Melanoma Recognition via Visual Attention

Paper (preprint)
Paper (camera ready)
IPMI 2019 poster
Code available on Github
Project webpabe

Abstract

We propose an attention-based method for accurate melanoma recognition. The attention modules, which are learned together with other network parameters, estimate attention maps that highlight image regions of interest that are relevant to lesion classification. These attention maps provide a more interpretable output as opposed to only outputting a class label. Additionally, we propose to utilize prior information by regularizing attention maps with regions of interest (ROIs) (e.g., lesion segmentation or dermoscopic features). Whenever such prior information is available, both the classification performance and the attention maps can be further refined. To our knowledge, we are the first to introduce an end-to-end trainable attention module with regularization for melanoma recognition. We provide both quantitative and qualitative results on public datasets to demonstrate the effectiveness of our method. The code is available at https://github.com/SaoYan/IPMI2019-AttnMel.

Citation

@inproceedings{yan2019melanoma,
title={Melanoma Recognition via Visual Attention},
author={Yan, Yiqi and Kawahara, Jeremy and Hamarneh, Ghassan},
booktitle={International Conference on Information Processing in Medical Imaging},
pages={793–804},
year={2019},
organization={Springer}
}