Enhanced Dermatology Image Quality Assessment via Cross-Domain Training
A cross-domain image-quality study combining dermatology and general-purpose datasets to improve robustness with limited clinical data.
Enhanced Dermatology Image Quality Assessment via Cross-Domain Training
Authors
Ignacio Hernández Montilla, Alfonso Medela, Paola Pasquali, Andy Aguilar, Taig Mac Carthy, Gerardo Fernández, Antonio Martorell, Enrique Onieva.
Abstract
Reliable image-quality assessment is important when dermatology images are captured across different devices and settings. This work combines dermatology and non-dermatology image-quality datasets and introduces Legit.Health-DIQA-Artificial, assembled from several dermatology sources and annotated by human observers.
The experiments found that cross-domain training produced the best performance across domains and helped address the small scale and varied distortions of available dermatology data.
Publication
Published as a peer-reviewed conference paper in the proceedings of ICBRA 2025, pages 1-9.
Available at https://doi.org/10.1145/3774976.3774977.
Related Research
AEDV 2022: Deep-learning algorithm for optimising triage and referral
Deep-learning algorithm for optimising the triage and referral of patients with skin conditions. Presentation delivered in Spanish.
Read publicationAEDV 2022: Automatic urticaria scoring with artificial intelligence
Automatic urticaria scoring using artificial intelligence for precise hive counting. Presentation delivered in Spanish.
Read publicationAEDV 2023: Validation of a deep-learning algorithm for melanoma diagnosis
Results from validating a deep-learning algorithm for melanoma diagnosis. Presentation delivered in Spanish.
Read publicationInterested in medical AI?
Learn more about Legit.Health and our work in dermatology AI.
Explore Legit.Health