DSA image registration using non-uniform MRF model and pivotal control points

Area of Science:

• Medical Imaging
• Image Processing
• Computer Vision

Background:

• Motion artifacts in Digital Subtraction Angiography (DSA) degrade image quality.
• Conventional non-rigid registration for DSA is inefficient due to spatially non-uniform control point requirements.

Purpose of the Study:

• To propose a novel non-uniform Markov Random Field (MRF) model for efficient DSA registration.
• To introduce the concept of pivotal and non-pivotal control points to optimize registration.
• To reduce motion artifacts in DSA images.

Main Methods:

• A non-uniform MRF model for DSA registration was developed.
• Quad-trees were utilized to generate a non-uniform grid of control points.
• The concept of pivotal and non-pivotal control points was implemented for improved efficiency.

Main Results:

• The proposed MRF formulation produced a smooth displacement field, enhancing artifact reduction compared to independent control point registration.
• Pivotal control points improved computational performance without compromising artifact reduction.
• The approach was validated on clinical datasets with quantitative analysis and GPU performance assessment.

Conclusions:

• The novel non-uniform MRF with pivotal control points offers an efficient and effective solution for DSA registration.
• This method significantly reduces motion artifacts in DSA images.
• The approach demonstrates improved computational performance on GPUs.