🏥 Python CT Simulation on Low Dose/ Limited Angle/ Sparse View Reconstruction using CUDA Astra Toolbox

🚀 A comprehensive Python toolkit for CT (Computed Tomography) image processing, sinogram generation, and reconstruction with various noise and artifact simulations.

📋 Overview

This project provides two main scripts for CT image processing:

1. 📊 sino_creation.py - Interactive CT sinogram creation and reconstruction from DICOM IMA files

2. 🔧 noise_reconstruction.py - Advanced CT reconstruction with noise and artifact simulations

The toolkit supports various CT reconstruction scenarios including normal FBP reconstruction, low-dose CT (LDCT), limited angle reconstruction, and sparse view reconstruction.

✨ Features

📊 Sinogram Creation (sino_creation.py)

• 📁 Load and process DICOM IMA files
• 🎯 Generate high-quality sinograms using ASTRA toolbox
• ✅ Optional FBP reconstruction for verification
• 📈 Visualization comparing original, sinogram, and reconstructed images
• 🔄 Batch processing support with glob patterns
• ⚙️ Configurable preprocessing (Gaussian filtering)

🔧 Noise and Artifact Reconstruction (noise_reconstruction.py)

• 💉 Low-Dose CT (LDCT): Simulates reduced radiation dose with Poisson noise
• 📐 Limited Angle: Reconstruction from constrained angular range (default: 120°)
• 🎯 Sparse View: Reconstruction with reduced projection angles
• ⭐ Normal FBP: High-quality baseline reconstruction
• 📊 Comprehensive visualization comparing all reconstruction types

🛠️ Installation

💻 System Requirements

• 🐍 Python 3.7+
• 🎮 CUDA-compatible GPU (optional, for faster reconstruction)
• 💾 Sufficient RAM for processing large CT datasets

📦 Prerequisites

# Install required Python packages
pip install numpy matplotlib scikit-image tqdm pydicom

# Install ASTRA Toolbox (for CT reconstruction)
# Option 1: Using conda (recommended) 🌟
conda install -c astra-toolbox astra-toolbox

# Option 2: Using pip
pip install astra-toolbox

📁 Project Structure

CT_Simulation/
├── 📊 sino_creation.py          # Sinogram generation script
├── 🔧 noise_reconstruction.py   # Noise/artifact reconstruction script
├── 📂 sample/                   # Sample data directory
│   ├── 🏥 sample.IMA           # Sample DICOM file
│   └── 🖼️ *.png                # Sample visualizations
├── 📤 output/                   # Generated outputs
│   ├── 📊 sinograms/           # Generated sinograms (.npy)
│   ├── ⭐ normal/              # Normal reconstructions
│   ├── 💉 ldct/                # Low-dose reconstructions
│   ├── 📐 limited_angle/       # Limited angle reconstructions
│   ├── 🎯 sparse_view/         # Sparse view reconstructions
│   └── 📈 visualizations/      # Comparison plots
├── 📦 archive/                  # Archived/legacy scripts
└── 📝 *.log                    # Processing logs

🚀 Usage

1️⃣ Sinogram Creation (sino_creation.py)

Generate sinograms from DICOM IMA files with optional reconstruction and visualization.

🔰 Basic Usage

# Process single IMA file
python sino_creation.py sample/sample.IMA --output_dir ./output

# Process multiple files with pattern matching 🔄
python sino_creation.py "sample/*.IMA" --output_dir ./output

# Process with reconstruction verification ✅
python sino_creation.py sample/sample.IMA --output_dir ./output --reconstruct

# Process with visualization (includes reconstruction) 📈
python sino_creation.py sample/sample.IMA --output_dir ./output --visualize

🔧 Advanced Options

# Full pipeline with all options 🚀
python sino_creation.py "data/*.IMA" \
    --output_dir ./results \
    --reconstruct \
    --visualize \
    --verbose \
    --log_file processing.log

⚙️ Parameters

• 📁 input: Input IMA file path or glob pattern
• 📤 --output_dir, -o: Output directory (default: ./output)
• ✅ --reconstruct, -r: Perform FBP reconstruction for verification
• 📈 --visualize, -v: Create comparison visualization
• 🔊 --verbose: Enable detailed logging (default: True)
• 📝 --log_file: Custom log file path (default: sino_creation.log)

📤 Output Files

• 📊 Sinograms: output/sinograms/{filename}.npy
• 🔄 Reconstructions: output/reconstructions/{filename}.npy (if --reconstruct)
• 📈 Visualizations: output/visualizations/{filename}_comparison.png (if --visualize)

2️⃣ Noise and Artifact Reconstruction (noise_reconstruction.py)

Perform various CT reconstruction methods from existing sinogram files.

