Deformable Registration for Longitudinal Breast MRI Screening.

Area of Science:

• Medical Imaging
• Radiology
• Biomedical Engineering

Background:

• Breast MRI offers high sensitivity for cancer screening but suffers from high recall rates and unnecessary biopsies.
• Comparing current and prior MRI scans can reduce follow-up procedures but is challenging due to breast tissue deformation.
• Automated image registration is crucial for accurate side-by-side comparison and computer-aided diagnosis (CAD) systems.

Purpose of the Study:

• To propose a framework for optimizing deformable image registration parameters for 3D T1-weighted breast MRI.
• To identify the optimal parameter set for breast MRI registration using the Elastix tool.
• To evaluate the accuracy of the optimized registration in clinical patient data.

Main Methods:

• A finite element (FE) modeling approach was used to simulate breast tissue deformation and generate ground truth data.
• An exhaustive search evaluated 4050 parameter combinations across 7 registration parameters in Elastix.
• The optimal parameter set was validated on 13 patients' consecutive breast MRI scans using radiologist-identified landmarks.

Main Results:

• Registration accuracy varied significantly across parameter sets, with median errors from 0.29 mm to 2.50 mm.
• The best parameter sets achieved median registration errors as low as 0.29 mm and 95th percentile errors around 3.71 mm.
• Clinical validation demonstrated a registration accuracy of 3.4 [1.8–6.8] mm using the optimized parameter set.

Conclusions:

• A systematic framework effectively optimizes deformable registration parameters for breast MRI.
• The identified optimal parameter set significantly improves registration accuracy for longitudinal breast MRI analysis.
• Accurate registration facilitates more reliable comparison of breast MRI scans, potentially improving diagnostic efficiency and reducing unnecessary procedures.