This is the implementation of the paper "Efficient and Accurate Pneumonia Detection Using a Novel Multi-Scale Transformer Approach"

Requirements

• Python 3.8.18
• PyTorch 1.8.1
• cuda 11.5
• tensorboard 2.12.1
• numpy 1.22.3
• sklearn 1.3

Datasets used

The datasets used for this project are:

• Kermany (Chest X-Ray Images (Pneumonia)) dataset
• Cohen (COVID-19 image data collection) dataset github
• Cohen (COVID-19 image data collection) dataset Kaggle
• Chest Xray Masks and Labels dataset

Prepare Dataset

Verify the dataset structure Make sure your dataset is organized as follows:

../                      
├── common/             
├── data/               
|   ├── dataloader.py
|   ├── Kermany/
│   |   ├── train/
│   |   │   ├── NORMAL
│   |   |   └── PNEUMONIA
│   |   └── test/
│   |       ├── NORMAL
│   |       └── PNEUMONIA
|   ├── Cohen/
│   │   ├── train/
│   │   │   ├── COVID19
│   │   │   ├── NORMAL
│   │   |   └── PNEUMONIA
│   │   └── test/
│   │       ├── COVID19
│   │       ├── NORMAL
│   │       └── PNEUMONIA
│   └── Chest_Xray_Masks_and_Labels/
│       ├── train/
│       │   ├── images
│       |   └── masks
│       └── test/
│           ├── images
│           └── masks
├── model/
|   ├── MSR.py
|   ├── backbone_utils.py
|   └── transformer.py
├── README.md           
├── train_Cohen.py
├── train_Kermany.py
├── train_seg.py
├── test_Cohen.py
├── test_Kermany.py
└── test_seg.py             

If your dataset structure is different, adjust the --data_dir argument accordingly.

Segmentation Task

Training

To train the model, use the following command:

python train_seg.py --bsz 8
                    --lr 1e-3
                    --niter 30
                    --data_dir data/Chest_Xray_Masks_and_Labels
                    --logpath "your_experiment_name"

Testing

To test the model, use the following command:

python test_seg.py  --bsz 8
                --data_dir data/Chest_Xray_Masks_and_Labels
                --logpath "your_experiment_name"

Classification Task (Kermany)

Training

To train the model, use the following command:

python train_kermany.py --bsz 64
                        --lr 5e-5
                        --niter 100
                        --layers {50,101}
                        --data_dir data/Kermany
                        --logpath "your_experiment_name"

Testing

To test the model, use the following command:

python test_kermany.py  --bsz 64
                        --layers {50,101}
                        --data_dir data/Cohen
                        --logpath "your_experiment_name"

Classification Task (Cohen)

Training

To train the model, use the following command:

python train_cohen.py --bsz 64
                      --lr 5e-5
                      --niter 100
                      --layers {50,101}
                      --data_dir data/Cohen
                      --logpath "your_experiment_name"

Testing

To test the model, use the following command:

python test_cohen.py  --bsz 64
                      --layers {50,101}
                      --data_dir data/Kermany
                      --logpath "your_experiment_name"

Arguments

• --bsz: Batch size for training.
• --lr: Learning rate for the optimizer.
• --niter: Number of training iterations (epochs).
• --layers: Number of layers in the ResNet backbone (e.g., 50 or 101).
• --data_dir: Directory where the dataset is located.
• --logpath: Directory to save the best model checkpoint.

Citation

If you use this repository in your work, please cite the following paper:

@article{saber2025efficient,
  title={Efficient and accurate pneumonia detection using a novel multi-scale transformer approach},
  author={Saber, Alireza and Fateh, Amirreza and Parhami, Pouria and Siahkarzadeh, Alimohammad and Fateh, Mansoor and Ferdowsi, Saideh},
  journal={Sensors},
  volume={25},
  number={23},
  pages={7233},
  year={2025},
  publisher={MDPI}
}