3D-FACE-Classification

Advanced Analysis and Classification of Facial and Mimic Muscle Anatomy using 3D MRI Data

A comprehensive collection of processing and analysis scripts for automated segmentation, classification, and quantitative characterisation of facial anatomy, muscle groups, and bone structures from 3D MRI data.

Stark H, Jandausch K, Roth A, Biedermann U, Krämer M, Reichenbach JR, Aschenbach R, Volk GF, Fischer MS, Guntinas-Lichius O (2026) The Anatomy of Facial Muscles Revisited: High-Resolution Magnetic Resonance Imaging and Computed Tomography Studies on Body Donors. Journal of Imaging 12(9), 459. DOI:10.3390/jimaging12090459

Overview

3D-FACE-Classification is a specialised research project focused on the detailed analysis, segmentation, and classification of facial anatomy from high-resolution 3D MRI (Magnetic Resonance Imaging) datasets. The project provides an automated pipeline for:

• Multi-rater segmentation consensus - Combining independent expert annotations
• Muscle group classification - Categorizing facial musculature into functional groups
• Quantitative anatomy - Measuring tissue thickness, distribution, and spatial relationships
• Anatomical variation analysis - Understanding differences across subjects
• Biomechanical characterization - Analysing muscle organisation for functional understanding

This project combines medical image analysis, consensus methodology, statistical analysis, and detailed anatomical classification to create comprehensive 3D models of facial anatomy.

Project Context

Facial Anatomy Research

Understanding facial anatomy is essential for:

• Medical research - Understanding facial function and dysfunction
• Surgical planning - Precise anatomical knowledge for safe intervention
• Dermatology - Muscle-related ageing and ageing-related changes
• Orthodontics - Masticatory muscle organisation and function
• Speech-language pathology - Understanding mimic and speech-related muscles
• Cosmetic medicine - Evidence-based approach to facial aesthetics

MRI Technology in Facial Imaging

3D MRI provides:

• High soft tissue contrast - Excellent muscle-fat differentiation
• Non-invasive imaging - No radiation exposure
• High resolution - Sub-millimeter spatial resolution achievable
• 3D volumetric data - Complete spatial information
• In vivo imaging - Analysis of living subjects
• Reproducibility - Consistent image acquisition protocols

Multi-Rater Consensus Approach

The project implements best practices for handling multiple expert annotations:

• Rater agreement analysis - Assessing reproducibility and consistency
• Consensus building - Combining ratings from multiple experts
• Inter-rater reliability - Statistical measures of agreement
• Quality control - Identifying and managing discrepancies
• Anatomical validation - Cross-referencing with established anatomy

Repository Contents

Overview of Processing Pipeline

Raw 3D MRI Data
        ↓
[mrt_orientation.inc] - Coordinate system alignment & landmark detection
        ↓
[repair_transformation.macro] - Standardise coordinate systems across raters
        ↓
[rater.macro] - Analyse individual rater segmentations
        ↓
[best.macro] - Create consensus from 2+ raters
[best_x3.macro] - Create consensus from 3+ raters
        ↓
[thickness.macro] - Calculate muscle thickness maps
[distance.macro] - Calculate bone-muscle distance maps
        ↓
[stat_*.macro] - Statistical analysis by anatomical region
[stat_*_sub.macro] - Statistical analysis subdivided into 16 regions
        ↓
Anatomical Classification & Results

Script Categories

A. Data Preparation & Consensus Building

1. mrt_orientation.inc - MRI Coordinate System Alignment

Purpose: Standardises MRI coordinate systems and detects anatomical landmarks for consistent spatial orientation.

Key Functions:

• Loads bone segmentation in the original coordinate system
• Detects anatomical landmarks from marker files (.mrk.json format):
  • Eye landmarks (right eye upper/lower)
  • Anatomical reference points
  • Multiple marker file format support (fallback chain)
• Applies coordinate transformations:
  • Coordinate system inversion
  • Global-to-local coordinate conversion
  • Scaling adjustments for isotropic spacing
• Establishes reference coordinate system for all subsequent analyses

Input Files:

• Bone segmentation files (NIFTI format)
• Anatomical marker files (JSON format)
  • *_1.mrk.json, *_auge rechts oben.mrk.json (right eye upper)
  • *_2.mrk.json, *_auge rechts unten.mrk.json (right eye lower)
  • Additional anatomical markers as needed

Output:

• Standardised coordinate system definition
• Transformation matrices
• Landmark coordinates in standard space

Technical Details:

Coordinate Transformation:
1. Apply reflection: <-1, -1, 1> (mirror across specific axes)
2. Convert global to local coordinates: scalar.g2l
3. Apply scaling: <2, 2, 2> (achieve isotropic spacing)

2. repair_transformation.macro - Coordinate System Standardization

Purpose: Repairs inconsistencies in coordinate systems used by different raters and standardises all data to a common reference system.

