A 3D α-hull approach to juxta-pleural nodule detection in chest Computed Tomography images

Introduction
Lung cancer is one of the main causes of death among both men and women
[1] [2], with about 28% and 19% of all cancer-related deaths in the United
States [3] and in European Union [4], respectively. The survival rate is
estimated to be between 10% and 15% after 5 years from diagnosis, with
an increase up to 50% if the cancer is detected in its early stage [5]. In this
scenario, early diagnosis plays an important role in increasing the survival
rate of people afiected by lung cancer.
ComputedTomography(CT)isconsideredthebestimagingmodalityforthe
detection of lung nodules, particularly after the introduction of the multi-
detector-row and helical CT technologies [6]. Therefore, screening programs
based on low-dose CT are regarded as a promising approach for detecting
early-stage lung cancers and reducing the number of lung cancer deaths.
A recent work [7] indeed demonstrates the real efiectiveness of screening
programs for lung cancer carried out with low-dose CT. The work shows a
reduction of 5-year mortality of more than 20% for subjects in the screening
program with low-dose CT.
The development of a Computer-Assisted Detection (CAD) system to
automatically identify lung nodules can enhance diagnosis accuracy in the
usual clinical practice providing valuable assistance to the radiologist by
giving a second opinion on a diagnosis. Along these lines, the MAGIC-5
1
Introduction 2
Collaboration
1
, funded by the Italian INFN (Istituto Nazionale di Fisica
Nucleare), adopted a development strategy that led to a complete CAD
algorithm for automated lung nodule identiflcation.
Lung cancer, most commonly, manifests itself as non-calcifled pulmonary
nodules. Among the difierent types of nodules, the juxta-pleural ones, which
originate in the pleura but expand into the lung parenchyma, due to their
position and their high density, are a "di–cult" type of lung nodule to be
detectedbyautomaticsystems. ThepurposeofimplementingaCADsystem
that deals exclusively with the issue of nodules in contact with the pleura
is therefore an attempt to develop a synergetic and parallel work to those
of other generic lung CADs of the Collaboration. The aim of this work is
thereforetocontributetothedevelopmentofaCADsystemfortheautomatic
detection of juxta-pleural nodules in CT scans.
The basic steps of the CAD system presented in this work are:
1. segmentationofthelungparenchyma,afterwhichjuxta-pleuralnodules
appear as concavities of the lung surface;
2. concavity closing in the binary segmented mask with the alpha-hull
concavitypatchingmethodinordertoincludeinthesegmentedvolume
those concavities that may be related to juxta-pleural nodules. By
applying the difierence operation between the closed and the original
mask, a list of nodule candidates is obtained;
3. feature (candidate characteristic parameters) extraction and
calculation for each nodule candidate;
4. nodule candidate classiflcation by using Artiflcial Neural Networks.
In this work, in order to overcome the limitations imposed by the
2D analysis previously performed by our research group, a 3D approach
1
Medical Applications on a Grid Infrastructure Connection, now M5L.
Introduction 3
is performed. The method proposed consists of stacking 2D single slice
binary masks whose concavities have been closed by the multiscale alpha-
hull algorithm in order to obtain a 3D reconstruction of the closed lung.
By applying a difierence operation with the segmented original volume, we
detect a list of concavities (either natural or due to nodules).
Thestudyoftheperformanceofthenodulehuntingbasedonalpha-hullfora
flxed range of fi values is also presented. This study leads to the assessment
of the optimum parameter optimizing detection sensitivity, which identifles
all juxta-pleural nodules in the CT data set.
In order to reduce the number of false positives (regions reported as healthy
byradiologistsandincorrectlyidentifledaspathologicalbytheCADsystem)
a research on the most efiective characteristics of the nodules is carried out.
Finally, the classiflcation step is performed, preceded by 3D features
extraction, analysis and flltering. The classifler developed in this work, is an
Artiflcial Neural Network. Training and test of the network are performed.
Byvaryingthenumberofneuronsinthehiddenlayer, thestructurewiththe
best performance is determined. The results obtained are then compared to
other CAD systems performance.
Chapter 1
Lung cancer and medical
imaging diagnostics
1.1 Introduction
This chapter introduces the important points and aspects of this thesis,
presenting the problem of lung nodules and its possible solutions, both in
terms of medical-diagnostics and mathematical-physical point of view. We
also discuss the need for a screening-program policy and the importance of
X-ray Computed Tomography as a tool in medical imaging diagnostics.
1.2 The neoplastic disease
A neoplasm is an abnormal mass growth of tissue as a result of neoplasia.
