MONAI-Based Medical Image Classification

Overview

This project demonstrates the use of MONAI (Medical Open Network for AI) for training a DenseNet121 model on medical imaging data. The pipeline includes:

• Preprocessing medical images using MONAI transforms
• Training a DenseNet model with PyTorch and MONAI
• Evaluating model performance using AUC-ROC metrics

Dataset

This project assumes the presence of a medical imaging dataset, which is preprocessed using MONAI transforms. Example datasets include:

• NIH Chest X-ray Dataset (Link)
• PhysioNet Datasets (Link)
• MIMIC-III (Medical ICU Data) (Link)

Project Steps

1️⃣ Data Preprocessing

• Use MONAI transforms (LoadImage, EnsureChannelFirst, ScaleIntensity, ToTensor) to preprocess images.
• Cache datasets for efficient loading.

2️⃣ Model Training

• Use DenseNet121 as the backbone model.
• Define loss function (CrossEntropyLoss) and optimizer (Adam).
• Train the model for multiple epochs while tracking loss.

3️⃣ Model Evaluation

• Use AUC-ROC metric to evaluate performance.
• Store the best-performing model based on AUC.
• Validate the model on unseen test data.

Installation & Usage

1. Install Required Dependencies

pip install monai torch torchvision numpy matplotlib

2. Run Training Script

python train_monai.py

3. Visualize Results

To plot the AUC-ROC curve, modify the evaluation section:

import matplotlib.pyplot as plt
from sklearn.metrics import roc_curve, roc_auc_score

fpr, tpr, _ = roc_curve(y_true.cpu(), y_pred.cpu())
auc_score = roc_auc_score(y_true.cpu(), y_pred.cpu())

plt.plot(fpr, tpr, label=f"AUC = {auc_score:.4f}", color='blue')
plt.plot([0, 1], [0, 1], linestyle="--", color="gray")
plt.xlabel("False Positive Rate")
plt.ylabel("True Positive Rate")
plt.title("ROC Curve")
plt.legend()
plt.show()

Results

• The model achieves high AUC scores, indicating strong classification performance.
• Overfitting is prevented using validation monitoring.
• The best-performing model is saved as best_model.pth.

Contributions

Feel free to fork this repository, improve the training pipeline, or experiment with different datasets. 🚀

License

This project is open-source and available under the MIT License.