Evaluation of a deformable registration algorithm for subsequent lung computed tomography imaging during...

Area of Science:

• Medical imaging analysis
• Radiotherapy research
• Computational anatomy

Background:

• Accurate lung segmentation and deformable image registration (DIR) are crucial for analyzing changes in lung computed tomography (CT) images during cancer treatment.
• Evaluating the performance of these algorithms and the consistency of different assessment techniques is essential for clinical application.

Purpose of the Study:

• To evaluate a lung segmentation algorithm and a DIR algorithm for lung CT images.
• To compare different evaluation techniques for these algorithms.
• To investigate the correlation between evaluation methods and their clinical utility.

Main Methods:

• Acquired 69 subsequent CT images from 15 lung cancer patients undergoing radiochemotherapy.
• Automated lung segmentations were compared to manual contours.
• Performed DIR using a specialized fast algorithm, requiring lung segmentation as input.
• Evaluated DIR using landmark distances, lung contour metrics, and vector field inconsistencies.

Main Results:

• Automated lung segmentation required manual correction in 66% of cases, though it assisted delineation.
• Landmark-based DIR evaluation showed high accuracy (2.9 mm error).
• Contour metrics and vector field inconsistencies (0.9 mm median) indicated satisfactory DIR performance.
• No clear correlation was found between the three DIR evaluation methods.

Conclusions:

• Automatic lung segmentation is challenging but aids manual processes.
• The evaluated DIR algorithm provides reliable results for longitudinal lung CT data.
• Clinical use of DIR requires fast evaluation, potentially combining automated methods to identify unacceptable results.