Related Experiment Video
Updated: Jun 28, 2025
09:32
Resolving Water, Proteins, and Lipids from In Vivo Confocal Raman Spectra of Stratum Corneum through a Chemometric Approach
Published on: September 26, 2019
7.2K
Parameter-based transfer learning for severity classification of atopic dermatitis using hyperspectral imaging
Eun Bin Kim1, Yoo Sang Baek2, Onesok Lee1,3
1Department of Software Convergence, Graduate School, Soonchunhyang University, Asan City, Chungcheongnam-do, South Korea.
Summary
Hyperspectral imaging (HSI) effectively classifies atopic dermatitis (AD) severity using transfer learning. Training models on psoriasis data achieved high accuracy, demonstrating HSI
Area of Science:
- Dermatology
- Medical Imaging
- Machine Learning
Background:
- Atopic dermatitis (AD) is a chronic inflammatory skin condition with structural changes.
- Noninvasive methods are needed to evaluate AD.
- Hyperspectral imaging (HSI) captures skin structural features using light wavelength variations.
Purpose of the Study:
- To classify the severity of atopic dermatitis (AD) using hyperspectral imaging (HSI).
- To optimize classification results by exploring different source- and target-domain datasets with parameter-based transfer learning.
Main Methods:
- Utilized psoriasis, skin cancer, eczema, and AD datasets as source domains.
- Employed hyperspectral images as the target domain for wavelength-specific AD classification.
- Evaluated 96 combinations of sources, models, and targets for severity classification performance.
Main Results:
- Achieved 83% classification performance using ResNet50 trained on an augmented psoriasis dataset (source) and NIR dataset (target).
- Obtained 81% accuracy with ResNet50 trained on an unaugmented psoriasis dataset (source) and R dataset (target).
- ResNet50 showed potential as a generalized model, and the psoriasis dataset proved effective for training.
Conclusions:
- Demonstrated the feasibility of classifying AD severity using hyperspectral images.
- Showcased the scalability of hyperspectral imaging and transfer learning for domain exploration in AD research.
