Computer Aided Diagnosis System for Lung Nodule Detection using CT Scan

Image
Sandeep V
1
, Rajeswari
2

1
PG Student, AIT Bangalore, [email protected], 9841!8!1"
#
Pro$essor% &'(, )* (ept, AIT Bangalore, ra+es,[email protected]

Abstract
/ecent .ears, t-e image processing tec-ni0ues ,idel.
used in a several medical $ields $or image improvement
0ualit. in earlier detection and treatments stages,
,-ere t-e time is ver. important $actor to 1no, and
discover t-e a2normalit. o$ issues in targeted image,
especiall. in a various cancer tumours suc- are lung
cancer, 2reast cancer, and man. more. T-e Image
0ualit. and a accurac. is t-e core $actor $or researc-,
image 0ualit. as ,ell as improvements are depend on
t-e en-ancement o$ stage ,-ere t-e lo, pre3processing
tec-ni0ue is used a 2ased on t-e Ga2or $ilter ,it-3in a
Gaussian rule. 4ollo,ing segmentations purpose ,e
are using ,aters-ed algorit-m and principles, an
en-anced image region o$ t-e o2+ect o$ interest t-at
used as t-e 2asic $oundation are in$ormation o$ $eature
e5traction is o2tained 2. using G6*7 % S87. As
rel.ing on a general and $eatures, t-e normalit. a
comparison is made. In t-is pro+ect, t-e main detecting
$eatures $or accreted image comparison are pi5el
percentage and mas13la2elling to $ind out t-e lung
nodules using *T scan images.
Keywords Lung nodule detection, Image processing
techniques, Feature extraction or nodule,
!nhancement"techniques, #atershed"algorithm$
1. Introduction
Lung cancer is a disease o the a%normal cells that are
multipl&ing and growing in"to the tumour$ 'ancer cells
can %e a carried awa& rom lungs in %lood, or l&mph
luid that are surrounds the lung tissue$ L&mph lows
through a l&mphatic (essels, which are drain in"to
l&mph nodes that located in the lungs and in the centre
o a chest$ )he Lung cancer oten spreads in to toward
the centre o the chest, %ecause a natural low o l&mph
is out o the lungs is towards to the centre o the chest$
a *etastasis occurs when the cancer cell lea(es a site
where it+s %egan and mo(es in"to a l&mph node or
another part o the %od& through the a %lood stream ,1-$
. 'ancer that starts in the lungs is called primar&
cancer o lung$ )here are se(eral di t&pes o lung
cancers, and these di(ided into a two main group/ 1$
Small cell lung cancer and, 2$ non"small cell lung$
'ancer has three su%t&pes/ is 'arcinoma,
.denocarcinoma 0 Squamous cell carcinomas$
)he ran1 order o the cancers or the %oth males and
emales among 2ordanian in the &ear o 2334 indicates
that there were 567 cases o the lung cancer accounting
or a 89$9 :; o all diagnosed newl& cancer cases in the
2334$ )he Lung cancer is aected 274 815$5 :; male
and 64 82$<:; o emales, with a males to emale ratio
is 6/1 which Lungs cancer ran1ed second in the males
and 13th among in emales ,2-$ Fig"image 1 shows a
general description o lung cancer detections s&stems
that contain our %asic stages or steps$ )he irst stage
starts with a ta1ing the collections o ') images
8normal image and a%normal image; in the a(aila%le
=ata%ase rom the I*>. ?ome VI. 0 !L'.@ @u%lic
.ccess, ,5-$ )he second step applied to a se(eral
techniques o image enhancements, to get %est qualit&
o the image results and clearness$ )he third step
applied to the image segmentation algorithms which
pla& an eecti(e rule in image processing stage, and
the ourth step to o%tains the a general eatures rom
enhanced image and segmented image which gi(es
indicators o a normalit& or a%normalit& o image$

4igure1. 6ung cancer image processing steps
Image capture (CT image)
Image enhancement (using Gabor
filter)
Image segmentation (using
watershed algorithm)
