Model-Based Registration of X-ray
Mammograms and MR Images of the
Female Breast
N.V. Ruiter, T.O. Müller, R. Stotzka, H. Gemmeke,
Forschungszentrum Karlsruhe, Germany
Motivation
Support multimodal
breast cancer
diagnosis:
MR image
• Locate lesion in
complementary
modality
X-ray mammogram
 Registration:
Find geometric correspondence
Registration Problem
Images not directly comparable !
• MR image:
• Volume image
• Undeformed breast
• Prone position
• X-ray mammography:
• 2D projection
• Large deformation
• Only one projection per deformation
 Add information using a
biomechanical model
Process of Registration
MR image
Biomechanical model
Deformed model
Deformation process
X-ray mammogram
Projection of artificial MR image
Problems of Deformation Model
• Large deformation of soft tissue
• Details of deformation process unknown:
•
•
•
•
Exact patient position
3D shape of breast
Thickness of deformed breast
Compression force
• Tissues of the breast:
• Only large scale structures
resolved in MR image
• Material models in literature:
• Inconclusive
• Not plausible for large deformation
L‘
Plate compression
Skin
Fatty
tissue
Glandular
tissue
Specification of Deformation Model
• Adapt global parameters:
• Projection angle by non-linear scaling
• Volume is preserved  estimate thickness
• Finite Element Model
• Large deformations (>>5%), (nearly) incompressible
materials
• Material model
• Evaluation of models for breast tissue:
 Neo Hookean model,
homogeneous tissues
• Two step modeling of deformation
1st step: Mammographic deformation
2nd step: Fine tuning using mammogram
Mammographic
deformation
Results with Clinical Datasets
• Six clinical data sets:
Lesion position in X-ray mammograms and MR image
known
• Smallest visible lesions in MR images: 5 mm
• Prediction of lesion position
• MRI  X-ray:
• Mean center distance: 4.3 mm (2.3 – 6 mm)
• Mean area overlap: 81%
• X-ray  MRI:
• Mean center distance: 3.9 mm (1.6 – 6.4 mm)
• Mean volume overlap: 91%
Results with Clinical Datasets
MRI X-ray : craniocaudal example
24mm
Direct MR projection
X-ray mammogram
 Center lesion distance
2.3 mm
 Lesion overlap 100%
MR projection after simulation
Conclusions
• Registration overcomes 3D deformation
• Successful first evaluation:
Localization with approx. 5 mm deviation
(smallest visible lesion)
 Clinical evaluation
• Possible applications:
• Support multimodal breast cancer diagnosis
(also alternative 3D modalities)
• Simulation of breast deformation
Thank you !

Talk_M6-6 - INFN - Torino Personal pages