Texture Generation

texture_generation - MRI Texture Feature Extraction for Radiomics

Part of the micaflow processing pipeline for neuroimaging data.

This module computes advanced texture features from MRI data that can be used for tissue characterization, lesion analysis, radiomics applications, and quantitative imaging biomarker development. It performs automatic tissue segmentation and extracts quantitative imaging features including gradient magnitude and relative intensity maps, which capture local intensity variations and tissue contrast properties respectively.

What are Texture Features?

Texture features quantify spatial patterns and intensity variations in medical images: - Provide information beyond simple intensity statistics - Capture tissue heterogeneity and structural organization - Enable quantitative characterization of pathology - Support radiomics and machine learning applications - Complement traditional volumetric measurements

Computed Features:

  1. Gradient Magnitude: - Measures local intensity changes - Highlights edges and boundaries - Captures tissue transitions - Sensitive to structural organization - Range: 0 to maximum gradient

  2. Relative Intensity (RI): - Normalized intensity relative to tissue peaks - Accounts for scanner/protocol variations - Centers around GM-WM boundary - Range: Typically 0-200 (100 = background) - Provides contrast-normalized values

How It Works:

  1. Load input MRI and brain mask

  2. Calculate brain-masked mean intensity for background reference

  3. Compute gradient magnitude (edge detection)

  4. Calculate relative intensity (normalized contrast mapping)

  5. Apply smoothing to relative intensity map

  6. Save all feature maps as NIfTI files

Relative Intensity Calculation:

  1. Calculate mean intensity within the brain mask as background reference (BG)

  2. Normalization: RI = 100 * (I / BG) (Centers the mean intensity at 100)

  3. Smooth with Gaussian (σ=3mm FWHM)

  4. Re-mask to ensure background remains clean

Command-Line Usage:

# Basic usage micaflow texture_generation

–input <path/to/image.nii.gz> –mask <path/to/brain_mask.nii.gz> –output <path/to/output_prefix>

# Example with T1w image micaflow texture_generation

–input T1w_preprocessed.nii.gz –mask brain_mask.nii.gz –output subject01_textures

# Example with short flags micaflow texture_generation

-i T1w.nii.gz -m mask.nii.gz -o output/features

Python API Usage:

>>> from micaflow.scripts.texture_generation import run_texture_pipeline
>>>
>>> # Basic usage
>>> run_texture_pipeline(
...     input="preprocessed_t1w.nii.gz",
...     mask="brain_mask.nii.gz",
...     output_dir="output_texture_maps"
... )
>>>
>>> # With variables
>>> input_file = "data/T1w.nii.gz"
>>> mask_file = "data/mask.nii.gz"
>>> output_prefix = "results/subject01"
>>> run_texture_pipeline(input_file, mask_file, output_prefix)

Pipeline Integration:

Texture generation typically follows preprocessing:

Structural MRI Pipeline: 1. Preprocessing (N4 bias correction, denoising) 2. Brain extraction (skull stripping) 3. Tissue segmentation (optional: use synthseg instead) 4. Texture feature extractiom 5. Statistical analysis or machine learning

Radiomics Pipeline: 1. Image acquisition and quality control 2. Preprocessing and standardization 3. ROI/lesion segmentation 4. Feature extraction (texture_generation) 5. Feature selection 6. Model training/prediction

Exit Codes:

0 : Success - texture features computed successfully 1 : Error - invalid inputs, file not found, or processing failure

See Also:

  • synthseg : Alternative segmentation method

  • n4_bias_correction : Recommended preprocessing

  • denoise : Noise reduction before feature extraction

Command Line Usage

micaflow texture_generation [options]

Source Code

View the source code: GitHub Repository

Description

║ (Radiomics) ║

This script generates quantitative texture feature maps from MRI data for radiomics, tissue characterization, and quantitative imaging biomarker development. Features include gradient magnitude and relative intensity.

Full Help

╔════════════════════════════════════════════════════════════════╗
║                  TEXTURE FEATURE EXTRACTION                    ║
║                        (Radiomics)                             ║
╚════════════════════════════════════════════════════════════════╝

This script generates quantitative texture feature maps from MRI data
for radiomics, tissue characterization, and quantitative imaging biomarker
development. Features include gradient magnitude and relative intensity.

