Skip to content

ISIC 2019 - Skin Lesion Analysis Towards Melanoma Detection

License

Notifications You must be signed in to change notification settings

wanghsinwei/isic-2019

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ISIC 2019 - Skin Lesion Analysis Towards Melanoma Detection

This is a Keras with TensorFlow backend implementation for ISIC 2019 Challenge Task 1: classify dermoscopic images among nine different diagnostic categories without meta-data. The proposed approaches implemented in this repository were ranked 25th out of 64 unique teams with Balanced Multiclass Accuracy of 0.505 (0.636 for the first team). The ODIN (Out-of-DIstribution detector for Neural networks) method used in the first approach was re-implemented using Keras. This is also my capstone project of Udacity Machine Learning Nanodegree. For more details please refer to the project report.

Getting Started

Dependencies

Datasets

Diagnostic Category Amount
Angiofibroma or fibrous papule 1
Angioma 12
Atypical melanocytic proliferation 12
Lentigo NOS 70
Lentigo simplex 22
Scar 1

After downloading the dataset, all melanosis images can be retrieved from the dataset by using the notebook.

Diagnostic Category Amount
Melanosis 16

Directory & File Structure of Input Data

  • Put all data under a root folder like the tree structure shown below.
    • ISIC_2019_Training_Input and ISIC_2019_Test_Input folders contain images of ISIC 2019 Training Data and Test Data respectively.
    • Out_Distribution folder contains all out-of-distribution images from both ISIC Archive and Seven-Point datasets.
    • ISIC_2019_Training_GroundTruth_DuplicateRemoved.csv is a copy of ISIC_2019_Training_GroundTruth.csv but removing two entries: ISIC_0067980 and ISIC_0069013.
RootFolder
│   ISIC_2019_Training_GroundTruth.csv
│   ISIC_2019_Training_GroundTruth_DuplicateRemoved.csv
│
└───ISIC_2019_Test_Input
│   │   ISIC_0034321.jpg
│   │   ISIC_0034322.jpg
│   │   ...
│   │
└───ISIC_2019_Training_Input
│   │   ISIC_0000000.jpg
│   │   ISIC_0000001.jpg
│   │   ...
│   │
└───Out_Distribution
│   │   7pt_Fgl059.jpg
│   │   7pt_Fhl002.jpg
│   │   ...
│   │   ISIC_0001115.jpg
│   │   ISIC_0001129.jpg
│   │   ...

Training and Predicting

Follow the steps starting from the Common Parameters cell in the notebook approach_1.ipynb and approach_2.ipynb to reproduce training and predicting processes of approach-1 and 2 respectively. You might need to change a few parameters according to the environment.

Command-line

Here is an example code of training DenseNet201, Xception and ResNeXt50 models using approach-1. Assuming the path to the data root folder is /home.

python3 main.py /home --approach 1 --training --epoch 100 --batchsize 32 --maxqueuesize 10 --model DenseNet201 Xception ResNeXt50

Use -h to show usage messages.

python3 main.py -h

Testing Results

Please visit ISIC 2019 Challenge Leaderboards for complete rankings.

Team Rankings

Ranked 25th out of 64 unique teams with Balanced Multiclass Accuracy of 0.505 (0.636 for the first team) Team Rankings

Approach Rankings

Rank Approach Name (129 approaches) Balanced Multiclass Accuracy
52 Convolutional Ensemble with Out-of-Distribution Detector 0.505
57 Skin Lesion Classification using Ensemble of Convolutional Neural Networks 0.499

License

MIT