy regulations and the scarcity of rare disease data. Recent approaches focus on deep learning-based image generation models to create synthetic data, increasing its diversity and quality for medical applications. This study proposes an improved diffusion-based model for high-quality image generation across diverse domains. Inspired by the 8-channel VAE from Mefusion, we modified the VAE structure in Stable Diffusion to reduce artifacts. To address the loss of detailed representations in the Latent Diffusion model's compression process, we introduced multilevel embeddings and adapter layers. These additions improved synthetic data quality in the dermatology domain. Using the HAM10000 dataset, we generated synthetic data for seven skin disease conditions and conducted classification experiments to evaluate its utility. The classification accuracy using synthetic data alone was comparable to using original data. Training with both synthetic and original data improved accuracy from 87% to 90%. Our results confirm that synthetic data from our diffusion model is effective for dermatological training. Visual and quantitative evaluations further highlight its potential for medical applications.

키워드
Computer scienceImage (mathematics)DiffusionDiseaseData modelingArtificial intelligenceComputer visionMedicineDatabasePathology
타입
article
IF / 인용수
- / 0
게재 연도
2025

주식회사 디써클

대표 장재우,이윤구서울특별시 강남구 역삼로 169, 명우빌딩 2층 (TIPS타운 S2)대표 전화 0507-1312-6417이메일 info@rndcircle.io사업자등록번호 458-87-03380호스팅제공자 구글 클라우드 플랫폼(GCP)

© 2026 RnDcircle. All Rights Reserved.