n.wikipedia
rmal areas ing a large urs, to comp m is compos the nose an ratory tract.
th, and pass
human resp m Wikipedia a.org/wiki/R
are then re amount of plete the ide sed with no and mouth,
. The air c ses through
iratory syst a, the free en Respiratory_
cognized th series is a entification.
ose, mouth, and termi coming from h the pharyn
em.
ncyclopedia _system.
hrough the s difficult an
throat, airw inates at th
m external nx into the t
a,
scan series, nd huge wo
way and lun he alveoli.
environme trachea. Tra
, slice ork. It
ngs as It is ent is achea
and b
bronchi are chi at the ca l of the seco lobe of eac her into dist ner than bef ms hollow ca Human air bifurcates a rachea is di ches into tw ree. After b eoli are reco It is an easy dle of throat m mouth and ough the se mes difficu mes smalle irway lumen meter of airw easing with
Moreover, ed. And the
s across w
the compon arina, which ond thoracic ch lung. The al part of th fore. They avities for g rway is a tr at the carina
ivided into t wo or more b
branching ab ognized as th y task to seg t and it is a d nose to ma
egmentation ult after seve er and small
n in differe way lumen.
each branc with curren resolution within slices
nents of airw h is a cartila c vertebra. T e bronchiole he lobes. Af end up at t as exchangi ree-like tunn a which is r
two parts. T bronchioles bout 17 tim he leaf node gment and r rod shaped ain left and n of trachea
eral bifurcat ler in the di ent branchin
And from t hing level.
nt CT scan of images s increases
way. The tr aginous ridg The main br es are subdi fter division
the alveoli.
ing with the nel that air recognized Those two a s, and each b mes or less,
e of the tree recognize th
hollow tub right bronch a can be de
tions. It is b istal bronch ng order. In
the table, th
nners, the qu becomes h
from 80 t
rachea then ge at the low
ronchi bran ivided abou ns, the bron . Alveoli ar e blood [38]
passes thro as the root are called m branch poin there are al e [8].
he trachea. T e as shown hi.
ealt easily, t because that us. In figur n table 2.2,
he airway lu
uantity of s higher than to a few h
separated in wer end of t nch into bron
t more than nchioles bec re anatomic ].
ough. It sta node. After main bronchi nt is recogni lveoli where
The trachea in figure 2.
the segmen t the diamet e 2.8, it sho it shows th umen diame
slices that l ever before undreds. A
nto left and the trachea nchioles, on n 20 times an come shorte cal structure
arts from tra r the bifurc i. Each bron ized as a no e gas excha
a is located .7 which co
ntation of ai ter of bronc ows the diam
he decreme eter is obvi
lung occupi e. The regi According t
d right onnect
irway chiole meter enting iously
ied is on of o the
(a)
(c)
adva lung quan
F
(c)
ancement of occupied.
ntity of slice There is no
Figure 2.8 D
) Inferior lo
f technolog An autom es, and to re o a fixed o
Diameter of (a) Trache obar bronchu
gy, there is atic segmen place this la optimal thre
(b)
(d)
f airway lum ea, (b) Main
us, (d) Ante
a significan ntation way abor-intensi eshold valu
men in the s n bronchus,
eromedial b
nt differenc y is in nee ive and time e, because
same patient
asal bronch
ce in how m ed to deal
e-consumin the intensit
nt.
hus.
many slices with such ng work.
ty of the ai s that
large
irway
Bifur
rcation er
meter
en is a little solute value nsity of wa nuation of
predetermin Thus, there osed in an racy change els. Nowada
h draft of th Manually shold differs
Recent app r is 2D and Three-dime
Trachea M B
0
1.98 cm
bit higher t e. The inten ater is defin
X-ray. Thu nable becau e are sever n interactiv ed, but it sti ays, basicall he airway. T
selecting s from each proaches can
3D hybrid a ensional typ
Table 2.2
Main Bronchus
1
1.78 cm
than air inte sity of air is ill requires ly, is to appl The selected
one is a people.
n be classif approach.
pe is the ap
Diameter o
Inferior lobar bronchus
2
1.59 cm
ensity. The r s defined to HU. Other i ntioned bef rtifacts ment determine [4]. Shorte an operator ly a manual d threshold straightforw
fied into two
pproaches th
of airway lum
s
Anter basal
3
0.87 c
range of lum o -1,000 Hou
ntensity of fore, the o tioned in se the thresho ening the o r to deal wit lly selected value is us ward appro
o major typ
hose segmen men.
romedial l bronchus
cm
men intensi unsfield uni f tissue is a optimal thr ection 2.1.
old. One of operation ti th the select fixed thresh ually aroun oach, becau
pes: one is 3
nting the ai
Pulmon alveolus
Distal terminat 200 to
m [46]
ity over air its (HU), an according t reshold val
f the metho ime withou
tion of cand hold to acqu nd -1,000 to
use the op
3D approach
irway tree w nary
didate uire a o -700 ptimal
h; the
with a
CT im
mage series wing algorith
ts for a 3D -segmented ay wall is, because of es and artifa Hybrid typ mentation. I m perform 2D
0, 34, 59].
mentation to r bronchi [3 The segme el belongs to ause of the t water or oth hea is a ring
rence betw s is distribu ify a fixed s. The airw nguish.