Key Operations:

• Identifies coordinate system deviations
• Applies correction transformations
• Standardises all rater segmentations to a single reference (rater HS - Heiko Stark)
• Validates transformation accuracy
• Ensures inter-rater data compatibility

Background: Different raters may use different MRI coordinate system conventions. This script identifies and corrects these differences, ensuring all segmentations are comparable.

Input: Individual rater segmentation files with varying coordinate systems Output: Transformed segmentations in a standardised reference frame

3. best.macro - Binary Consensus (2+ Raters)

Purpose: Creates a consensus muscle segmentation combining annotations from a minimum of 2 independent raters.

Key Operations:

• Loads all mask files for a subject with the file naming convention mask*.nii.gz
• Excludes derived files:
  • mask_sub.nii.gz (subdivision masks)
  • mask_XX.nii.gz (consensus result)
  • mask_X3.nii.gz (3-rater consensus)
  • mask_*split.nii.gz (split versions)
• Sums rater segmentations
• Thresholds to binary consensus (value ≥ 2)
• Saves consensus mask as mask_XX.nii.gz

Algorithm:

For each subject directory:
  1. Load all individual rater masks
  2. Sum masks: mask_total = mask_rater1 + mask_rater2 + ...
  3. Apply threshold: consensus = (mask_total ≥ 2) ? 1 : 0
  4. Save as mask_XX.nii.gz

Input: Individual rater segmentation files Output: mask_XX.nii.gz - Binary consensus mask

File Naming Convention:

• Individual raters: mask_HS.nii.gz, mask_KS.nii.gz, mask_DS.nii.gz, etc.
• 2+ consensus: mask_XX.nii.gz
• 3+ consensus: mask_X3.nii.gz

4. best_x3.macro - High-Confidence Consensus (3+ Raters)

Purpose: Creates a more stringent consensus requiring agreement from 3 or more independent raters.

Key Operations:

• Similar to best.macro but requires a minimum of 3 raters
• Generates mask_X3.nii.gz with higher confidence
• Useful for controversial anatomical regions
• Better for validation and benchmark datasets

Algorithm:

Consensus = (mask_total ≥ 3) ? 1 : 0

Use Cases:

• Creating high-quality reference/gold-standard segmentations
• Validating algorithm performance
• Analysing agreement patterns
• Identifying anatomically ambiguous regions

B. Muscle & Tissue Characterization

5. thickness.macro - Muscle Thickness Calculation

Purpose: Computes local muscle thickness at each voxel using distance transform analysis.

Key Operations:

• Loads consensus muscle mask (mask_XX.nii.gz)
• Sets maximum scan length: 60 mm (thickness radius)
• Computes distance transform via local thickness measurement
• Applies fullscan=false for optimisation (incomplete sphere scans acceptable)
• Saves thickness map: Thickness/{ID}_thickness.nii.gz
• Enables parallel processing (thread.max := 40)

Thickness Analysis:

• Local thickness (LT): Perpendicular distance to nearest muscle boundary
• Medial axis: Central axis through muscle structure
• Thickness distribution: Histogram of thickness values
• Mean thickness: Average across muscle region
• Thickness variation: Spatial variation of muscle thickness

Input: Consensus muscle segmentation Output:

• 3D thickness map showing local muscle thickness at each point
• Unit: millimetres
• Range: 0-60 mm (maximum scan radius)

Interpretation:

• Low values (0-5 mm): Thin muscle tissue at boundaries
• Medium values (5-15 mm): Typical muscle thickness
• High values (15-60 mm): Thick muscle bundles or muscle groups
• Zero values: Outside muscle region

6. distance.macro - Bone-Muscle Distance Analysis

Purpose: Calculates the 3D spatial distance from muscle segmentations to bone surfaces, revealing muscle-to-bone relationships.

Key Operations:

• Computes distance from each muscle voxel to the nearest bone boundary
• Generates distance field in 3D space
• Saves as distance maps for statistical analysis
• Reveals anatomical muscle organisation relative to skeleton
• Important for understanding functional muscle groups

Applications:

• Analysing muscle attachment sites
• Understanding muscle-bone relationships
• Identifying muscle layers and stratification
• Biomechanical significance of muscle positioning

Output: 3D distance field showing muscle proximity to bone surfaces

C. Statistical Analysis

The repository includes comprehensive statistical analysis scripts for multiple levels of anatomical classification:

7. stat.macro - Muscle Group Statistics

Purpose: Generates statistical summaries for major muscle group classifications.

