Computer Aided Detection of Bone Metastases in the Thoracolumbar Spine - Spinal Imaging and Image Analysis - page 99

Spinal canal

(b)

(a)

0.8 0.4

Spinal canal

0.8 0.9

0.4

(c)

0.7 0.9

(e)

(d)

Fig. 3 Spinal canal extraction. a CT image; b watershed result; c spinal canal candidates;
d directed acyclic graph (DAG), number on the edge is the weight between two nodes; and
e extracted spinal canal (blue color)

db ðÞ ¼ average I x ð ðÞ Þ; x 2 b ð 1 Þ

here I(x) is the image intensity of a pixel x inside the basin b. Given a and b are two
neighboring basins, they will be merged if both conditions in Eq. 2 are satis

ed

j db ðÞ da ðÞ j \ d ð c i Þ da ðÞþd m ; 8 c i 2 Na ðÞ; c i 6 ¼ b

ð 2 Þ

j db ðÞ da ðÞ j \ d ð c i Þ db ðÞþd m ; 8 c i 2 Nb ðÞ; c i 6 ¼ a

δ m is the merging threshold. After that, all
basins that meet the criteria in Eq. 3 and surrounded by bone pixels are recorded as
potential candidates for the spinal canal.

here N(a) denotes neighbors of basin a,

d ð c i Þ d ð b Þ [ d d ; 8 c i 2 N ð b Þ ð 3 Þ

here δ d is the depth contrast threshold. Figure 3 b, c show the result of the watershed

algorithm and the candidates for spinal canals.
As showed in Fig. 3 c, multiple canal candidates may exist in one slice due to the
partial volume effect or loss density vertebra body region such as lytic bone lesion.
We propose a method to extract the correct spinal canal using directed graph search.
We