weinajin / IPMI2019-AttnMel

IPMI2019 paper - Melanoma Recognition via Visual Attention

Melanoma Recognition via Visual Attention

---

Code for our IPMI paper:

Yiqi Yan, Jeremy Kawahara, and Ghassan Hamarneh, “Melanoma Recognition via Visual Attention”. To appear in International Conference on Information Processing in Medical Imaging (IPMI), 2019.

---

Pre-traind models

Google drive link

How to run

1. Dependences

• PyTorch (>= 0.4, 1.0 preferred)
• torchvision
• tensorboardX
• Pillow
• scikit-learn

2. Data preparation

ISIC 2016 download here; organize the data as follows

• data_2016/
  • Train/
    • benign/
    • malignant/
  • Test/
    • benign/
    • malignant/

ISIC 2017 download here; organize the data as follows

• data_2016/
  • Train/
    • melanoma/
    • nevus/
    • seborrheic_keratosis/
  • Val/
    • melanoma/
    • nevus/
    • seborrheic_keratosis/
  • Test/
    • melanoma/
    • nevus/
    • seborrheic_keratosis/
  • Train_Lesion/
    • melanoma/
    • nevus/
    • seborrheic_keratosis/
  • Train_Dermo/
    • melanoma/
    • nevus/
    • seborrheic_keratosis/

Under the folder Train_Lesion is the lesion segmentation map (ISIC2017 part I); under the folder Train_Dermo is the map of dermoscopic features (ISIC2017 part II). The raw data of dermoscopic features require some preprocessing in order to convert to binary maps. What is under Train_Dermo is the union map of four dermoscopic features. Note that not all of the images have dermoscopic features (i.e, some of the maps are all zero).

3. Training

Switching between ISIC2016 and ISIC2017 data: modify the code following the comments at the beginning of the training script

switch between ISIC 2016 and 2017 modify the following contents:

1. import from data_2016/import from data_2017
2. num_aug: x2 for ISIC 2017; x5 for ISIC 2016
3. root_dir of preprocess_data
4. mean and std of transforms.Normalize

1. Training without any attention map regularization (with only the classification loss, i.e, AttnMel-CNN in the paper):

• train on ISIC 2016

python train.py --dataset ISIC2016 --preprocess --over_sample --focal_loss --log_images

• train on ISIC 2017 (by default)

python train.py --dataset ISIC2017 --preprocess --over_sample --focal_loss --log_images

2. Training with attention map regularization (AttnMel-CNN-Lesion or AttnMel-CNN-Dermo in the paper):

We only train on ISIC2017 for these two models.

• AttnMel-CNN-Lesion (by default)

python train_seg.py --seg lesion --preprocess --over_sample --focal_loss --log_images

• AttnMel-CNN-Dermo

python train_seg.py --seg dermo --preprocess --over_sample --focal_loss --log_images

3. Testing

python test.py --dataset ISIC2016  

or

python test.py --dataset ISIC2017  

LICENSE

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.