Prostate Metastasis Segmentation from CT Images using Deep Learning

This repository implements a deep learning pipeline for segmenting prostate metastasis from CT images. The process leverages Python, MONAI, Torch, and CUDA to build an efficient and accurate segmentation system.

Tools and Libraries

• Python: Primary programming language.
• MONAI: Medical Open Network for AI library for deep learning in medical imaging.
• Torch: PyTorch deep learning framework.
• CUDA: GPU acceleration for faster processing.

---

Steps

1. DATA Preparation

Prepare the dataset by organizing CT images and corresponding segmentation masks. Ensure the following:

• Data is stored in a structured format (e.g., NIFTI, DICOM).
• Dataset is split into training, validation, and testing subsets.

2. Preprocessing

Perform preprocessing steps to normalize and augment the data:

• Resample images to a consistent voxel spacing.
• Normalize intensity values.
• Apply data augmentation techniques like flipping, rotation, and cropping to increase dataset diversity.

3. Training Pipeline

Train the deep learning model using MONAI and Torch:

• Define the segmentation model architecture.
• Set hyperparameters (learning rate, batch size, number of epochs).
• Utilize GPU acceleration with CUDA for training.
• Monitor training metrics (e.g., loss, Dice coefficient).

4. Evaluation using Dice Coefficient

Evaluate the model performance:

• Use the Dice Similarity Coefficient to assess segmentation accuracy.
• Compare predictions against ground truth masks.
• Adjust the model as needed based on evaluation results.

5. Saving Segmentation to RT-STRUCT using Postprocessing Pipeline

Generate RT-STRUCT files from the segmentation results:

• Post-process the segmented images (e.g., smoothing, contour extraction).
• Convert segmentation masks to RT-STRUCT format for clinical use.
• Save the output in a standardized directory structure.

---

How to Use

1. Clone the repository:

   git clone https://github.com/mohandemadx/Prostate-Metastasis-Segmentation-from-CT-images-using-Deep-Learning.git

2. Install dependencies:

   pip install -r requirements.txt

3. Follow the steps outlined above to prepare data, train the model, and evaluate results.

---

Acknowledgments

This project leverages MONAI and PyTorch, with significant contributions from the medical imaging and deep learning communities.