Key Operations:

• Loads individual rater classifications (mask*.nii.gz)
• Loads anatomical subdivision mask (mask_sub.nii.gz)
• Classifies voxels into muscle groups:
  1. Unknown - Unclassified tissue
  2. Mimic muscles - Facial expression muscles
  3. Platysma - Neck muscle extending into lower face
  4. Masticatory muscles - Muscles of mastication
  5. Eye muscles - Orbicularis and extraocular muscles
  6. Tongue muscles - Lingual musculature

Statistical Output:

• Count of voxels per muscle group per rater
• Generates CSV output: stat.csv
• Headers: ID, N_Unknown, N_Mimik, N_Platysma, N_Mastication, N_Eye, N_Tongue

Input: Rater segmentation files and subdivision mask Output: CSV file with muscle group counts

8. stat_rater.macro - Rater Agreement Analysis

Purpose: Analyses inter-rater agreement and reliability metrics.

Key Operations:

• Compares individual rater segmentations against consensus
• Calculates voxel-wise agreement statistics:
  • Agreement count (value = 1)
  • Partial agreement (value = 2)
  • Disagreement (value = 3)
• Outputs agreement metrics per rater
• Generates: stat_rater.csv

Output Metrics:

• Voxel count for each agreement level
• Rater reliability indicators
• Consistency assessment

9. stat_*_sub.macro - Regional Subdivision Analysis

Purpose: Provides fine-grained spatial analysis by subdividing muscles into 16 regional parts.

Scripts:

• stat_sub.macro - General subdivision statistics
• stat_bone_sub.macro - Bone structure analysis by region
• stat_distance_sub.macro - Distance field analysis by region
• stat_fat_sub.macro - Adipose (fat) tissue analysis
• stat_fat_in_sub.macro - Internal fat analysis
• stat_fat_out_sub.macro - External fat analysis
• stat_fat_intensities_sub.macro - Fat tissue MRI intensity analysis
• stat_intensities_sub.macro - Overall MRI intensity analysis
• stat_thickness_sub.macro - Thickness analysis by regional subdivision

Regional Subdivision Scheme: The face is subdivided into 16 regions (4 x 4 grid or hierarchical subdivision):

• Anterior/Posterior division
• Superior/Inferior division
• Left/Right subdivision
• Regional mapping for spatial analysis

Output: Detailed statistics for each of 16 regions:

• Voxel counts
• Intensity statistics
• Thickness measurements
• Distance measurements
• Tissue composition

D. Helper Files & Documentation

10. mrt_orientation_checkup.inc - Validation Helper

Purpose: Validates coordinate system alignment and transformation correctness.

Key Operations:

• Verifies landmark detection success
• Checks coordinate system transformations
• Validates geometric consistency
• Diagnostic output for troubleshooting

Input: Transformed anatomy data Output: Validation report and diagnostic metrics

11. imagexd.macro - Command Reference

Complete reference documentation for all imagexd macro language commands used throughout the project.

File Organisation

Directory Structure:

3D-FACE-classification/
├── Processing Scripts
│   ├── best.macro                      # Consensus generation
│   ├── best_x3.macro                   # High-confidence consensus
│   ├── repair_transformation.macro     # Coordinate standardization
│   ├── thickness.macro                 # Thickness calculation
│   └── distance.macro                  # Distance field computation
│
├── Analysis Scripts (Rater Level)
│   ├── rater.macro                     # Individual rater analysis
│   └── stat_rater.macro                # Rater agreement metrics
│
├── Analysis Scripts (Muscle Groups)
│   ├── stat.macro                      # Major muscle group statistics
│   └── stat_*_sub.macro                # Regional subdivision analysis
│       ├── stat_sub.macro
│       ├── stat_bone_sub.macro
│       ├── stat_distance_sub.macro
│       ├── stat_fat_sub.macro
│       ├── stat_fat_in_sub.macro
│       ├── stat_fat_out_sub.macro
│       ├── stat_fat_intensities_sub.macro
│       ├── stat_intensities_sub.macro
│       └── stat_thickness_sub.macro
│
├── Helper Files
│   ├── mrt_orientation.inc             # Coordinate alignment
│   ├── mrt_orientation_checkup.inc     # Validation
│   └── imagexd.macro                   # Command reference
│
└── LICENSE                             # GPL-3.0

