UC4 — 3D Medical Image Registration & Segmentation

Objectives

• Understand why multi-modal MRI registration matters for brain tumor diagnosis

• Learn about the registration pipeline: N4 bias correction, HD-BET, ANTsPy

• Know the four MRI modalities and their clinical roles

• Understand the SRI-24 atlas alignment workflow

Repository: wp7-UC4-medical-image-registration Contributors: Saruar Alam (UiB)

The Problem

Brain tumor diagnosis and monitoring rely on multiple MRI modalities, each highlighting different tissue characteristics. Before these modalities can be analyzed together — to delineate tumor boundaries or estimate tumor volume — the images must be spatially aligned.

This registration step is critical for both clinical practice and automated segmentation research.

MRI Modalities

Each modality contributes a different clinical perspective:

Modality	Full Name	Clinical Role
T1	T1-weighted	Detailed anatomical structure
T1Gd	T1 with gadolinium contrast	Highlights active tumor tissue (enhancing regions)
T2	T2-weighted	Sensitive to fluids, revealing edema and infiltration
FLAIR	Fluid-Attenuated Inversion Recovery	Differentiates CSF from lesions, especially near ventricles

Registration Pipeline

The pipeline combines two established medical imaging tools:

        graph TD
    A[Raw MRI Scans<br>T1, T1Gd, T2, FLAIR] --> B[N4 Bias Correction<br>Remove intensity non-uniformities]
    B --> C[HD-BET Brain Extraction<br>AI-based skull stripping]
    C --> D[Rigid Registration<br>All modalities → T1Gd reference]
    D --> E[Atlas Registration<br>T1Gd → SRI-24 standard atlas]
    E --> F[Propagate Transforms<br>Apply to all modalities]
    F --> G[Aligned Multi-Modal Volumes<br>Ready for segmentation]
    

Step-by-step:

1. N4 Bias Correction — Removes intensity non-uniformities caused by magnetic field inhomogeneities

2. HD-BET Brain Extraction — AI-based skull stripping to isolate brain tissue

3. Rigid Registration — Aligns all modalities to the T1Gd reference using ANTsPy

4. Atlas Registration — Registers T1Gd to the SRI-24 standard brain atlas

5. Transform Propagation — Applies all computed transformations to the remaining modalities

Tools Used

Tool	Purpose
HD-BET	High-Definition Brain Extraction Tool — AI-based skull stripping
ANTsPy	Advanced Normalization Tools for Python — registration and transformation
nibabel	NIfTI image I/O
SRI-24 Atlas	Standard brain atlas for spatial normalization

Data: BraTS Dataset

The pipeline is designed for the Brain Tumor Segmentation (BraTS) challenge dataset:

• Multi-institutional MRI scans

• Expert-annotated tumor boundaries

• Multiple modalities per subject

Environment Setup

UC4 uses a Conda environment for reproducibility:

git clone https://github.com/NAICNO/wp7-UC4-medical-image-registration.git
cd 3D-medical-image-registration-segmentation
conda env create -f environment.yml
conda activate 3d-image-registration-segmentation

Key dependencies: antspyx, nibabel, numpy, scipy, hd-bet

Status

UC4 provides the core registration pipeline with a Conda environment, an orchestrator notebook with synthetic 3D data for demonstration, a full test suite, and CI/CD pipeline. UC4 is the only WP7 use case in healthcare, where reproducible computational pipelines are especially important for clinical research.

Keypoints

• Brain tumor diagnosis requires spatial alignment of multiple MRI modalities

• Four MRI modalities (T1, T1Gd, T2, FLAIR) each reveal different tissue characteristics

• HD-BET provides AI-based brain extraction; ANTsPy handles registration

• Pipeline: N4 bias correction → brain extraction → rigid alignment → atlas registration

• SRI-24 atlas enables standardized spatial normalization across subjects

• Only WP7 use case in the healthcare domain