Nodule detection (using SVM and
3)
!eatures e"traction (using G#CM)
Lung cancer is most dangerous widespread cancer in
the world, according to stages o disco(er& o cancer
cells in the lungsA the process o earl& detections o the
disease pla&s a (er& quic1 and important role and
essential role to a(oid a serious ad(ance stages to
reduce its percentage o distri%utions$
)he aim o this project was a detect eature or an
accurate o images comparison and pixels percentage
o mas1"la%elling and earl& detection o lung nodule
and lung cancer$
. !aterial " !et#ods
In this project, to o%tain more and more accurate results
we are di(ided our wor1 into the ollowing three steps/
1$ )he Image !nhancement stage/ to ma1e an image
%etter enhance it rom a noising, corruption and
intererence$ )he ollowing three methods and steps
used or this purpose onl&/ Ba%or ilter it 8has %est
results;, the .uto enhancement" algorithm is another
technique, and one more method FF) Fast Fourier
)ransorm it 8shows the worst result or image
segmentations;$
2$ )he Image Segmentation stage/ to di(ides and
segment the enhanced image, used algorithms on the
RCI o the images 8just two lungs, the methods used
are/ threshold approach and its *ar1er"'ontrolled,
#atershed Segmentation 8this has a %etter approach
and %etter results than threshold;$
5$ Features !xtraction step/ to get the general eatures
o enhanced %& Ba%or and watershed image using
BL'* and SV*$
)o ind out the lung nodule siDe, shape, area o the
nodule$ #hether it is cancers lung nodule are not all
this things we are inding in this eature extraction
.1 Image $n#ancement
)he image @re"processing is a stage starts with the
image enhancementA the main aim o image
enhancement is to impro(e a interpreta%ilit& or
perception o a inormation included in t image or
humans (iewers, or to pro(ide a %etter qualit& input to
other automated image"processing 8I@; techniques$
Image enhancement techniques as di(ided into a two
%road categories/ 1$Spatial domain$ 2$ requenc&
methods$ .nd unortunatel&, there is no other general
theor& or a determining what is a EgoodF image
enhancement is when it will come to a human
perception$ I it loo1s good to human e&e, it is good$
?ow"e(er, when the image enhancement technique, are
used pre"processing tools or other image" processing
technique, the quantitati(e measures can determined
which techniques are the most appropriate ,<-$ In an
image enhancement stages we are used the ollowing 2
technique/ or Ba%or ilter, 1$.uto"enhancement and 2$
Fast Fourier transorms techniques$
.1.1 %abor &ilter
Image presentation is %ased on the Ba%or unction
constitutes an excellent local multi"scale
decompositions in the terms o logons that
simultaneousl& 8and optimall&; localiDations in space
and requenc& domains ,6-$
)he Ba%or ilter is a linear ilter and whose impulse
response is as deined %& harmonic unction multiplied
%& a Baussian unction$ )he multiplications 0
con(olution"propert& and 8'on(olution"theorems;, the
Fourier transorm o Ba%or ilterGs impulse and its
response o the con(olutions o a Fourier"transorm
8F); and the harmonic unction$ and Fourier transorm
o the Baussian"unctions$ Figure 2 shows 8a; the
original image ') image 8%; the enhanced Ba%or Filter
image$

9a: 'riginal Image 92: )n-anced 2. Ga2or
4igure #. T-e result o$ appl.ing Ga2or en-ancement
tec-ni0ue
)he .uto enhancement technique is strongl& depends
on the su%jecti(e and o%ser(ation and a statistical
operation such as, 1$mean and 2$(ariance calculation$
)he enhancement percentage in this project was equal
to the 54$326:$
. Image Segmentation
)he Image segmentation is an essential process or the