Neoplasia (neo is the greek for "new" and plas ‡a means "growth") is the
abnormal proliferation of cells. A tumor
1
can be a:
1
Tumor (Latin for swelling, one of the cardinal signs of inammation) originally meant
any form of swelling, neoplastic or not. Current English, however, both medical and
non-medical, uses tumor as a synonym of neoplasm.
4
1.2 The neoplastic disease 5
† Benign tumor
It is a type of tumor that doesn’t metastasize, that is, that doesn’t
afiect any part of non-adjacent organs. Basically, benign tumors are
typically surrounded by an outer surface (flbrous sheath) that inhibits
their ability to behave in an invasive manner, penetrating into nearby
healthycells. Cellgrowthisthenlimitedtothepathologicalareawhere
it is formed and therefore benign tumors are isolated. Indeed, many
kindsofbenign tumorsareconsideredtobe harmlesstohumanhealth.
† Potentially malignant tumor (pre-cancer)
It is also called carcinoma in situ and it is deflned as a proliferation
of atypical epithelial cells with various morphological and biological
characters of malignancy. Anyway, it is an early form of cancer that
doesn’t have inflltrating capacity, i.e. the ability to invade tissues
located beyond the basement membrane, and consequently blood
vessels or lymph vessels. Hence, the neoplastic cells proliferate in their
normal habitat or "in situ" (Latin for "in their place"). However, this
kindofneoplasmhasthepossibilitytobecomeprogressivelyaninvasive
carcinoma.
† Malignant tumor (cancer)
It is the most dangerous type of neoplasm. In this type of pathology,
there is an uncontrolled growth of cells that invade adjacent tissues.
Moreover, most of malignancy tumors invade surrounding tissues and
spreadinthehostorganism, givingrisetoothermalignancies, separate
and apart from the primary tumor. These secondary malignancies are
called metastases.
Lung carcinoma is a diagnostic category that includes all kind of
neoplasms that originate from epithelial tissues that make up the bronchi
and lung parenchyma. This type of disease has gradually become more
1.2 The neoplastic disease 6
and more frequent over the years. Indeed, in the beginning of the last
century, malignant tumors of the lung accounted for only 2% of the tumors
observed in the course of autopsy. Nowadays, tumor is considered to be
one of the main causes of death in the industrialized Countries. The rate
of people dying every year for cancer is estimated to be ten millions around
the world. Moreover, the more the development of industrial-technological
nations increases, the more the rate of tumor incidence rises up. In this
scenario, lung cancer is the leading cause of cancer-related mortality [8].
There are many risk factors that contribute to the onset of lung cancer.
Among them, we may include genetic factors, exposure to environmental
pollution or to ionizing radiation, uranium, asbestos, and cigarette smoke,
the last one being considered the most important etiological factor (7 cases
out of 10). It has been shown that a 35-year-old man who smokes more
than 25 cigarettes per day, has a risk of 13% to die before being 75 years
old for lung cancer. Furthermore, this risk increases with the number of
cigarettes smoked, duration and age of onset of smoking, nicotine content,
and presence or absence of the cigarette fllter. The relative risk of smokers
with respect to non-smokers is 14, instead for heavy smokers it rises up to
20. Smoking cessation leads to a progressive reduction in the risk of lung
cancer but never to its complete cancellation. Figures 1.1 and 1.2 show the
rate of death related to certain cancers, difierentiated by sex and weighted
for age, in America from 1930 to 2006 [9].
As can easily be seen from the trend of the two graphs, the incidence
of lung and bronchus cancer in men has reached a peak during the 90’s
and now is slightly decreasing. In women, instead, it is on the rise because
the tendency of the woman to smoke is growing more and more, and it
has surpassed breast cancer as the leading cause of cancer-related mortality.
When the lung cancer manifests its flrst symptoms, most of the time it is
1.2 The neoplastic disease 7
Figure 1.1: Rate of death related to difierent types of cancers in men, from 1930
to 2006 in America [9].
alreadyinanadvancedstateinwhichasurgicaloperationisnomorepossible.
During this stage, the survival rate is estimated to be between 10% and 15%
after 5 years. However, these percentages are considerable increased if the
cancer is detected at its early stage. For lung cancer, in this case, a 5-year
survival exceeding 50% and in the very early cases up to 90% is attained [5].
This flnding highlights the importance of identifying the disease in an initial
state in order to reduce mortality. Therefore, early diagnosis and computed
tomographic screening policy can play a key role in this scenario.
1.3 Low-Dose Computed Tomographic Screening 8
Figure 1.2: Rate of death related to difierent types of cancers in women, from
1930 to 2006 in America [9].