Input Data Structure:

Subject_Data/
├── Subject_001/
│   ├── mask_HS.nii.gz                  # Rater 1 segmentation
│   ├── mask_KS.nii.gz                  # Rater 2 segmentation
│   ├── mask_DS.nii.gz                  # Rater 3 segmentation
│   ├── mask_XX.nii.gz                  # 2+ consensus (generated)
│   ├── mask_X3.nii.gz                  # 3+ consensus (generated)
│   ├── mask_sub.nii.gz                 # Anatomical subdivisions
│   ├── Bones/
│   │   └── Subject_001 Segmentation-Bones.nii.gz
│   ├── Thickness/
│   │   └── Subject_001_thickness.nii.gz (generated)
│   ├── Distance/
│   │   └── Subject_001_distance.nii.gz (generated)
│   └── Markers/
│       ├── F_1.mrk.json                # Landmark files
│       ├── F_2.mrk.json
│       └── ... (additional markers)
│
├── Subject_002/
│   └── (similar structure)
│
└── ...

Output Files:

Results/
├── stat.csv                            # Muscle group statistics
├── stat_rater.csv                      # Rater agreement
├── stat_*_sub.csv                      # Regional analysis results
├── Thickness/
│   ├── Subject_001_thickness.nii.gz
│   ├── Subject_002_thickness.nii.gz
│   └── ...
└── Distance/
    ├── Subject_001_distance.nii.gz
    ├── Subject_002_distance.nii.gz
    └── ...

Technical Details

Software Requirements

Primary Tool: imagexd

Website: https://stark-jena.de/research-interests/software/imagexd/

Features Used:

• 3D image I/O (NIFTI format with gzip compression)
• Scalar field operations (segmentation, masking)
• Distance transform algorithms
• Statistical analysis functions
• Landmark/marker file parsing (JSON)
• Directory and file batch processing
• Parallel multi-threaded computation
• Image alignment and transformation

File Formats

NIFTI Format (.nii.gz):

• Standard neuroimaging format
• Includes spatial metadata (voxel spacing, orientation)
• Gzip-compressed for storage efficiency
• 3D volumetric data support
• Single or multi-channel support

Marker Format (.mrk.json):

• JSON-based anatomical landmark file
• Contains 3D coordinates of anatomical points
• Used for coordinate system alignment
• Compatible with common MRI software (ITK-SNAP, etc.)

CSV Format:

• Comma/tab-separated statistical results
• Headers with clear column descriptions
• Compatible with spreadsheet and statistical software

Data Requirements

MRI Specifications:

• Resolution: Typically 1-2 mm isotropic voxels
• Contrast: T1-weighted or FLAIR sequences
• Field of Strength: 1.5T or higher
• Coverage: Complete facial region including neck
• Quality: High signal-to-noise ratio for accurate segmentation

Segmentation Requirements:

• Independent annotation by trained anatomists/radiologists
• Consistent anatomical definitions across raters
• Coverage of major facial muscle groups
• Bone segmentation for reference

Marker/Landmark Requirements:

• Precise anatomical point identification
• Consistent landmark placement across subjects
• Multiple marker file format support for flexibility

Processing Configuration

Threading:

thread.max := 40
// Adjust based on available CPU cores
// Typical: cores_available - 2 (leave room for system processes)

Thickness Measurement Parameters:

scalar.fullscan false
// false: scan incomplete spheres (faster)
// true: scan only complete spheres (more accurate)

Maximum scan length: 60 mm
// Typical muscle thickness rarely exceeds 40-50 mm

Distance Measurement:

• Euclidean distance in 3D space
• Computed from each muscle voxel to the nearest bone boundary
• Used for muscle layer/stratification analysis

Usage Guide

Prerequisites

1. Install imagexd from https://stark-jena.de/research-interests/software/imagexd/
2. Prepare MRI data with anatomical segmentations from multiple raters
3. Create landmark files with anatomical reference points
4. Organize data in standard directory structure

Data Preparation

Step 1: Organise Input Data

Create directory structure with subject folders:
- Each folder contains rater segmentations: mask_HS.nii.gz, mask_KS.nii.gz, etc.
- Include bone segmentations: Bones/ subdirectory
- Place marker files in the root of the subject directory

Step 2: Run Coordinate Alignment

imagexd mrt_orientation.inc
# This validates and aligns coordinate systems

Step 3: Repair Coordinate Systems (if needed)

imagexd repair_transformation.macro
# Standardise all rater segmentations to common reference

