DermatoAI-ISIC2018

Dataset

ISIC 2018 Task 3 Dataset

Citation HAM10000 Dataset: (c) by ViDIR Group, Department of Dermatology, Medical University of Vienna; https://doi.org/10.1038/sdata.2018.161

MSK Dataset: (c) Anonymous; https://arxiv.org/abs/1710.05006; https://arxiv.org/abs/1902.03368

Dataset source: ISIC Challenge

Usage Guide

1. Data: Download the data from https://challenge.isic-archive.com/data/#2018.

2. Install Python: Make sure Python is installed on your system. If not, you can download and install Python from the official Python website: https://www.python.org/downloads/

3. Create a virtual environment:

   python -m venv env

4. Activate the virtual environment

   For Windows

   env\Scripts\activate

   For macOS/Linux

   source env/bin/activate

5. Install the dependencies

   pip install -r requirements.txt

Configuration

• Prepare a JSON configuration file with the following structure (an example is provided as config/config-example.json):

  {
      "random_seed": 42,
      "data_paths": {
          "train": "./data/ISIC2018_Task3_Training_Input",
          "validation": "./data/ISIC2018_Task3_Validation_Input",
          "test": "./data/ISIC2018_Task3_Test_Input"
      },
      "label_paths": {
          "train": "./data/ISIC2018_Task3_Training_GroundTruth/ISIC2018_Task3_Training_GroundTruth.csv",
          "validation": "./data/ISIC2018_Task3_Validation_GroundTruth/ISIC2018_Task3_Validation_GroundTruth.csv",
          "test": "./data/ISIC2018_Task3_Test_GroundTruth/ISIC2018_Task3_Test_GroundTruth.csv"
      },
      "num_classes": 7,
      "class_names": ["MEL", "NV", "BCC", "AKIEC", "BKL", "DF", "VASC"],
      "wandb": {
          "api_key": "YOUR_WANDB_API_KEY_HERE"
      }
  }

Training

python src\train.py
        --architecture <vit or resnet50>
        --batch_size <batch_size> 
        --epochs <num_epochs> 
        --lr <learning_rate> 
        --dropout_rate <dropout_rate> 
        --weight_decay <weight_decay> 
        --step_size <step_size> 
        --gamma <gamma> 
        --patience <patience> 
        --subset <subset_size> 
        --config_filepath <path_to_config_file> 
        --run_name <experiment_name> 
        --checkpoint_filename <path_to_save_best_model> 
        [--track_experiment]

Arguments

• --architecture (default: 'resnet50'): Architecture to use. Choose between 'vit' for Vision Transformer and 'resnet50' for ResNet-50.
• --batch_size (default: 8): Input batch size for training.
• --epochs (default: 100): Number of epochs to train.
• --lr (default: 0.0001): Learning rate.
• --dropout_rate (default: 0.1): Dropout rate.
• --weight_decay (default: 0.01): Weight decay for regularization.
• --step_size (default: 20): Step size for the learning rate scheduler.
• --gamma (default: 0.1): Multiplicative factor for learning rate decay.
• --patience (default: 10): Patience for early stopping.
• --subset (default: None): Use a subset of the full dataset for training, specify the subset size.
• --config_filepath (default: './config/config.json'): Path to the configuration file.
• --run_name (default: 'resnet-batch8-lr0_0001-dropout0_1'): Name of the experiment run.
• --checkpoint_filename (default: './weights/best_model.pth'): File path to save the best model weights.
• --track_experiment (optional): Flag to track the experiment using Weights & Biases.

Evaluation

python src\evaluate.py 
        --architecture <vit or resnet50>
        --batch_size <batch_size> 
        --config_filepath <path_to_config_file> 
        --checkpoint_filename <path_to_saved_model> 
        --cm_filename <filename_for_confusion_matrix> 
        --run_name <experiment_name> 
        [--track_experiment]

Arguments

• --architecture (default: 'resnet50'): Architecture to use. Choose between 'vit' for Vision Transformer and 'resnet50' for ResNet-50.
• --batch_size (default: 8): Input batch size for evaluation.
• --config_filepath (default: './config/config.json'): Path to the configuration file.
• --checkpoint_filename (default: './weights/best_model.pth'): Path to the saved model weights file.
• --cm_filename (default: 'confusion_matrix'): Filename (without extension) for saving the generated confusion matrix image.
• --run_name (default: 'resnet-batch8-lr0_0001-dropout0_1'): Name of the experiment run to be used for identifying the Weights & Biases run ID.
• --track_experiment (optional): Flag to track the evaluation using Weights & Biases.

Inference

python src/inference.py
        --architecture <vit or resnet50>
        --img_path <path_to_image>
        --config_filepath <path_to_config_file>
        --checkpoint_filename <path_to_saved_model>

Arguments

• --architecture (default: 'resnet50'): Specify the architecture to use. Choose between 'vit' for Vision Transformer and 'resnet50' for ResNet-50.
• --img_path (default: './data/ISIC2018_Task3_Test_Input/ISIC_0034524.jpg'): Path to the image file to be classified.
• --config_filepath (default: './config/config.json'): Path to the configuration file.
• --checkpoint_filename (default: './weights/best_model.pth'): Path to the saved model weights file to be used for inference.