Performance Evaluation of Deformable Image Registration Algorithms Using Computed Tomography of Multiple Lung...

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Image Analysis

Background:

• Deformable image registration (DIR) is crucial for assessing organ motion in 4D CT scans.
• Accurate DIR is essential for image-guided radiotherapy, especially for lung tumors.

Purpose of the Study:

• To quantitatively assess the performance of 10 DIR algorithms.
• To compare DIR algorithm accuracy using 4D CT images with implanted fiducial markers.
• To evaluate DIR for tracking lung tumors during respiratory motion.

Main Methods:

• Utilized 4D CT images from 5 patients with fiducial markers (FMs).
• Physicians established ground-truth positions for FMs and tumors.
• 10 DIR algorithms were applied to estimate FM and tumor positions.
• Target registration errors (TREs) were calculated and compared to ground-truth.

Main Results:

• Optical flow algorithms achieved TREs within ground-truth uncertainty for FMs (1.82 ± 1.05 to 1.98 ± 1.17 mm).
• Optical flow and demons algorithms showed TREs of 1.29 ± 1.21 to 1.78 ± 1.75 mm for tumor positions.
• Tumor volume tracking showed significant differences between deformed positions and mean differences (4.55 to 7.55 times higher).

Conclusions:

• The iterative optical flow method provided the best performance for FM tracking in 4D CT.
• Optical flow and demons algorithms demonstrated superior performance for tumor volume tracking.
• Quantitative evaluation confirmed the effectiveness of DIR algorithms in assessing organ deformations.