Main Processing Pipeline

Step 1: Generate Consensus Segmentations

# Create 2+ rater consensus
imagexd best.macro
# Generates: mask_XX.nii.gz

# Create 3+ rater consensus (optional, for higher confidence)
imagexd best_x3.macro
# Generates: mask_X3.nii.gz

Step 2: Calculate Tissue Measurements

# Calculate muscle thickness
imagexd thickness.macro
# Generates: Thickness/*.nii.gz

# Calculate bone-muscle distances
imagexd distance.macro
# Generates: Distance/*.nii.gz

Step 3: Analyse Individual Raters

# Generate rater-specific statistics
imagexd rater.macro
# Generates: stat_rater.csv

Step 4: Statistical Analysis

General muscle group statistics:

imagexd stat.macro
# Generates: stat.csv
# Output: voxel counts per muscle group

Regional subdivision statistics:

# General regional statistics
imagexd stat_sub.macro

# Bone structure by region
imagexd stat_bone_sub.macro

# Distance measurements by region
imagexd stat_distance_sub.macro

# Fat tissue analysis
imagexd stat_fat_sub.macro
imagexd stat_fat_in_sub.macro
imagexd stat_fat_out_sub.macro
imagexd stat_fat_intensities_sub.macro

# Intensity analysis
imagexd stat_intensities_sub.macro

# Thickness by region
imagexd stat_thickness_sub.macro

Batch Processing

Automated full pipeline:

Create a shell script process_all.sh:

#!/bin/bash

# Define dataset directory
DATASET="/path/to/data"

cd "$DATASET"

# Step 1: Coordinate alignment
echo "Aligning coordinate systems..."
imagexd mrt_orientation.inc > orientation.log 2>&1

# Step 2: Repair transformations (if needed)
# imagexd repair_transformation.macro > repair.log 2>&1

# Step 3: Consensus generation
echo "Generating consensus segmentations..."
imagexd best.macro > consensus.log 2>&1
imagexd best_x3.macro > consensus_x3.log 2>&1

# Step 4: Tissue characterisation
echo "Calculating tissue measurements..."
imagexd thickness.macro > thickness.log 2>&1
imagexd distance.macro > distance.log 2>&1

# Step 5: Statistical analysis
echo "Running statistical analysis..."
imagexd rater.macro > rater.log 2>&1
imagexd stat.macro > stat.log 2>&1
imagexd stat_sub.macro > stat_sub.log 2>&1
imagexd stat_bone_sub.macro > stat_bone_sub.log 2>&1
imagexd stat_distance_sub.macro > stat_distance_sub.log 2>&1
imagexd stat_fat_sub.macro > stat_fat_sub.log 2>&1
imagexd stat_fat_in_sub.macro > stat_fat_in_sub.log 2>&1
imagexd stat_fat_out_sub.macro > stat_fat_out_sub.log 2>&1
imagexd stat_fat_intensities_sub.macro > stat_fat_intensities_sub.log 2>&1
imagexd stat_intensities_sub.macro > stat_intensities_sub.log 2>&1
imagexd stat_thickness_sub.macro > stat_thickness_sub.log 2>&1

echo "Processing complete!"

Run with:

chmod +x process_all.sh
./process_all.sh

Customization

Modify threading for your system:

Edit thickness.macro or other scripts:
thread.max := 40  # Change to your CPU core count

Adjust thickness measurement parameters:

Edit thickness.macro:
new := scalar.measure.thickness 60  # Change 60 to a different max radius
scalar.fullscan false  # Change to true for complete spheres only

Change output locations: Modify paths in individual scripts:

a := "Thickness/" / x "-" y " thickness.nii.gz"
// Change "Thickness/" to desired output directory

Research Applications

Clinical Applications

Facial Pathology:

• Understanding muscle degeneration in muscular dystrophies
• Analysing muscle changes in facial nerve palsy
• Studying ageing-related changes in facial musculature
• Assessment of post-surgical outcomes

Cosmetic Medicine:

• Objective measurements for botulinum toxin injection planning
• Assessment of muscle atrophy and rejuvenation
• Personalised treatment planning
• Outcome documentation

Orthodontics & Orthognathics:

• Understanding masticatory muscle organisation
• Planning surgical interventions
• Functional analysis of jaw muscles
• Post-surgical follow-up

Research Applications

Anatomy & Variation:

• Documenting normal anatomical variation
• Sex-based differences in facial anatomy
• Age-related changes in muscle structure
• Ethnic variation in facial anatomy

