Attention Residual Learning for Skin Lesion Classification(IEEE Transactions on Medical Imaging )

Introduction

This work is an unofficial code implemented by Pytorch. Some issues are different with original paper:

• We don't employed additional dermoscopy images which is collected in ISIC Archive.
• The data augment isn't same with original paper but the performance of our work is approximative.

Preparation

First of all, clone the code

https://github.com/Vipermdl/ARL

Then, create a folder for dataset:

cd ARL && mkdir data

Note: We currently only support ISIC 2017. To make things easy, we provide simple dataset loader that inherits torch.utils.data.Dataset making it fully compatible with the torchvision.datasets API.

A soft-link is recommended.

ln -s /path/to/isic2017 ./data/isic2017

Then, we sliced the images for patches to train:

python generate_patch_images.py

It should have this basic structure

$ISIC2017/
$ISIC2017/ISIC-2017_Test_v2_Data/
$ISIC2017/ISIC-2017_Training_Data/
$ISIC2017/ISIC-2017_Validation_Data/
$ISIC2017/ISIC-2017_Training_Data_Patch/
$ISIC2017/ISIC-2017_Test_v2_Part3_GroundTruth.csv/
$ISIC2017/ISIC-2017_Training_Part3_GroundTruth.csv/
$ISIC2017/ISIC-2017_Validation_Part3_GroundTruth.csv/
$ISIC2017/ISIC-2017_Training_Part3_GroundTruth_patch.csv/

prerequisites

• Python 3.6
• Pytorch 0.4.0 or higher
• CUDA 8.0 or higher
• Install all the python dependencies using pip: pip install -r requirements.txt

Train

python train_mel.py

Test

python predict2017_mel.py

Performance Comparision

We re-implemented with ARLNet50 for the task of Melanoma Classification:

Work	params	AUC	ACC	Sensitivity	Specificity
Original paper	2.3	0.875	0.850	0.658	0.96
Our work	2.35	0.872	0.850	0.487	0.937

Below is AUC figure:

Authorship

This repository is produced by Dongliang Ma, if you have any question, please contact with me.