Medical Imaging Demo - User Guide

Overview

The medical imaging demo (examples/scientific/medical_imaging_demo.xdl) demonstrates XDL’s capabilities for medical image analysis and 3D visualization. It generates a synthetic CT head scan with realistic anatomical structures and provides interactive 3D visualization.

Features

Data Generation

  • 3D CT Volume: 64×64×32 voxels (configurable)
  • Anatomical Structures:
    • Skull (cortical bone)
    • Brain tissue (gray and white matter)
    • Air-filled cavities
  • Hounsfield Units: Physically realistic HU values
    • Air: -1000 HU
    • Soft tissue: 20-45 HU
    • Bone: 700-3000 HU

Analysis Capabilities

  1. Tissue Segmentation
    • Air (HU < -500)
    • Soft tissue (-100 ≤ HU < 200)
    • Bone (HU ≥ 200)
  2. Multi-Planar Reconstruction (MPR)
    • Axial slices (transverse plane)
    • Coronal slices (frontal plane)
    • Sagittal slices (lateral plane)
  3. CT Windowing
    • Brain window (center: 40 HU, width: 80 HU)
    • Bone window (center: 500 HU, width: 2000 HU)
  4. Volume Statistics
    • Mean and standard deviation of HU values
    • Min/max HU range
    • Tissue-specific statistics
  5. Image Quality Assessment
    • Signal-to-noise ratio (SNR)
    • Dynamic range
    • Voxel count and volume dimensions
  6. 3D Volume Visualization ✨
    • Interactive ray-casting volume renderer
    • Grayscale colormap (standard for medical imaging)
    • Real-time threshold adjustment
    • Multi-tissue visualization

Running the Demo

Basic Execution 

# Run the complete demo with 3D visualization
xdl examples/scientific/medical_imaging_demo.xdl

Environment Variables 

# Use specific VIZ3D backend
VIZ3D_BACKEND=threejs xdl examples/scientific/medical_imaging_demo.xdl

# Available backends:
# - threejs (default): Three.js WebGL
# - webgpu: Native WebGPU renderer
# - browser: Browser-based WebGPU

Workflow Steps

Step 1: Generate Synthetic CT Data

Creates a 64×64×32 volume with:

  • Air-filled background (-1000 HU)
  • Skull shell (bone: 1400-1600 HU)
  • Brain tissue (20-45 HU):
    • Gray matter (outer): ~40 HU
    • White matter (inner): ~30 HU

Step 2: Tissue Segmentation

Segments volume into three tissue classes:

  • Air: HU < -500 (~55% of volume)
  • Soft tissue: -100 ≤ HU < 200 (~33% of volume)
  • Bone: HU ≥ 200 (~12% of volume)

Step 3-5: Multi-Planar Reconstruction

Extracts representative slices in three orthogonal planes:

  • Axial: At z = 16 (mid-head level)
  • Coronal: At y = 32 (mid-frontal)
  • Sagittal: At x = 32 (mid-lateral)

Step 6: CT Windowing Presets

Defines standard clinical window settings:

Brain Window:

  • Center: 40 HU
  • Width: 80 HU
  • Range: 0-80 HU
  • Purpose: Optimize gray/white matter contrast

Bone Window:

  • Center: 500 HU
  • Width: 2000 HU
  • Range: -500 to 1500 HU
  • Purpose: Visualize skull fractures and bone detail

Step 7: Volume Statistics

Computes comprehensive statistics:

  • Mean HU value (~-351 HU, air-dominated)
  • Standard deviation (~837 HU)
  • HU range: -1000 to +1600 HU
  • Total voxels: 131,072 (64×64×32)

Step 8: Tissue-Specific Statistics

Calculates mean HU for each tissue type:

  • Bone: ~1500 HU
  • Soft tissue: ~38 HU (brain)
  • Air: -1000 HU (exact)

Step 9: Image Quality Metrics

Assesses scan quality:

  • Signal-to-noise ratio (SNR)
  • Dynamic range: ~2600 HU
  • Voxel count and dimensions

Step 10: 3D Volume Visualization ✨

Interactive 3D rendering of the CT head scan:

