Table 1:

Step-by-step algorithm describing the procedure for segmenting OAR structures of the head and neck for an incoming target CT scan using the Structured Weighted Voting (SWV) and Global Weighted Voting (GWV) atlas fusion techniques using CT-radiodensity image similarity or MIND descriptor.

INPUT: Unseen target CT scan, Multi-atlas set with CT images and expert delineations of OARs

OUTPUT: Segmented CT scan with OAR segmentations and consensus map

Step 1: Deformably register all atlases to the target scan by using Plastimatch

Step 2: Propagate atlas labels to the target scan

Step 3: Calculate atlas similarity matrix using either CT-radiodensity or MIND descriptor image feature in Eq. 2

Step 4: Calculate weights and dynamically select top-ranked atlases using Eq. 3

IF fusion method == SWV THEN

Set τ = 0.5

ELSE IF fusion method == GWV THEN

Set τ = 0.8

Step 5: Fuse propagated atlas labels using Eq. 1 to produce the consensus map

Step 6: Extract segmentations using default consensus threshold = 33