Medical image alignment based on landmark- and approximate contour-matching

Area of Science:

• Medical imaging
• Image registration
• Computational anatomy

Background:

• Accurate medical image registration is crucial for diagnosis and treatment planning.
• Traditional methods like iterative closest point (ICP) and thin plate spline (TPS) interpolation struggle with sparse landmark data.
• Enhancing image similarity and preserving local structures are key challenges in medical image registration.

Purpose of the Study:

• To introduce a landmark- and contour-matching (LCM) registration method.
• To improve image similarity in pairs with sparse landmark information by combining landmark data with approximate point correspondences.
• To extend the LCM method for both 2D and 3D medical image registration.

Main Methods:

• Developed a model for registering 2D medical images using landmark and contour-approximating landmarks.
• Extended the model for 3D cardiac image registration.
• Validated the LCM method on 2D hand X-rays and 3D porcine cardiac MRI, using Dice similarity coefficient, target registration error, and interior angle for evaluation.

Main Results:

• Reduced target registration error from 27.12 mm to 0.01 mm.
• Achieved an average 112% improvement in image similarity (Dice coefficients).
• Preserved local curvature at major landmarks and reduced image deformities, as indicated by interior angle measurements.

Conclusions:

• The LCM method overcomes limitations of purely landmark-based techniques.
• It provides accurate registration results, robust to landmark localization errors.
• The method enhances image similarity and preserves anatomical integrity in medical image registration.