EUVIMEDEUVIMED entdecken
Open Access

FairDITA: Disentangled Image-Text Alignment for Fair Skin Cancer Diagnosis

Jiwon Park, Seunggyu Lee, Younghoon Lee

Journal of Imaging Informatics in Medicine · 2025

DOI: 10.1007/s10278-025-01693-2 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Imaging Informatics in Medicine, 2025. Verfasst von Jiwon Park und 2 weiteren Autor:innen. Beteiligte Einrichtung: Department of Data Science, Seoul National University of Science and Technology, 232, Gongneung-ro, Nowon-gu, Seoul, 01811, Korea. Hauptthemen laut MEDLINE: Skin Neoplasms, Image Interpretation, Computer-Assisted, Deep Learning. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 34 Literaturangaben. Der Volltext ist frei in PubMed Central verfügbar. Gefördert von: Seoul national university of science and technology.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Recent advances in deep learning have significantly improved skin cancer classification, yet concerns regarding algorithmic fairness persist because of performance disparities across skin tone groups. Existing methods often attempt to mitigate bias by suppressing sensitive attributes within images. However, they are fundamentally limited by the entanglement of lesion characteristics and skin tone in visual inputs. To address this challenge, we propose a novel contrastive learning framework that leverages explicitly constructed image-text pairs to disentangle lesion condition features from skin tone attributes. Our architecture consists of a shared text encoder and two specialized image encoders that independently align image features with the corresponding textual descriptions of lesion characteristics and skin tone. Furthermore, we measure the semantic distance between lesion conditions and skin color embeddings in both image- and text-embedding spaces and perform optimal representation alignment by matching the distances in the image space to those in the text space. We validated our method using two benchmark datasets, PAD-UFES-20 and Fitzpatrick17k, which span a wide range of skin tones. The experimental results demonstrate that our approach consistently improves both classification accuracy and fairness across multiple evaluation metrics.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Skin NeoplasmsImage Interpretation, Computer-AssistedDeep LearningHumansAlgorithmsSkin Pigmentation

Stichwörter der Autor:innen

Machine learning fairnessMultimodal representation learningSkin cancer diagnosis

Autor:innen und Institutionen

  • Department of Data Science, Seoul National University of Science and Technology, 232, Gongneung-ro, Nowon-gu, Seoul, 01811, Korea.
  • Gen AI R&D Team, INTERX, Ludwig-Erhard-Str.30-34, 65760, Eschborn, Germany.
  • Department of Industrial Engineering, Seoul National University of Science and Technology, 232, Gongneung-ro, Nowon-gu, Seoul, 01811, Korea. yhoon.lee@seoultech.ac.kr.

Förderung

  • Seoul national university of science and technology

Zitieren

APA

Park, J., Lee, S., & Lee, Y. (2025). FairDITA: Disentangled Image-Text Alignment for Fair Skin Cancer Diagnosis. Journal of Imaging Informatics in Medicine. https://doi.org/10.1007/s10278-025-01693-2

Vancouver

Park J, Lee S, Lee Y. FairDITA: Disentangled Image-Text Alignment for Fair Skin Cancer Diagnosis. Journal of Imaging Informatics in Medicine. 2025. doi:10.1007/s10278-025-01693-2

BibTeX
@article{Park2025,
  author = {Park, Jiwon and Lee, Seunggyu and Lee, Younghoon},
  title = {FairDITA: Disentangled Image-Text Alignment for Fair Skin Cancer Diagnosis},
  journal = {Journal of Imaging Informatics in Medicine},
  year = {2025},
  doi = {10.1007/s10278-025-01693-2}
}
Kostenlos starten →