Impact of deformable image registration accuracy on thoracic images with different regularization weight parameter...

Area of Science:

• Medical Imaging
• Radiology
• Computational Anatomy

Background:

• Deformable image registration (DIR) is crucial for analyzing anatomical changes in medical imaging.
• Optimizing DIR parameters, such as regularization weights, is essential for accurate thoracic image analysis.

Purpose of the Study:

• To evaluate the impact of varying regularization weights on the accuracy of thoracic image registration using commercial DIR software.
• To assess the stability and performance of the ANATOMY CONstrained Deformation Algorithm (ANACONDA) in thoracic DIR.

Main Methods:

• Utilized 4-dimensional (4D) CT scans from 10 patients, with 300 manually identified anatomic landmarks.
• Applied ANACONDA software to deform peak-inhale to peak-exhale thoracic images across a range of regularization weights (4 to 40,000).
• Calculated registration error (RE) by comparing software-calculated displacements with landmark-defined displacements; measured computation time.

Main Results:

• Higher regularization weights significantly reduced registration error (RE) in thoracic images.
• RE decreased sharply with increasing regularization weight before plateauing.
• No significant difference in RE was observed between regularization weights of 400 and other tested values (4, 40, 4000, 40,000).
• Average computation time ranged from 8.4 to 12.2 seconds.

Conclusions:

• The default regularization weight setting for ANACONDA demonstrates stability for thoracic image registration.
• The findings suggest that ANACONDA provides reliable DIR accuracy in the thoracic region across a wide range of regularization weights.
• Further optimization of regularization weights may not be critical for ANACONDA in this anatomical area.