🔰 Basic Usage

# Normal FBP reconstruction only ⭐
python noise_reconstruction.py output/sinograms/sample.npy --output_dir ./output

# Enable specific reconstruction types 🎯
python noise_reconstruction.py output/sinograms/sample.npy \
    --output_dir ./output \
    --ldct \
    --limited_angle \
    --sparse_view

# Process multiple sinograms with visualization 📊
python noise_reconstruction.py "output/sinograms/*.npy" \
    --output_dir ./output \
    --ldct \
    --limited_angle \
    --sparse_view \
    --visualize

⚙️ Parameters

• 📁 input: Input sinogram file (.npy) or glob pattern
• 📤 --output_dir, -o: Output directory (default: ./output)
• 💉 --ldct: Enable Low-Dose CT reconstruction
• 📐 --limited_angle: Enable Limited Angle reconstruction
• 🎯 --sparse_view: Enable Sparse View reconstruction
• 📈 --visualize, -v: Create comprehensive comparison visualization
• 🔊 --verbose: Enable detailed logging (default: True)
• 📝 --log_file: Custom log file path (default: noise_reconstruction.log)

🔬 Reconstruction Types

1. ⭐ Normal FBP: High-quality baseline reconstruction (always performed)
2. 💉 Low-Dose CT (LDCT):
   • 📉 Dose reduction factor: 50% (configurable)
   • 🎲 Poisson noise simulation for photon counting statistics
3. 📐 Limited Angle:
   • 📏 Angular range: 120° (configurable)
   • 🎯 Starting angle: 0° (configurable)
4. 🎯 Sparse View:
   • 📊 Sampling factor: Every 24th angle (configurable)

📤 Output Files

• ⭐ Normal: output/normal/{filename}.npy
• 💉 LDCT: output/ldct/{filename}.npy (if --ldct)
• 📐 Limited Angle: output/limited_angle/{filename}.npy (if --limited_angle)
• 🎯 Sparse View: output/sparse_view/{filename}.npy (if --sparse_view)
• 📈 Visualization: output/visualizations/{filename}_comprehensive.png (if --visualize)

🔄 Complete Workflow Example

Here's a complete workflow from DICOM file to various reconstructions:

# Step 1: Generate sinogram from DICOM file 📊
python sino_creation.py sample/sample.IMA \
    --output_dir ./output \
    --visualize

# Step 2: Perform various reconstructions with noise/artifacts 🔧
python noise_reconstruction.py output/sinograms/sample.npy \
    --output_dir ./output \
    --ldct \
    --limited_angle \
    --sparse_view \
    --visualize

# Results will be available in: 📁
# - output/sinograms/sample.npy (original sinogram)
# - output/normal/sample.npy (normal reconstruction)
# - output/ldct/sample.npy (low-dose reconstruction)
# - output/limited_angle/sample.npy (limited angle reconstruction)
# - output/sparse_view/sample.npy (sparse view reconstruction)
# - output/visualizations/ (comparison plots)

⚙️ Configuration

🔧 Reconstruction Parameters

Key parameters can be modified in the script headers:

📊 Sinogram Creation (sino_creation.py):

NUM_ANGLES = 2160              # Number of projection angles
ANGLE_RANGE_DEG = (0.0, 360.0) # Angular range
DETECTOR_SPACING = 1.0         # Detector pixel spacing
APPLY_GAUSSIAN_FILTER = False  # Preprocessing filter
GAUSSIAN_SIGMA = 0.4           # Filter parameter

🔧 Noise Reconstruction (noise_reconstruction.py):

# Low-dose parameters 💉
DOSE_REDUCTION_FACTOR = 0.5    # 50% dose reduction
POISSON_NOISE_SCALE = 1.0      # Noise scaling

# Limited angle parameters 📐
LIMITED_ANGLE_RANGE_DEG = 120.0  # Angular range
LIMITED_ANGLE_START_DEG = 0.0    # Starting angle

# Sparse view parameters 🎯
SPARSE_VIEW_FACTOR = 24        # Sampling factor

⚡ Performance Notes

• 🎮 GPU Acceleration: CUDA-enabled GPUs significantly speed up reconstruction
• 💾 Memory Usage: Large datasets may require substantial RAM
• ⏱️ Processing Time: Varies with image size and number of angles
  • 📊 Sinogram generation: ~1-5 seconds per image
  • 🔄 FBP reconstruction: ~0.1-1 seconds per reconstruction

📝 Logging

Both scripts generate detailed logs:

• 📊 sino_creation.log: Sinogram generation process
• 🔧 noise_reconstruction.log: Reconstruction process

Use --verbose flag for detailed console output. 🔊

🔬 Other Algorithms

I compared lots of algorithms, here are a comparison plot:

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🤝 Contributing

Feel free to reach me (Jie) out if you have any questions or suggestions! This is NOT a very professional project, but I'm happy to improve it. 😊

---

💡 Tip: Start with the sample data in the sample/ directory to get familiar with the toolkit before processing your own DICOM files!