most image anal&sis and su%sequent tas1s$ In the
particular man&"existing techniques or image
description and recognition and depends highl& on the
segmentation results ,9-$ Segmentation or di(ides
image into its constituent regions and o%jects$
Segmentation o the medical images in the orm o 2=,
slice %& slice has a man& useul applications or
medical proessionals such as/ (isualiDations 0 (olume
estimations o o%jects o interested area, detection o
a%normalities 8e$g$ tumour+s, pol&ps, etc$;, the tissue
quantiication and the& classiication, and man& more
,4-$ )he goal and aim o segmentation is to simpli& or
change the representation o image into something that
in to the more meaningul and easier anal&ses$ )he
Image segmentation is t&picall& and mainl& we used to
locate o%jects and %oundaries o selected image
8cur(es, lines, etc$; with the images$ *ore precise,
segmentations o image is the process o the assigning
the la%el to the each and e(er& pixel wit in the image
such pixels with the same la%el shares certain (isuals
characteristic ,7-$ Result o segmentation image is the
set o segments that are collecti(el& co(ers the entries
image, or the set o contours rom extracted the image
8detection edge;$ C .ll pixels in a gi(en regions are
similar to respect to some characteristics or computed
propert&, such the colour, intensit&+s, or texture o$
.djacent o regions to dierent with respect to the
characteristic8s; same$
Segmentation algorithms %ased on one o two %asic
properties o intensit& o (alues/ discontinuit& and the
similarit&$ )he irst 81; categor& is to partition o the
image %ased on the a%rupt changes in the intensit&,
such as edges in image edges o the image$ )he second
82; categories is %ased on the partitioning o the image
into its regions that are similar to according to a
predeined criterions$ )he ?istogram threshold
approach alls to under this categor&$
..1 'aters#ed Segmentation Approac#
*ar1er"dri(en watershed segmentation technique
extracts a seeds that indicate a presence o o%jects or a
%ac1ground at image speciic locations$ *ar1er
locations are set to %e regional minima within
topological surace 8t&picall&, gradient o original IHp
image;, the watershed algorithm is applied to ,11-$
Separating the touching o%jects in an image is one o
most diicult image processing operation, where a
watershed transorm is oten to appl& such pro%lem$
*ar1er controlled watershed segmented approach has 2
t&pes/ 1$!xternal associated with the %ac1ground and 2$
Internal associated with an o%ject o interest$
9a: )n-anced image 2. Ga2or 92: Segmented image 2.
;aters-ed
4igure <. =ormal )n-anced image 2. Ga2or $ilter and
its Segmentation using 7ar1er3*ontrolled ;aters-ed
approac-
)he Image Segmentation using a watershed transorm
wor1s well i we can identi& or mar1 %ac1ground
locations and oreground o%jects to ind a Ecatchment
%asinsF and a Ewatershed ridge linesF with in the image
%& treating it as a surace image where the light pixels
are high and the dar1 pixels are the low shows the
segmented image %& a watershed$
.( &eatures $)traction
Image eatures !xtraction stage is an (er& important
stage that uses techniques and algorithms to detect and
isolate the (arious desired portions shapes 8eatures; o
a gi(en images$ )o predict the pro%a%ilit& o lung
nodule cancer presence, the ollowing two methods are
used/ 8BL'*; 0 8SV*; and mas1ing, %oth methods
are %ased the acts that strongl& related to the lung
anatom& and it inormation o lung ') imaging$
.(.1 Nodule Detection
Lung nodule detection is a (er& diicult step within the
'.= s&stems de(elopment$ .ctuall&, the ') lung