  • Window: 1280×720 HD display
  • Colormap: Grayscale (medical imaging standard)
  • Camera: Positioned for optimal head viewing
  • Volume: Full 64×64×32 CT scan
  • Features visible: Skull, brain tissue, anatomical structure

3D Visualization Controls

Mouse Controls

  • Left Click + Drag: Rotate camera around head
  • Mouse Wheel: Zoom in/out
  • Right Click + Drag: Pan camera

GUI Controls

The viewer includes sliders for:

Threshold Slider (key control for medical imaging):

  • Minimum (-1000 HU): Shows everything (air + tissue + bone)
  • -500 HU: Remove air, show tissue + bone
  • 200 HU: Show only bone (skull)
  • 500 HU: Show dense bone only

Opacity Slider:

  • Adjust transparency of entire volume
  • Lower values: See through structures
  • Higher values: Solid rendering

Usage Tips:

  1. Start with threshold at minimum to see full volume
  2. Increase threshold to ~-500 to remove air and see head outline
  3. Increase to ~200 to visualize skull structure
  4. Adjust opacity to see internal/external features

Keyboard Shortcuts

  • ESC: Close viewer
  • R: Reset camera to default position
  • Space: Pause/resume auto-rotation (if enabled)

Clinical View Presets

Use threshold settings to simulate clinical views:

Soft Tissue View (threshold: -500 HU):

  • Shows brain tissue and skull
  • Removes air background
  • Good for overall anatomy

Bone View (threshold: 200 HU):

  • Shows skull only
  • Useful for fracture detection
  • Visualize bone structure

Dense Bone View (threshold: 500 HU):

  • Shows only cortical bone
  • Highest density structures
  • Clear skull outline

Output Example 

> Step 10: Launching 3D visualization...

VIZ3D: Initialized (1280x720)
  Volume loaded:  64  x  64  x  32
  HU range: [ -1000 ,  1600 ]

  Rendering 3D CT volume...
  - Dark: Air (-1000 HU)
  - Gray: Soft tissue (20-45 HU)
  - Bright: Bone (700-3000 HU)
  - Structures: Skull and brain tissue visible

  Adjust threshold slider to:
    • Show only bone (threshold > 200 HU)
    • Show soft tissue (threshold > -500 HU)
    • Include air (threshold at minimum)

🚀 Launching Three.js volume visualization...
Controls:
  - Left mouse: Rotate camera
  - Mouse wheel: Zoom in/out
  - GUI sliders: Adjust threshold and opacity

✓ Three.js volume visualization launched
  Volume: 64×64×32
  Colormap: GRAYSCALE

✓ 3D visualization complete!

Key Results

Typical output from the demo:

Key Findings:
  • Volume size:  64 x 64 x 32
  • Mean HU:  -351 HU
  • Bone coverage:  12.2%
  • Soft tissue coverage:  33.1%
  • Air coverage:  54.7%
  • SNR:  0.048

Tissue-Specific Mean HU:
  • Bone:  1500 HU (n=16044)
  • Soft tissue:  38 HU (n=43324)
  • Air:  -1000 HU (n=71704)

Medical Interpretation

Anatomical Features

Skull Structure:

  • Ellipsoidal shell representing cranium
  • Cortical bone with realistic HU values (1400-1600)
  • Shell thickness: ~3 voxels (realistic for skull)

Brain Tissue:

  • Gray matter (outer): ~40 HU (typical)
  • White matter (inner): ~30 HU (typical)
  • Realistic tissue differentiation
  • Spherical geometry inside skull

Air Spaces:

  • Surrounding volume: -1000 HU (air density)
  • Represents extracranial space
  • Useful for windowing demonstration

Clinical Applications

This demo simulates workflows for:

  1. Trauma Imaging
    • Skull fracture detection (bone window)
    • Hemorrhage identification (brain window)
    • 3D reconstruction for surgical planning
  2. Neurological Assessment
    • Gray/white matter differentiation
    • Volume measurements
    • Tissue density analysis
  3. Surgical Planning
    • 3D visualization of anatomy
    • Measurement tools
    • Multi-planar reconstruction
  4. Education and Training
    • Teaching CT interpretation
    • Understanding Hounsfield units
    • Windowing technique practice

