skrantidatta / attention-based-skin-cancer-classification

Attention-based-Skin-Cancer-Classification

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

Results

Soft Attention maps of Skin lesion in Inception ResNet V2 on HAM10000 data

Citation

@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} }

Datasets

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

attention-based-skin-cancer-classification's Issues

Requirements/ Environment

Can you provide a requirements.txt or better yet an environment.yml (or otherwise installation instructions)?

how much each model takes to train

Did you use any GPU to train these models. These are taking very long to train on Simple CPU

class weights

Hi @skrantidatta ,

Thanks for sharing the paper and code. Would you please explain the reason to choose class weight = 5.0 for melanoma while other classes are 1.0. Thanks

What kind of attention is this？

Hi, Your code is great! I'm interested in your code. Can you tell me what kind of attention mechanism this is？Is this multiheads attention mechanism？

Kernel crashing !

I have done the preprocessing and all other necessary steps mentioned in the repository but when I am trying to run the cell for training the gradcam model the kernel is crashing. I don't know the cause for the crash, did you face a similar issue while training ?

Reading zip file into Colab doesn't work for me

Hello dear Skrantidatta,
Thank you so much for sharing your knowledge!
I have issue in running this code cell and I get error about not working:
What can I do?
" from zipfile import ZipFile
filename="/content/drive/MyDrive/ISIC-2017_Training_Data.zip"
with ZipFile(filename,'r') as zip:
zip.extractall()
print("done")"
Error is about not finding the zip file, however I already have it in the direction..
Thank you so much for answering!