Skin Lesions Classification using Multichannel Dermoscopic Images
| Main Authors: | Moura, Luís Vinícius, Dartora, Caroline Machado, Silva, Ana Maria Marques |
|---|---|
| Format: | Proceeding poster |
| Bahasa: | eng |
| Terbitan: |
, 2019
|
| Subjects: | |
| Online Access: |
https://zenodo.org/record/3464578 |
Daftar Isi:
- Visual inspection of visible light dermoscopic images can be challenging due to the similarity between melanoma and benign skin lesions, especially in the early stages. This paper aims to investigate the influence of independent use of color channels in feature extraction for skin lesions classification in dermoscopic images. Visible dermoscopic RGB images from 1850 benign and malignant (melanoma) skin lesions from ISIC database were used. RGB channels were split and analyzed independently to identify the relevance of each color channel information. Radiomics features were extracted, normalized, and selected by their scores for an unsupervised feature-ranking algorithm. Selected features were used to train a random forest classifier, which was cross-validated with K-fold method and optimized using exhaustive grid search. The results showed no significant differences in skin lesions classification using each color channels, although R channel produced the most relevant information, with an accuracy of 0.86 and an AUC of 0.93.