Validation of three deformable image registration algorithms for the thorax.

Area of Science:

• Medical imaging
• Image processing
• Thoracic imaging

Background:

• Deformable image registration (DIR) is crucial for calculating lung ventilation using 4D CT scans.
• Evaluating DIR accuracy requires comparing expert landmark correspondences and image differences.

Purpose of the Study:

• To validate and compare the accuracy of three DIR algorithms: optical flow (OF), diffeomorphic morphons (DM), and diffeomorphic demons (DD).
• To assess DIR performance for correlating voxels between respiratory phases in 4D CT images of the thorax.

Main Methods:

• DIR algorithms (OF, DM, DD) were applied to 4D CT images across respiratory phases (end-expiration, end-inspiration, mid-expiration, mid-inspiration).
• Validation used a 4D thorax model with radiologist-delineated landmarks.
• Registration accuracy was quantified by target registration error (TRE) and image differences (HU).

Main Results:

• Average TRE for OF, DM, and DD was 1.6 ± 0.9 mm, 1.4 ± 0.6 mm, and 1.4 ± 0.7 mm, respectively, indicating errors within two voxels.
• All DIR methods significantly outperformed rigid registration (median TRE of 5.0 mm).
• Voxel differences were within 100 HU for 87-93% of voxels across the algorithms.

Conclusions:

• The three DIR methods (OF, DM, DD) achieve accurate registrations in the thorax region.
• Mean TRE below two voxels suggests comparable accuracy among the evaluated DIR algorithms for thoracic imaging.