Multiscale deformable registration for dual-energy x-ray imaging

Area of Science:

• Medical Imaging
• Radiology
• Image Processing

Background:

• Dual-energy (DE) imaging enhances lung nodule detection by reducing anatomical noise.
• Double-shot DE imaging offers superior performance but is susceptible to motion artifacts due to temporal separation of image acquisition.
• Motion artifacts degrade image quality and diagnostic performance in DE chest imaging.

Purpose of the Study:

• To develop and evaluate a deformable registration technique for aligning high-energy (HE) and low-energy (LE) images in DE chest imaging.
• To mitigate motion artifacts in DE images caused by the temporal gap between HE and LE data acquisition.
• To improve the diagnostic performance of DE chest imaging through enhanced image quality.

Main Methods:

• A multi-pass deformable registration algorithm was developed, utilizing mutual information optimization for large-scale motion and normalized cross-correlation for finer scale corrections.
• The algorithm aligns HE images to LE images prior to DE image decomposition.
• The technique was evaluated in 129 patients using an experimental DE imaging prototype.

Main Results:

• The registration algorithm demonstrated statistically significant improvements in image alignment compared to baseline and cardiac-gating methods.
• Modulation transfer function (MTF) analysis indicated enhanced DE image quality, including noise reduction and edge enhancement.
• The algorithm effectively reduced motion artifacts, particularly in the cardiac region.

Conclusions:

• The developed deformable registration technique significantly improves image alignment in DE chest imaging.
• This method enhances DE image quality, offering potential for improved diagnostic performance in detecting subtle lung nodules.
• The algorithm represents a valuable tool for overcoming motion-related limitations in DE imaging.