Biomechanics:

• Understanding forces generated by facial muscles
• Analysing muscle fibre organisation
• Modelling facial movement
• Functional analysis of muscle groups

Image Processing & AI:

• Algorithm validation and benchmarking
• Creating gold-standard segmentation datasets
• Training deep learning models
• Evaluating segmentation accuracy

Educational Purposes:

• Creating detailed 3D anatomical models
• Medical student training
• Surgical planning and education
• Anatomical visualisation

Athlete & Performance Analysis

Facial Expression:

• Understanding smile mechanics
• Analysing expression production
• Studying speech-related muscles
• Performance analysis for actors/athletes

Output Data Interpretation

Thickness Maps

Interpretation:

• Units: Millimeters
• Range: 0-60 mm (based on measurement parameters)
• Low values (0-5 mm): Muscle tissue at boundaries, thin muscle layers
• Medium values (5-15 mm): Typical mimic muscle thickness
• High values (15-60 mm): Thick muscle bundles, muscle groups

Clinical Significance:

• Thin muscles: More vulnerable to atrophy, visible movement
• Thick muscles: Greater contractile force, better preservation with age

Distance Maps

Interpretation:

• Units: Millimeters
• Zero values: Muscle directly adjacent to bone
• Small values (0-10 mm): Deep muscle layer, closely associated with skeleton
• Large values (10-50 mm): Superficial layers, distant from bone

Clinical Significance:

• Deep muscles: Primary movers, more stable
• Superficial muscles: Expression muscles, movement generators

Statistical Outputs (CSV)

stat.csv - Muscle Group Counts

ID              N_Unknown  N_Mimik  N_Platysma  N_Mastication  N_Eye  N_Tongue
Subject_001_HS  1250       45600    2300        8900           3200   1100

stat_rater.csv - Rater Agreement

ID              Voxels_Agreement  Voxels_Partial  Voxels_Disagreement
Subject_001_HS  42300             3200            1500

stat_*_sub.csv - Regional Analysis

Region  Thickness_Mean  Thickness_SD  Distance_Mean  Fat_Volume  Bone_Volume
1       8.5             2.3           15.2           450         120
2       9.1             2.1           14.8           480         135
...

Scientific Background

Facial Anatomy Overview

Major Muscle Groups:

1. Mimic Muscles (facial expression)

   • Orbicularis oculi (eye)
   • Orbicularis oris (mouth)
   • Buccinator (cheek)
   • Corrugator supercilii (eyebrow)
   • Frontalis (forehead)
   • Platysma (neck/lower face)

2. Masticatory Muscles

   • Masseter (elevation, closure)
   • Temporalis (elevation, closure)
   • Medial pterygoid (elevation, closure)
   • Lateral pterygoid (opening, lateral movement)
   • Mylohyoid (supporting musculature)

3. Eye Muscles

   • Orbicularis oculi (closure, blinking)
   • Levator palpebrae superioris (opening)
   • Mueller's muscle (sympathetic contribution)

4. Pharyngeal & Lingual Muscles

   • Tongue muscles (intrinsic and extrinsic)
   • Pharyngeal constrictor muscles
   • Soft palate muscles

Muscle Organisation

Layering:

• Deep layer: Attached to bone, primary movers
• Intermediate layer: Crossing muscle fibres, coordinative function
• Superficial layer: Expression muscles, attached to skin/fascia

Fibre Organisation:

• Parallel fibers: Greater contractile force
• Fusiform: Balanced strength and motion range
• Complex arrangements: Coordinated multi-directional movement

MRI Imaging Principles

T1-Weighted Images:

• Muscle: Intermediate signal (medium grey)
• Fat: High signal (bright white)
• Bone: Low signal (dark)
• Best for: Anatomical detail, tissue identification

FLAIR Sequences:

• Suppresses CSF signal
• Enhanced muscle-to-fluid contrast
• Better for facial region imaging

Related Research & References

Associated Projects

• Cloud2 - 3D visualization and geometric analysis

  • https://github.com/heikostark/Cloud2
  • Used for visualising 3D facial anatomy

• imagexd - Core image processing framework

  • https://stark-jena.de/research-interests/software/imagexd/
  • All processing executed through this tool

• VHP-Classification - Connective tissue analysis

  • https://github.com/heikostark/VHP-classification
  • Similar multi-rater consensus methodology

• Gordon1966 - Biomechanical muscle modeling

  • https://github.com/heikostark/Gordon1966
  • Integration with FEBio for muscle simulation

