derm7pt

derm7pt preprocess the Seven-Point Checklist Dermatology Dataset and converts the data into a more acessible format.

derm7pt is a Python module that serves as a starting point to use the data as described in,

J. Kawahara, S. Daneshvar, G. Argenziano, and G. Hamarneh, “Seven-Point Checklist and Skin Lesion Classification using Multitask Multimodal Neural Nets,” IEEE Journal of Biomedical and Health Informatics, vol. 23, no. 2, pp. 538–546, 2019. [pdf] [doi]

Download the Data

http://derm.cs.sfu.ca

The images and meta-data (e.g., seven-point checklist criteria, diagnosis) can be downloaded from the external site above.

The actual images and meta-data are not stored in this repo.

Minimal Example

import sys, os
import pandas as pd
sys.path.insert(0, os.path.abspath(os.path.join(os.getcwd(), '..'))) # To import derm7pt
from derm7pt.dataset import Derm7PtDatasetGroupInfrequent

# Change this line to your data directory.
dir_release = '/local-scratch/jer/data/argenziano/release_v0'

# Dataset after grouping infrequent labels.
derm_data = Derm7PtDatasetGroupInfrequent(
    dir_images=os.path.join(dir_release, 'images'), 
    metadata_df=pd.read_csv(os.path.join(dir_release, 'meta/meta.csv')), 
    train_indexes=list(pd.read_csv(os.path.join(dir_release, 'meta/train_indexes.csv'))['indexes']), 
    valid_indexes=list(pd.read_csv(os.path.join(dir_release, 'meta/valid_indexes.csv'))['indexes']), 
    test_indexes=list(pd.read_csv(os.path.join(dir_release, 'meta/test_indexes.csv'))['indexes']))

# Outputs to screen the preprocessed dataset in a Pandas format.
derm_data.df

This will group infrequent class labels together and assign numeric values to each class label.

You can see the output in this minimal example notebook.

You can find a more comprehensive example here that includes an example of how to classify some of the seven-point checklist.

Installation Instructions

You can see the dependencies and versions derm7pt was tested on here.

To use derm7pt:

Download the data and unzip it to your folder (we will use the folder /local-scratch/jer/data/argenziano/release_v0 for this example)
Clone this repository
Run the minimal_example.py. Make sure to change the directory to match your data folder.

Steps #2 and #3 are shown below,

git clone https://github.com/jeremykawahara/derm7pt.git
cd derm7pt
python minimal_example.py '/local-scratch/jer/data/argenziano/release_v0'

This should output a view of the data that is similar to what is shown in this notebook.

Related Publications

More information about this data can be found in our publication, and if you use the data or code, please cite our work,

@article{Kawahara2018-7pt,
author = {Kawahara, Jeremy and Daneshvar, Sara and Argenziano, Giuseppe and Hamarneh, Ghassan},
doi = {10.1109/JBHI.2018.2824327},
issn = {2168-2194},
journal = {IEEE Journal of Biomedical and Health Informatics},
month = {mar},
number = {2},
pages = {538--546},
publisher = {IEEE},
title = {Seven-point checklist and skin lesion classification using multitask multimodal neural nets},
volume = {23},
year = {2019}
}

You can read more about the seven-point checklist here:

G. Argenziano, G. Fabbrocini, P. Carli, D. G. Vincenzo, E. Sammarco, and M. Delfino, “Epiluminescence microscopy for the diagnosis of doubtful melanocytic skin lesions. Comparison of the ABCD rule of dermatoscopy and a new 7-point checklist based on pattern analysis,” Arch. Dermatol., vol. 134, no. 12, pp. 1563–1570, 1998.

Clarifying Notes

The following notes are all related to this publication:

J. Kawahara, S. Daneshvar, G. Argenziano, and G. Hamarneh, “Seven-Point Checklist and Skin Lesion Classification using Multitask Multimodal Neural Nets,” IEEE Journal of Biomedical and Health Informatics, vol. 23, no. 2, pp. 538–546, 2019. [pdf] [doi]

In Section B. Mini-Batches Sampled and Weighed by Label we set k=1.

This means the mini-batch has 24k = 24 samples, since there are 24 unique labels.