images, the nodules are requentl& attached to the %lood
(essel or to the pleura, urther their gre& tone is so it is
similar to (essels section that traditional intensit&+s"
%ased methods are inappropriate$ Instead,
)he !ecti(e nodule detection algorithm must ta1e
%oth the 1$gre& le(el and 2$the o%ject shape into the
account$ In our s&stem '.= we adopt a method that
uses 5= shapes inormation to identi& o spherical
regions within a gi(en gre& le(el$ .nd In this we are
use a Baussian 5= unction as smoothing ilter$
(. Conclusions
)he image impro(ement techniques are de(eloping or
earlier disease detections and the treatment stages, the
time actor is ta1en in account to disco(er the
a%normalit& issues in the target images$ Image accurac&
and qualit& is a core actors o this research r project,
the qualit& o image assessment as well as the
enhancement stages where were adapted to low pre"
processing technique %ased on the Ba%or ilter with in
the Baussian rules and unction$ )he proposed
technique and method is eicient or image
segmentation and principles to %e a region o the
interest oundations or eature extraction and to
o%taining$ )he proposed technique gi(es a (er&
promising result+s comparing with other used
techniques or old techniques$ Rel&ing on the general
eatures, the normalit& comparison is made on$ )he
main detected eatures or the accurate images
comparison on pixels percentage and mas1 with high
accurac& and ro%ust operations$in the end o results we
are inding in this project alse positi(e lung nodule
cancer $our research till now we are completed all the
enhancement and segmentation, urther we are inding
area o the nodule and siDe o the nodule and what
stage o the cancer$
*. +eferences
,1- Ion"Small 'ell Lung 'ancer, .(aila%le at/
http/HHwww$1atemacint&reoundation$orgH pdHnon"small"
cell$pd, .dapted rom Iational 'ancer Institute 8I'I; and
@atients Li(ing with 'ancer 8@L#';, 2339, 8accessed 2ul&
2311;$
,2- )arawneh *$, Iimri C$, .rqou% J$, Kaghal *$, 'ancer
Incidence in 2ordan 2334, .(aila%le at/
http/HHwww$moh$go($joH*C?HFilesH@u%licationH2ordan
:23'ancer:23Registr&L2334:23ReportL1$pd, 2334
,5- Lung 'ancer =ata%ase, .(aila%le at/
https/HHeddie$(ia$cornell$eduHcgi"%inHdatacHsignon$cgi,
8accessed 2ul& 2311;$
,<- BonDaleD R$'$, #oods R$!$, =igital Image @rocessing,
Mpper Saddle Ri(er, I2 @rentice ?all, 2334$
,6- 'risto%al B$, Ia(arro$ R$, Space and requenc& (arient
image enhancement %ased in Ba%or representation, @attern
Recognition Letters, !lse(ier, 177<, 16, p$ 295"299$
,N- Jrishan .$, !(aluation o Ba%or ilter parameters or
image enhancement and segmentation, in !lectronic
Instrumentation and 'ontrol !ngineering, *aster$ @unja%/
)hapar Mni(ersit&, 2337
,9- Iunes O$=$C$, @PreD *$B$, *edical Image Segmentation
%& *ultile(el )hresholding >ased on ?istogram =ierence,
presented at 19th International 'onerence on S&stems,
Signals and Image @rocessing, 2313
,4- Ven1ateshwarlu J$, Image !nhancement using FuDD&
Inerence S&stem, in 'omputer Science 0 !ngineering,
*aster thesis, 2313
,7- Shapiro L$B$, Stoc1man B$'$, 'omputer Vision/ )heor&
and .pplications, @rentice ?all, 2331
,13- ?uang Q$, Bao #$, 'ai #$, )hresholding technique with
adapti(e window selection or une(en lighting image, @attern
Recognition Letters, !lse(ier, 233<,
,11- Le(ner I$, Khang ?$, 'lassiication"=ri(en #atershed
Segmentation, I!!! )ransactions on Image @rocessing, 2339
Aut#or ,iograp#ies
*r Sandeep V 8>$!$; has completed his !ngineering
rom Re(a Institute o )echnolog&, >angalore$
'urrentl& he is pursuing *")ech in =igital
'ommunication !ngineering rom .char&a Institute o
)echnolog&, >angalore$