In clinical applications, neural networks must focus on and highlight the most important parts of an input image. Soft-Attention mechanism enables a neural network to achieve this goal. This paper investigates the effectiveness of Soft-Attention in deep neural architectures. The central aim of Soft-Attention is to boost the value of important features and suppress the noise-inducing features. We compare the performance of VGG, ResNet, Inception ResNetv2 and DenseNet architectures with and without the Soft-Attention mechanism, while classifying skin lesions. The original network when coupled with Soft-Attention outperforms the baseline by 4.7% while achieving a precision of 93.7% on HAM10000 dataset. Additionally, Soft-Attention coupling improves the sensitivity score by 3.8% compared to baseline and achieves 91.6% on ISIC-2017 dataset.
All the experiments were executed on the Keras framework with tensorflow version 2.4.0.
https://arxiv.org/abs/2105.03358
@article{datta2021soft, title={Soft-Attention Improves Skin Cancer Classification Performance}, author={Datta, Soumyya Kanti and Shaikh, Mohammad Abuzar and Srihari, Sargur N and Gao, Mingchen}, journal={arXiv preprint arXiv:2105.03358}, year={2021} }
HAM10000 dataset:
Tschandl, P., Rosendahl, C. & Kittler, H. The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions. Sci. Data 5, 180161 doi:10.1038/sdata.2018.161 (2018).Available: https://www.nature.com/articles/sdata2018161, https://arxiv.org/abs/1803.10417
ISIC-2017 dataset:
Codella N, Gutman D, Celebi ME, Helba B, Marchetti MA, Dusza S, Kalloo A, Liopyris K, Mishra N, Kittler H, Halpern A. "Skin Lesion Analysis Toward Melanoma Detection: A Challenge at the 2017 International Symposium on Biomedical Imaging (ISBI), Hosted by the International Skin Imaging Collaboration (ISIC)". arXiv: 1710.05006 [cs.CV] Available: https://arxiv.org/abs/1710.05006