────────────────────────── USAGE ──────────────────────────
  micaflow texture_generation [options]

─────────────────── REQUIRED ARGUMENTS ───────────────────
  --input, -i   : Path to the input MRI image (.nii.gz)
  --mask, -m    : Path to the binary brain mask (.nii.gz)
  --output, -o  : Output prefix for texture feature maps

──────────────────── EXAMPLE USAGE ──────────────────────

# Example 1: Basic usage
micaflow texture_generation \
  --input T1w_preprocessed.nii.gz \
  --mask brain_mask.nii.gz \
  --output subject01_textures

# Example 2: With short flags
micaflow texture_generation \
  -i T1w.nii.gz \
  -m mask.nii.gz \
  -o output/features

# Example 3: From preprocessed data
micaflow texture_generation \
  -i data/preprocessed/T1w_n4.nii.gz \
  -m data/masks/brain_mask.nii.gz \
  -o results/radiomics/sub-01

────────── WHAT ARE TEXTURE FEATURES? ──────────────────

Texture features quantify spatial patterns in medical images:
• Go beyond simple intensity statistics
• Capture tissue heterogeneity and organization
• Provide quantitative imaging biomarkers
• Enable radiomics and machine learning
• Support precision medicine approaches

Common applications:
• Tumor characterization and grading
• Neurodegenerative disease staging
• White matter lesion analysis
• Age-related changes
• Treatment response monitoring

─────────────── COMPUTED FEATURES ──────────────────────

1. Gradient Magnitude:
• Measures local intensity changes
• Highlights edges and tissue boundaries
• Sensitive to structural organization
• Range: 0 to maximum gradient
• Higher values at GM/WM boundaries

2. Relative Intensity (RI):
• Normalized intensity relative to tissue peaks
• Accounts for scanner/protocol variations
• Centers around GM-WM boundary (value = 100)
• Range: Typically 50-150
• Provides contrast-normalized values

3. Tissue Segmentation (intermediate):
• Gray matter mask
• White matter mask
• CSF identification
• Used for feature computation

────────────────── HOW IT WORKS ────────────────────────

Processing pipeline:
1. Load input MRI and brain mask
2. Calculate mean intensity within mask (background reference)
3. Compute gradient magnitude (edge detection)
4. Calculate relative intensity (normalized contrast globally scaled)
5. Apply smoothing (σ=3mm FWHM)
6. Save feature maps as NIfTI files

Atropos segmentation:
• K-means clustering (3 tissue classes)
• MRF spatial prior (0.2 weight, 1x1x1 neighborhood)
• 3 iterations for convergence
• Masked to brain tissue only
• Output: 1=CSF, 2=GM, 3=WM

────────────────────── OUTPUT FILES ─────────────────────

Given output prefix "subject01_textures", creates:

subject01_textures_gradient-magnitude.nii.gz
  - Edge and boundary detection map
  - Higher values at tissue transitions
  - Range: 0 to ~50 (depends on contrast)

subject01_textures_relative-intensity.nii.gz
  - Normalized contrast map
  - Centered at GM-WM boundary (100)
  - Range: Typically 50-150
  - Smoothed with σ=3mm FWHM

────────────────────────── NOTES ─────────────────────────
• Preprocessing recommended: N4 bias correction, denoising
• Processing time: 2-5 minutes per subject
• Memory: ~2-4 GB RAM
• All operations masked to brain only
• No resampling (preserves input resolution)
• Values outside mask are zero
• Gradient uses first-order finite differences
• Peak finding uses 1st-99.5th percentile range

─────────────── PIPELINE POSITION ───────────────────────
Structural MRI Pipeline:
1. Preprocessing (N4, denoising)
2. Brain extraction (skull stripping)
3. Tissue segmentation (optional)
4. Texture feature extraction ← You are here
5. Statistical analysis or ML

Radiomics Pipeline:
1. Image acquisition and QC
2. Preprocessing and standardization
3. ROI/lesion segmentation
4. Feature extraction ← You are here
5. Feature selection
6. Model training/prediction

────────────────────── EXIT CODES ───────────────────────
0 : Success - texture features computed
1 : Error - invalid inputs, file not found, or processing failure

───────────── QUALITY CONTROL ───────────────────────────
Visual inspection:
1. Check gradient magnitude highlights tissue boundaries
2. Verify relative intensity centered around 100
3. Ensure smooth transitions (no edge artifacts)
4. Confirm brain mask coverage is complete
5. Look for segmentation errors

Expected ranges:
• Gradient magnitude: 0 to ~50
• Relative intensity: 50-150 (100 = background)
• Values outside mask: 0

───────────────── COMMON ISSUES ─────────────────────────
Issue: Poor segmentation quality
Solution: Use N4 bias correction first, verify mask quality

Issue: Extreme gradient values
Solution: Check image quality, reduce noise with denoising

Issue: Relative intensity not centered at 100
Solution: Verify tissue segmentation, check for artifacts

Issue: Processing takes very long
Solution: Normal for high-resolution images (2-5 min typical)