Although m cing manua
s combined hms [34]. M
seeded reg d, that is ca sometimes, f the limite acts such as pe is the ap
t can be co D filters to It can als o identify th
3].
entation som o airway or
trachea, it i her tissues g of airway ween the tra uted from
-threshold; h way and the
machine lea al operations
into a stack Many of them
gion growin alled the le , possible b d resolution
partial volu pproaches c
ombined in recover air so perform he main bron
metimes nee body while s not difficu
with highe wall filled w achea and s
-1,000 to -6 however, th e lung have
arning techn s, but the ac
k of volume m employ a ng procedur eakage into being broke
n of the CT ume affect [ combining 2 n different o rway walls,
in the rev nchi and th
eds a predef e doing regi ult to be seg er density th with air wit urrounding 600 HU. T here is no a e similar int
niques are u ccuracy of d
e. They are a classifier re [58]. Thi lung paren en in the low
T scanner, b [20].
2D analysis orders of th
and then pe verse order hen 2D techn
fined thresh ion growing gmented. It han air or th the inten tissue is s Therefore, it a fixed valu tensity, so t
useful in so detecting th
usually var or manually s technique nchyma, or wer order b but also the
s or recons hese compo erform 3D s r that mean niques reco
hold to dete g or segmen is surround greater than sity around
ignificant.
t seems not e exists tha that they ar
ome researc he airway co
ried of 3D r y determine e often resu r explosion bronchi. It i
e interferen
struction an onents. Som
segmentatio ns operatin onstructing l
ermine whet ntation proc
ded with mu n -100 HU d -1,000 HU
The intensi t too difficu at suitable f
re ambiguo
ch areas wit onnectivity
egion
good
d enough us ches.
The most shold canno
on to perso umed by th en.
An adaptiv lem [4]. Th they found ess adapted edure that id
The leakag matically in a
consuming In this work This algori er the resolu nsity each v have high nsity parts li igher than u low resolut The entire actual comp
sing a fuzz
time-consu ot be retrieve
on, and inf he processin
ve way was hey tried to d most suit
d to the en dentifying o ge is also ca
a short perio g algorithm
k, experime ithm is not ution is, the voxels are. M
intensity p ike the air i usual while tion due to a images will position. T
zy logic app
uming proc ed directly b fluenced by ng of airway
s by propos find the op ted threshol ntire lung.
outside the a alled as an e
od. Though just like us ents showed applicable fewer voxe More speci parts such a
in the alveo taking CT aliasing and l looks mild he average
proach [57]
cedure is to by any prio y artifacts y segmentat
sed Kiraly ptimal thres ld value ta It is cause airway.
explosion, b h it actually sing a seque d that it perf
while proc els within e
fically, vox as bones, v oli and airw scanning a d partial vol d and smoot e intensity
]. It segmen
o find the or knowledg mentioned tion depend
et al. that t shold by us akes place ed by the l
because the catches the ential search formed as sl cessing with ach slice w xels in the l vessels and way lumen. T
and samplin lume effect, th without d
becomes h
nts limited
best fitted ge because i in sections ds on how th
tried to solv sing increm
before the leakage of
identified v best one as h algorithm low as it sea h low resolu ould be, and ung and air
airway wa The low int g objects on , and vice ve detail inform igher than
order of ai
d threshold.
it may vary s 2.1. The he threshold
lve the thre mental thresh
region gro region gro
volume incr s expected, m or a brute
arched.
ution scans d also the h rway region all, but also tensity part n the fixed versa.
mation inste normal or
eshold holds, owing owing
reased it is a force
s. The higher n, not
resol
lution ones nsity in the b
mooth the chus. Ther eover, the d nomena me refore, the th
t to distal br Recently, F ons with an
ied by calcu mean value
sified. It trie osion of lea mean value over-estima inating con f. It often fa r words, the non-air obje able one, and In summar s but work sholds being abijanska p
. However, bronchi is m
entire scan e are not e distribution entioned ab hresholds ar ronchi.
