Pulmonary nodule registration: rigid or nonrigid?

Area of Science:

• Medical Imaging
• Radiology
• Computer-Aided Diagnosis

Background:

• Accurate registration of pulmonary nodules on serial chest CT scans is crucial for monitoring disease progression and treatment response.
• Existing rigid and nonrigid registration techniques have varying performance characteristics in terms of accuracy and computational efficiency.

Purpose of the Study:

• To compare the performance of a novel rigid registration algorithm against a traditional iterative closest point (ICP) rigid algorithm and a nonrigid Demons algorithm for pulmonary nodule registration.
• To evaluate the accuracy and computational efficiency of these different registration schemes.

Main Methods:

• A gradient descent-based rigid registration algorithm with separate handling of geometric transformations (translation, rescaling, shearing, rotation) was developed.
• The developed rigid algorithm, ICP, and Demons nonrigid algorithm were tested on 60 clinical lung CT datasets with follow-up scans.
• Nodule registration accuracy was assessed using 3D Euclidean distance, and computational cost was measured in seconds.

Main Results:

• The developed rigid affine approach achieved a nodule registration error of 9.6 mm in 5 seconds.
• The ICP rigid algorithm had an error of 9.8 mm and took 300 seconds.
• The nonrigid Demons algorithm showed an error of 10.0 mm and required 55 seconds.

Conclusions:

• The developed rigid registration method demonstrates high efficiency and sufficient accuracy for pulmonary nodule registration in longitudinal studies.
• Rigid registration is a practical and preferred choice for clinical applications due to its speed and adequate precision.