Iterative inversion of deformation vector fields with feedback control.

Area of Science:

• Medical imaging
• Computational anatomy
• Image registration

Background:

• Accurate inverse deformation vector field (DVF) calculation is crucial for 4D imaging, radiation therapy, and image registration.
• Existing iterative algorithms for DVF inversion face challenges in accuracy and efficiency.

Purpose of the Study:

• To improve the accuracy and efficiency of iterative algorithms for DVF inversion.
• To advance the understanding of divergence and latency conditions in DVF inversion.
• To introduce adaptive feedback control mechanisms for DVF inversion.

Main Methods:

• Developed a framework of fixed-point iteration algorithms with active feedback control for DVF inversion.
• Designed adaptive control mechanisms modulating the inverse consistency (IC) residual for improved convergence.
• Introduced three spectral measures, including the nontranslational displacement component (NTDC), for DVF characterization.

Main Results:

• NTDC-adaptive iterations demonstrated a larger convergence region and faster convergence than nonadaptive methods.
• Spatially variant control achieved error reduction by at least an order of magnitude within 10 steps.
• Numerical experiments showed quantitative agreement with analysis-based predictions.

Conclusions:

• Adaptive control is essential and effective for DVF inversion, especially with significant nontranslational displacement components (NTDCs).
• The proposed adaptive iterations and spectral measures offer potential improvements for deformable image registration methods.
• The study provides new insights into iterative DVF inversion algorithms using residual feedback control.