Deep Attention Model for the Hierarchical Diagnosis of Skin Lesions

Catarina Barata, Jorge S. Marques, M. Emre Celebi; The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 0-0

Abstract


Deep learning has played a major role in the recent advances in the dermoscopy image analysis field. However, such advances came at the cost of reducing the interpretability of the developed diagnostic systems, which do not comply with the requirements of the medical community nor with the most recent laws on machine learning explainability. Recent advances in the deep learning field, namely attention maps, improved the interpretability of these methods. Incorporating medical knowledge in the systems has also proved useful to increase their performance. In this work we propose to combine these two approaches in a formulation that: i) makes use of the hierarchical organization of skin lesions, as identified by dermatologists, to develop a classification model; and ii) uses an attention module to identify relevant regions in the skin lesions and guide the classification decisions. We demonstrate the potential of the proposed approach in two state-of-the-art dermoscopy sets (ISIC 2017 and ISIC 2018).

Related Material


[pdf]
[bibtex]
@InProceedings{Barata_2019_CVPR_Workshops,
author = {Barata, Catarina and Marques, Jorge S. and Emre Celebi, M.},
title = {Deep Attention Model for the Hierarchical Diagnosis of Skin Lesions},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}
}