Research References

Facial Anatomy:

• Anatomical classification systems
• 3D reconstruction techniques
• Biomechanical properties of facial muscles

Inter-Rater Agreement:

• Consensus methodology
• Reliability statistics (Dice coefficient, Jaccard index)
• Agreement assessment metrics

MRI Segmentation:

• Anatomical accuracy in medical imaging
• Multi-rater segmentation protocols
• Quality assessment methods

Research Website

For more information on 3D facial anatomy research:

• https://stark-jena.de/current-responsibilities/3d-face/
• Research focus and publications
• Collaborators and partnerships

Muscle Groups Defined

Classification System

Class 0 - Unknown/Unclassified

• Tissue not clearly assigned to a muscle group
• Requires further classification
• Often represents partial volume effects

Class 1 - Mimic Muscles

• Muscles of facial expression
• Primarily attached to skin
• Innervated by CN VII (facial nerve)
• Include: orbicularis oculi/oris, buccinator, corrugators

Class 2 - Platysma

• Large sheet muscle of lower face and neck
• Extends from mandible to clavicle/shoulder
• Important for facial expression and neck movement
• Often analysed separately due to unique anatomy

Class 3 - Masticatory Muscles

• Muscles specialised for jaw movement and biting
• Include: masseter, temporalis, medial/lateral pterygoid
• Innervated by CN V (trigeminal nerve)
• Critical for feeding and speech

Class 4 - Eye Muscles

• Muscles controlling eye and eyelid movement
• Include: orbicularis oculi, levator palpebrae
• Essential for vision protection and expression
• Involved in blinking and eye closure

Class 5 - Tongue/Pharyngeal Muscles

• Muscles of tongue and pharynx
• Critical for speech and swallowing
• Mix of intrinsic and extrinsic muscles
• Innervated by CN XII (hypoglossal) and others

Performance Considerations

Computational Requirements

Typical System Specifications:

• CPU: Modern multi-core processor (8+ cores recommended)
• RAM: 32 GB minimum, 64+ GB recommended
• Storage: 500 GB - 2 TB depending on dataset size
• GPU: Optional (accelerated processing with CUDA support)

Processing Times (Approximate per subject):

• Coordinate alignment: 1-2 minutes
• Consensus generation: 1-2 minutes (per operation)
• Thickness calculation: 5-10 minutes
• Distance calculation: 5-10 minutes
• Statistical analysis: 1-5 minutes

Total Pipeline: ~30-60 minutes per subject (depending on volume)

Memory Management

Memory-Intensive Operations:

• 3D distance transform
• Thickness measurement with large scan radii
• Multi-subject batch processing

Optimisation Tips:

1. Process one subject at a time
2. Adjust thread count based on available RAM
3. Use disk caching for intermediate results
4. Monitor memory usage during processing

Parallelization

Thread Settings:

thread.max := 40
// Optimal: 0.8 × number of available cores
// Example: 40 cores → thread.max := 32

Batch Parallelisation:

• Run multiple subjects simultaneously on separate cores
• Use GNU Parallel or similar tools
• Monitor total system memory

Data Quality Assurance

Rater Agreement Analysis

Metrics Computed:

• Dice Coefficient: Overlap between raters
• Jaccard Index: Union-to-intersection ratio
• Hausdorff Distance: Maximum distance between segmentations
• Volume agreement: Voxel count consistency

Interpretation:

• Dice ≥ 0.85: Good agreement
• Dice 0.70-0.85: Acceptable agreement
• Dice < 0.70: Poor agreement (review needed)

Quality Control Steps

Before Processing:

1. Verify file formats and naming conventions
2. Check coordinate system alignment
3. Validate landmark detection
4. Inspect individual rater segmentations visually

During Processing:

1. Monitor transformation accuracy
2. Verify consensus threshold appropriateness
3. Check output file completeness
4. Validate statistical results

After Processing:

1. Compare results to known anatomy
2. Check for outliers in statistical analysis
3. Validate thickness/distance measurements
4. Visual inspection of results

Troubleshooting

Common Issues

Problem: "File not found" errors

• Solution: Verify file naming follows conventions (mask_*.nii.gz)
• Check directory structure matches expected format
• Verify all required files are present

Problem: Coordinate alignment failures

• Solution: Check marker file format (.mrk.json)
• Verify landmark names match script expectations
• Validate marker file locations

Problem: Consensus threshold not met

• Solution: Ensure sufficient raters (minimum 2 for best.macro)
• Check rater segmentation quality
• Review rater agreement statistics