Technical Implementation

Array Operations

Uses XDL’s multi-dimensional array support:

ct_volume = FLTARR(nx, ny, nz)  ; 3D array
min_hu = MIN(ct_volume)         ; Works on entire volume
max_hu = MAX(ct_volume)         ; No manual loops needed
mean_hu = MEAN(ct_volume)       ; Direct computation
stddev_hu = STDDEV(ct_volume)   ; Statistical function

VIZ3D Functions Used 

VIZ3D_INIT, WINDOW_SIZE=[w, h], TITLE='...'
VIZ3D_COLORMAP, 'GRAYSCALE'
VIZ3D_CAMERA, POSITION=[x,y,z], TARGET=[x,y,z], FOV=45.0
VIZ3D_VOLUME, ct_volume, DIMENSIONS=[nx,ny,nz]
VIZ3D_RENDER, /INTERACTIVE, TITLE='...'

Performance

  • Generation: ~1-2 seconds for 64³ volume
  • Segmentation: ~0.5 seconds
  • Rendering: Real-time (60fps) with Three.js
  • Memory: ~50MB for 64×64×32 float array

Customization

Change Volume Size 

; Higher resolution (more detail)
nx = 128
ny = 128
nz = 64

; Lower resolution (faster)
nx = 32
ny = 32
nz = 16

Modify Anatomical Features 

; Larger skull
skull_outer_radius = 30.0  ; instead of 28.0

; Thicker skull
skull_inner_radius = 23.0  ; instead of 25.0

; Different tissue densities
gray_matter = 45.0   ; instead of 40.0
white_matter = 35.0  ; instead of 30.0

Different Colormap 

VIZ3D_COLORMAP, 'VIRIDIS'  ; Color instead of grayscale
VIZ3D_COLORMAP, 'TURBO'    ; Rainbow-like medical colormap

Applications

This demo illustrates workflows for:

  1. Clinical Radiology
    • CT scan interpretation
    • Multi-planar reconstruction
    • 3D visualization
  2. Medical Education
    • Teaching CT physics
    • Understanding Hounsfield units
    • Practicing windowing techniques
  3. Research
    • Image processing algorithm development
    • Segmentation validation
    • Quantitative analysis
  4. Surgical Planning
    • 3D anatomy visualization
    • Measurement and planning
    • Patient-specific modeling
  • geophysical_demo.xdl - Seismic 3D visualization
  • viz3d_demo1_gaussian.xdl - Simple 3D Gaussian
  • viz3d_showcase.xdl - Multiple 3D demos

Next Steps

To extend the demo:

  1. Import DICOM Data: Load real CT scans
  2. Advanced Segmentation: Deep learning-based tissue classification
  3. Image Registration: Align multiple scans
  4. Dose Planning: Radiation therapy planning
  5. Surface Rendering: Extract and render skull surface
  6. Measurement Tools: Distance, volume, density measurements

Technical Requirements

  • XDL interpreter with VIZ3D support
  • ~50MB RAM for 64³ volume
  • WebGL-capable GPU
  • Modern web browser (embedded Tauri window)

Troubleshooting

Visualization doesn’t appear 

# Verify backend
VIZ3D_BACKEND=threejs xdl examples/scientific/medical_imaging_demo.xdl

# Check logs
RUST_LOG=debug xdl examples/scientific/medical_imaging_demo.xdl

Poor visualization quality

  • Increase volume size (nx, ny, nz)
  • Adjust threshold to focus on structures of interest
  • Use appropriate windowing (brain vs bone)

Performance issues

  • Reduce volume size
  • Lower opacity setting
  • Close other GPU-intensive applications

Clinical Context

Hounsfield Scale Reference

  • Air: -1000 HU
  • Lung: -500 HU
  • Fat: -100 to -50 HU
  • Water: 0 HU
  • Soft tissue: 20-70 HU
  • Blood: 30-45 HU
  • Muscle: 10-40 HU
  • Gray matter: 37-45 HU
  • White matter: 20-30 HU
  • Bone: 700-3000 HU

Standard CT Windows

  • Brain: C=40, W=80
  • Subdural: C=75, W=215
  • Stroke: C=40, W=40
  • Bone: C=500, W=2000
  • Soft tissue: C=50, W=350
  • Lung: C=-600, W=1500

References

Author

Generated with XDL - Extended Data Language