Deformable 3D-2D registration for CT and its application to low dose tomographic fluoroscopy

Area of Science:

• Medical Imaging
• Image Registration
• Computational Imaging

Background:

• Standard deformable volume-to-volume (3D-3D) registration methods in medical imaging require high-quality images, often achieved with full data sampling.
• Deteriorated image quality due to reduced projection angles (e.g., in low-dose CT or tomographic fluoroscopy) leads to unacceptable 3D-3D registration outcomes.
• A registration approach robust to sparse projection data is needed for accurate medical image matching.

Purpose of the Study:

• To develop and validate a novel deformable volume-to-rawdata (3D-2D) registration method.
• To improve the robustness and accuracy of image registration when dealing with limited projection data.
• To enhance the performance of low-dose tomographic fluoroscopy through improved registration.

Main Methods:

• Proposed a deformable volume-to-rawdata (3D-2D) registration method optimizing alignment between a CT volume and acquired rawdata.
• Employed a cost function incorporating rawdata fidelity (sum of squared differences) and fluid-based diffusion regularization.
• Utilized alternating optimization: conjugate gradient descent for matching and Gaussian kernel convolution for regularization.

Main Results:

• The proposed 3D-2D registration method demonstrated high correlation with ground truth target positions, even with as few as 4-60 projections.
• Achieved improved matching in the rawdata domain compared to traditional 3D-3D registration methods.
• Results showed no introduced artifacts and greater stability with sparse rawdata than volume-to-volume approaches.

Conclusions:

• The deformable volume-to-rawdata (3D-2D) registration method offers superior robustness for sparse rawdata scenarios in medical imaging.
• This approach enhances the reliability of image registration in low-dose imaging applications like tomographic fluoroscopy.
• Improved registration can lead to better temporal resolution and overall robustness in low-dose tomographic fluoroscopy.