Adaptive Curriculum Query Strategy

Introduction

This project is the code for the paper Adaptive Curriculum Query Strategy for Active Learning in Medical Image Classification published in MICCAI2024. The implementation is based on Python and Pytorch framework.

Requirements

The main package and version of the Python environment are as follows

# Name                    Version         
python                    3.8.5                    
pytorch                   1.10.1         
torchvision               0.11.2         
cudatoolkit               10.2.89       
cudnn                     7.6.5           
matplotlib                3.3.2              
numpy                     1.19.2        
opencv                    4.6.0.66         
pandas                    1.1.3               
scikit-learn              0.23.2                
tqdm                      4.50.2             

The above environment is successful when running the code of the project. Pytorch has very good compatibility. Thus, I suggest that try to use the existing pytorch environment first.

---

Usage

1) Download Project

Runninggit clone https://github.com/HelenMa9998/Easy_hard_AL
The project structure and intention are as follows :

Adversarial active learning			# Source code		
    ├── seed.py			 	                                          # Set up random seed
    ├── query_strategies		                                    # All query_strategies
    │   ├── cdal_sampling.py                                    # Integrating Bayesian methods with high density and diversity selection
    │   ├── cluster_margin_sampling.py                          # selects a diverse set of examples on which the model is least confident
    │   ├── bayesian_active_learning_disagreement_dropout.py	  # Deep bayesian query method
    │   ├── entropy_sampling.py		                              # Entropy based query method
    │   ├── kcenter_greedy.py	                                  # samples have the smallest maximum distance to the labeled sample set, as a traditional diversity-based method
    │   ├── least_confidence.py                                 # selecting samples with the lowest confidence
    │   ├── margin_sampling.py                                  # samples that are close to the decision boundary
    │   ├── entropy_sampling_dropout.py		                      # Entropy-based MC dropout query method
    │   ├── random_sampling.py		                              # Random selection
    │   ├── strategy.py                                         # Functions needed for query strategies
    ├── data.py	                                                # Prepare the dataset & initialization and update for training dataset
    ├── handlers.py                                             # Get dataloader for the dataset
    ├── main.py			                                            # An example for code utilization, including the whole process of active learning
    ├── nets.py		                                              # Training models and methods needed for query method
    ├── supervised_baseline.py	                                # An example for supervised learning traning process
    └── utils.py			                                          # Important setups including network, dataset, hyperparameters...

2) Datasets preparation

1. Download the datasets from the official address:

   BraTS 2019 Dataset: https://www.adcis.net/en/third-party/messidor/

   BACH: https://iciar2018-challenge.grand-challenge.org/Dataset/

2. Modify the data folder path for the specific datasets in data.py

3) Run Active learning process

Please confirm the configuration information in the [utils.py]

  python main.py \
      --n_round 34 \
      --n_query 100 \
      --n_init_labeled 100 \
      --dataset_name Messidor \
      --traning_method supervised_val_loss \
      --strategy_name EntropySampling \
      --seed 42

The training results will be saved to the corresponding directory(save name) in performance.csv.
You can also run supervised_baseline.py by

python supervised_baseline.py