Fabijanska n initial thr ulating the m
and plus an ed to choos aking into th of these vo ated step o nditions to t failed while
e higher re ects would d leads the s ry, the meth
ks slow w g chosen. It proposed w
, the seed much highe n because enough gri of the vess ove, takes re difficult t
proposed a reshold [3].
mean value n interval. Id
se a better he lung enti oxels may n f choosing erminate th dealing wi solution the
be. So the segmentatio hod of Kiral ith low re t increases 1
orks well w
point withi er than the s there are v ids to repre sels and bro place in e to discover
a method t . It updates e of the clas
deally, it con threshold q irely. One re not converge g a thresho
he procedur ith the scan e scans are
threshold i on toward fa ly et al. pro
solution on 1 HU in eac with high r
in the trach seed point.
vessels and esent the im onchi is not
each region while doin
that takes t s the thresh
sified regio nverges wh quickly, but eason of the e due to the old [33]. M res beside th ns those hav , the lower is easily ov failure.
oposed work nes, becaus ch step. On resolution s
hea has low The aliasing d micro ve
mage with uniform di n with diff g region gro
the usage o hold which on. Having t ile all airwa t sometimes e failure of e incomplet Moreover, it he converge ving thin sl
difference er-updated
ks fine with se of the s the other h scans, but o
w intensity g will take essels near
low resolu istributed. T ferent influ owing from
of the clas h is going threshold ar ay candidate s failed wit
this algorit te airway w t does not ence of upd lice thickne
between th exceed the
h high resol slow grow hand, the m often fails
. The
sified to be dating ss. In he air most
lution wth of
ethod while
deali
ing low reso The propo hods. It is a hods mentio y are all in mplish the wever, post-p
is to modi robustness.
.2 Lung
Image regi loped progr eir rigid cha Though hu brospinal flu
cranium is traints the s, the shape age or cong Rigid regis mal brain [1 stration, sca
lt [37].
Taking loca
olution ones osed method
a trade-off oned above n an iterati e full segm
processing fy the regio
Registra
istration on ressively in aracteristic.
uman brain uid, luckily s solid tiss brain not t s of brains genital defec stration show
7]. It is a aling and sh
al deformat s.
d in chapte between sp perform on ive fashion mentation,
is substitut on growing
ation
n medical a n recent year
is consists y, it is surrou
sues those o move eas are quite th ct. The boun ws good en
compositio hearing, wh
ion into con
er 3 aims peed, accur nly once re n. They all e.g. anato table for ea g part only,
application rs. Human b
of soft tissu unded by th
are hard a sily, but al he same wit
ndary of bra nough result on of transl
hich is call
nsideration,
to combine racy and su egion growi
l require ex mically an ach other al trying to e
, especially brains can b
ues, e.g. gre he cranium,
and difficul so prevents thout much ain is circula
ts registerin ations and led affine r
non-rigid r
e both adv uccessfulne ing with the
xtraordinary nalysis and
lgorithms. H ease the com
y for brain be registere
ey matter, w and is show lt to be be s the brain
distorted, e ar and smoo ng pairs of h rotations. C registration,
registrations
vantages of ess. None o e entire vo y procedur d reconstru Hence, the mputational
n data, has ed easily be
white matte wn in figur ent. It not
being dam except accid
oth.
human brain Combining , shows a b
s are propos f both
dental
n and rigid better
sed to
deal regis into i
Elast defor
Ther With the lu
apply
with the stration whi individual f
There are tic-Body-Sp rmed variat
Lung is a re have no m hout the con
ungs in ever To get rid ying existed
deformable ch is a kern finite grids.
still othe pline regist tions.
soft and el many bones nstraint of b
rybody.
of the dif d methods o
Figure
ability. Th nel-based re
It registers er kernel-b tration [39]
astic organ s beside rib bones, the v
fficulties of of brain-subj
e 2.9 A brai
The major n egistration [
by moving based regis
. They are
n filling wit bs surroundi volume and
f local defo bjected regis
in slice.
non-rigid r [11]. It put t
each grid p strations li e proposed
th air. It is ing the lung d surface ha
ormations, r stration to lu
registration the source a points.
ike Thin-P with adva
much hollo g and show ave huge dif
researchers ung registra
is the B-s and target i
Plate-Spline anced usage
ower than b wn in figure
fferences am
at first att ations. Rigid
spline image
and es on
brain.
2.10.
mong
tempt d way
tried way
Cao resul
towa also propo resul
to overcom shows a sig Making ass et al. prop lt [27].
As mention ard inter-sub the charact osed a non-lt [21, 37].
me this issue gnificant im
sumptions a osed an ap
ned in sectio bjects ones.
teristics amo -linear land
Figure
e but the res mprovement and constrai proach tryi
on 1.2, a ga . Not only ong people dmark-based
2.10 A thor
sult is still u on lung reg ints may im ing to retain
ap does exis the position must be ta d approach
rax slice.
unable to pr gistration [5 mprove the p
n the pulm
st in extendi n and statu aken into co
trying to co
ogress. Afte 6].
performanc monary volu
ing intra-sub us has to be
onsideration onquer this
er that, non
ce of registr ume with a
bjects opera e taken care n. Ehrhardt issue with
-rigid
ation.
good
ations e, but et al.
a fair
Ch
are fo 3.1. J stage
hapte
The propos four stages o Just like an e are quite th