Medical Image Segmentation Toolkit

A modular Python toolkit for medical image segmentation, focusing on bone segmentation from CT scans. Implements and compares multiple classical segmentation algorithms including active contours, level sets, watershed, and edge detection methods.

Sample Results

Level-Set Segmentation (MorphACWE)	Watershed Segmentation
Chan-Vese level set isolating a vertebral body from a sagittal CT slice (smoothing=4, λ1=3, λ2=1)	Watershed contour overlay on a sagittal spinal CT scan

Project Structure

.
├── config.py                  # Centralized configuration and default paths
├── utils.py                   # Shared utilities (image loading, conversion)
├── requirements.txt           # Python dependencies
│
├── edge_detection/            # Edge detection algorithms
│   ├── canny_skimage.py       # Canny edge detection (scikit-image)
│   ├── canny_manual.py        # Canny from scratch (Gaussian, gradient, NMS, thresholding)
│   └── sobel.py               # Sobel and Laplacian edge detection (OpenCV)
│
├── active_contour/            # Active contour and level-set methods
│   ├── parametric_snake.py    # Parametric active contour (snake) via scikit-image
│   ├── morph_acwe.py          # Morphological Chan-Vese (region-based level set)
│   ├── morph_gac.py           # Morphological Geodesic Active Contour (edge-based)
│   ├── morph_acwe_3d.py       # 3D Chan-Vese with marching cubes visualization
│   └── level_set_evolution.py # Manual level-set PDE solver with curvature/balloon forces
│
├── watershed/                 # Watershed segmentation
│   ├── watershed_opencv.py    # OpenCV pipeline (Otsu + distance transform + watershed)
│   └── watershed_skimage.py   # scikit-image pipeline (mean-shift + peak local max + watershed)
│
├── itk_segmentation/          # ITK-based segmentation
│   ├── geodesic_active_contour.py  # Full ITK GAC pipeline (smoothing → gradient → sigmoid → fast marching → GAC)
│   └── watershed_itk.py       # ITK watershed with gradient magnitude preprocessing
│
├── medical_pipeline/          # End-to-end medical imaging pipelines
│   └── nifti_segmentation.py  # NIfTI volume segmentation with connected components + active contour
│
└── visualization_3d/          # 3D visualization
    └── bone_viewer.py         # PyVista-based volume rendering and multi-slice orthogonal views

Installation

git clone https://github.com/alirezashadmani/MedicalImaging.git
cd MedicalImaging
pip install -r requirements.txt

Optional dependencies (install only if needed):

pip install itk              # For itk_segmentation module
pip install nibabel cc3d     # For medical_pipeline module
pip install pyvista SimpleITK  # For visualization_3d module
pip install PyMCubes         # For 3D level-set visualization

Usage

Every script supports --help for full argument documentation. All image paths are passed via CLI arguments (no hardcoded paths).

Edge Detection

# Canny with multiple sigma comparison
python -m edge_detection.canny_skimage --image data/scan.png --sigma 1.0 3.0 5.0

# Manual Canny implementation (shows all intermediate stages)
python -m edge_detection.canny_manual --image data/scan.png --sigma 2.5 --low 40 --high 100

# Sobel + Laplacian (optionally with Canny overlay)
python -m edge_detection.sobel --image data/scan.png --with-canny

Active Contour / Level Set

# Parametric snake with circular initialization
python -m active_contour.parametric_snake --image data/scan.png --center 500 950 --radius 50

# Morphological Chan-Vese (region-based, no edge function needed)
python -m active_contour.morph_acwe --image data/scan.png --smoothing 5 --lambda1 3 --lambda2 1

# Morphological Geodesic Active Contour (edge-based)
python -m active_contour.morph_gac --image data/scan.png --alpha 100 --sigma 4.5

# 3D level set on volumetric data
python -m active_contour.morph_acwe_3d --volume data/confocal.npy --iterations 150

# Manual level-set evolution with curvature and balloon forces
python -m active_contour.level_set_evolution --image data/scan.png --iterations 50 --balloon 1.0

Watershed Segmentation

# OpenCV-based (with full pipeline visualization)
python -m watershed.watershed_opencv --image data/scan.png --show-steps

# scikit-image-based (mean-shift + distance transform)
python -m watershed.watershed_skimage --image data/scan.png --min-distance 20

ITK Segmentation

# Geodesic Active Contour (with optional intermediate outputs)
python -m itk_segmentation.geodesic_active_contour \
    --image data/scan.png --output result.png \
    --seed 56 92 --propagation 7.0 --save-intermediates

# ITK Watershed
python -m itk_segmentation.watershed_itk \
    --image data/scan.png --output result.png \
    --threshold 0.005 --level 0.5

Medical Pipeline (NIfTI)

# Full pipeline: NIfTI volume → connected components → convex hull → active contour
python -m medical_pipeline.nifti_segmentation \
    --img-path data/case.nii.gz \
    --seg-path data/seg.nii.gz \
    --full-path data/full_view.nii.gz \
    --obj-id 4 --alpha 18 --beta 0.01

3D Visualization

# Volume rendering (bone + ground truth overlay)
python -m visualization_3d.bone_viewer volume \
    --bone-image data/bone.nii.gz --gt-image data/gt.nii.gz

# Orthogonal slice viewer
python -m visualization_3d.bone_viewer slices --nrrd-file data/volume.nrrd

# Combined 4-panel view (XY, XZ, ZY slices + 3D volume)
python -m visualization_3d.bone_viewer combined \
    --nrrd-file data/slices.nrrd --volume-file data/volume.nrrd

Saving Output

All scripts support --output to save figures to disk instead of displaying:

python -m watershed.watershed_opencv --image data/scan.png --show-steps --output results/watershed.png

Algorithms Overview

Algorithm	Type	Module	Best For
Canny	Edge detection	edge_detection	Boundary detection, preprocessing
Sobel / Laplacian	Edge detection	edge_detection	Gradient-based edge maps
Parametric Snake	Active contour	active_contour	Smooth boundary segmentation
Chan-Vese (ACWE)	Level set	active_contour	Region-based segmentation without strong edges
Geodesic AC (GAC)	Level set	active_contour	Edge-based segmentation with balloon force
Watershed (OpenCV)	Region growing	watershed	Over-segmentation, marker-based
Watershed (skimage)	Region growing	watershed	Distance-transform-based seed detection
ITK GAC	Level set	itk_segmentation	Clinical-grade geodesic active contour
ITK Watershed	Region growing	itk_segmentation	ITK-native watershed with colormap output

Tech Stack

• Core: NumPy, SciPy, OpenCV, scikit-image, Matplotlib
• Level sets: morphsnakes
• Medical imaging: ITK, nibabel, SimpleITK, cc3d
• 3D visualization: PyVista, VTK

License

This project is for research and educational purposes.