Problem: Out of memory errors

• Solution: Reduce thread count
• Process fewer subjects simultaneously
• Increase available system RAM
• Use 64-bit imagexd version

Problem: Slow processing

• Solution: Increase thread count (if resources available)
• Verify disk I/O is not bottlenecked
• Check for competing processes
• Consider SSD for data storage

Problem: Inconsistent results across runs

• Solution: Verify identical input data
• Check for floating-point precision issues
• Validate environmental variables
• Review imagexd version consistency

Contributing

Contributions are welcome! Areas for enhancement:

Potential Contributions

• Additional anatomical classification schemes
• Improved inter-rater agreement algorithms
• Performance optimisations
• Extended statistical analysis methods
• Validation against manual measurements
• Documentation improvements
• Example datasets

How to Contribute

1. Fork the repository
2. Create a feature branch (git checkout -b feature/improvement)
3. Make improvements with clear documentation
4. Test thoroughly with sample data
5. Submit a pull request with a detailed description

Citation

If you use this work in your research, please cite:

@repository{3d-face-classification,
  title={3D-FACE-Classification: Advanced Analysis and Classification 
         of Facial and Mimic Muscle Anatomy using 3D MRI Data},
  author={Stark, Heiko},
  year={2026},
  url={https://github.com/heikostark/3D-FACE-classification}
}

Associated Research Topics:

• Facial anatomy analysis
• 3D medical image segmentation
• Multi-rater consensus methods
• Muscle classification systems
• Clinical imaging applications

License

This project is licensed under the GNU General Public License v3.0 (GPL-3.0).

You are free to:

• ✅ Use the software for any purpose
• ✅ Study how it works
• ✅ Modify and improve it
• ✅ Share it with others

With the requirement to:

• ℹ️ Provide attribution to the original author
• ℹ️ Disclose source modifications
• ℹ️ Use the same license for derivative works

See LICENSE file for full terms and conditions.

Author and Attribution

Heiko Stark
Institute of Zoology and Evolutionary Research
University of Jena

Research Interests:

• Medical image analysis
• 3D anatomical reconstruction
• Facial anatomy classification
• Multi-rater consensus methods
• Biomechanical analysis
• Computational geometry

Website: https://stark-jena.de/
Research Page: https://stark-jena.de/current-responsibilities/3d-face/

Acknowledgments

• imagexd Development Team: For the powerful image processing framework
• Anatomical Expert Raters: For careful manual segmentations
• MRI Technicians: For high-quality image acquisition
• Research Collaborators: For anatomical expertise and validation
• Free and Open Source Community: For foundational tools

Support and Feedback

For questions, bug reports, or feature suggestions:

1. GitHub Issues: https://github.com/heikostark/3D-FACE-classification/issues
2. Email: Via GitHub profile
3. Research Discussion: https://stark-jena.de/

Project Updates and Status

• Last Updated: May 2026
• Repository Status: Active
• License: GPL-3.0
• Maintained By: Heiko Stark
• Last Modification: November 2025

---

Quick Reference

File Naming Conventions

Filename Pattern	Description	Origin
mask_XX.nii.gz	2+ rater consensus	best.macro
mask_X3.nii.gz	3+ rater consensus	best_x3.macro
mask_[ID].nii.gz	Individual rater	Manual segmentation
mask_sub.nii.gz	Anatomical divisions	Reference file
*thickness.nii.gz	Thickness maps	thickness.macro
*distance.nii.gz	Distance maps	distance.macro
*.mrk.json	Anatomical landmarks	Marker files
*Segmentation-Bones.nii.gz	Bone segmentation	Reference anatomy

Script Execution Order

1. mrt_orientation.inc - Coordinate system validation
2. repair_transformation.macro (if needed) - Fix coordinate systems
3. best.macro - Generate consensus
4. best_x3.macro (optional) - High-confidence consensus
5. rater.macro - Individual rater statistics
6. thickness.macro - Thickness calculation
7. distance.macro - Distance calculation
8. Statistical analysis scripts in any order:
   • stat.macro
   • stat_*_sub.macro family

Output Files Generated

Results produced:
- mask_XX.nii.gz (consensus segmentation)
- mask_X3.nii.gz (high-confidence consensus)
- *_thickness.nii.gz (thickness maps)
- *_distance.nii.gz (distance maps)
- stat.csv (muscle group counts)
- stat_rater.csv (rater agreement)
- stat_*_sub.csv (regional statistics)

---

For complete documentation and tutorials, visit:
https://stark-jena.de/current-responsibilities/3d-face/