1.3 Low-Dose Computed Tomographic
Screening
Screening,inmedicine,isastrategyusedinapopulationtodetectadiseasein
individualswithoutsignsorsymptomsofthatdisease. Unlikewhatgenerally
happens in medicine, screening tests are performed on persons without any
clinical sign of disease. The intention of screening is to identify disease in
a community early, thus enabling earlier intervention and management in
the hope to reduce mortality and sufiering from a disease. Several types
of screening exist: universal screening involves screening of all individuals
in a certain category (for example, all children of a certain age). Case
1.3 Low-Dose Computed Tomographic Screening 9
flnding involves screening a smaller group of people based on the presence
of risk factors (for example, because a family member has been diagnosed
with a hereditary disease). In the flfties they started the flrst attempts at
screening for lung cancer [10], but only in the late seventies they have made
four large randomized and controlled trials to demonstrate the validity of
screening for lung cancer. Unfortunately, such studies, based on chest X-ray,
did not get the expected response because results were quite disappointing,
showing the same level of mortality in people who had undergone screening
and those who had not. This conclusion led to the disappointing end of
screening for lung cancer. Later, in the nineties, these types of studies were
re-evaluated. Moreover, with the progress in the X-ray technology, chest
X-ray had been superseded by CT technique. In fact, experimental data
seems to demonstrate the concrete contribution of Computed Tomography
for early investigations, in particular for the detection of lung cancer of
limited size that do not produce metastases. So far, the conventional CT
technologyhaditsbiggestobstaclesbecauseofthelongruntimes, highcosts
andradiationdoses. Today,mostoftheseproblemswerelargelysolvedbythe
introduction of the helical multi-layer computed technology with low-dose of
radiation. Inthiscontext,periodicscreeningprogramsatthelevelofprimary
and secondary prevention
2
, early detection of cancer and urgent need for
development in technology could be of help. The mass screening programs,
aimedatthesystematiccontrolofasymptomaticpopulationgroups,selected
accordingtospeciflcepidemiologicalcriteria,couldresultinsigniflcantefiects
in terms of early diagnosis and treatment.
Among the recent approaches of screening, there is the randomized
controlled trial known as ITALUNG-CT [11]. The ITALUNG-CT project is
2
The term primary prevention refers to the removal of the causes that lead to the
development of a disease, the term secondary prevention indicates the development of
early diagnosis.
1.3 Low-Dose Computed Tomographic Screening 10
trying to demonstrate the efiectiveness of screening using chest low-dose CT
in reducing mortality from lung cancer in heavy smokers or those who have
quitsmoking. ThestudywascarriedoutattheAOUCareggi,Pisa,andASL
at Pistoia. It is coordinated by ISPO (Istituto per lo Studio e la Prevenzione
Oncologica) and belongs to an international collaboration that includes
American and European similar studies. In this scenario, the flrst work that
demonstrates the real efiectiveness of screening for lung cancer, carried out
with low-dose CT, was recently published [7]. From August 2002 through
April 2004, 53.454 persons at high risk for lung cancer at 33 U.S. medical
centers were enrolled. Participants were randomly assigned to undergo three
annual screenings with either low-dose CT (26.722 participants) or single-
view posteroanterior chest radiography (26.732). Data were collected on
cases of lung cancer and deaths from lung cancer that occurred through
December 31, 2009. The rate of adherence to screening was more than 90%.
The rate of positive screening tests was 24.2% with low-dose CT and 6.9%
with radiography over all three rounds. A total of 96.4% of the positive
screening results in the low-dose CT group and 94.5% in the radiography
group were false positive results. The incidence of lung cancer was 645 cases
per 100.000 person per year (1.060 cancers) in the low-dose CT group, as
compared with 572 cases per 100.000 person per year (941 cancers) in the
radiography group (rate ratio, 1.13; 95% confldence interval [CI], 1.03 to
1.23). There were 247 deaths from lung cancer per 100.000 person-years
in the low-dose CT group and 309 deaths per 100.000 person-years in the
radiography group, representing a relative reduction in mortality from lung
cancerwithlow-doseCTscreeningof20.0%(95%CI,6.8to26.7; P=0.004).
The rate of death from any cause was reduced in the low-dose CT group, as
compared with the radiography group, by 6.7% (95% CI, 1.2 to 13.6; P =
0.02). Therefore, screening with the use of low-dose CT reduces mortality
from lung cancer [7].
1.4 Computed Tomography 11
1.4 Computed Tomography
Among the difierent types of tests for the diagnosis of lung cancer, there
is certainly the chest radiography. It is still one of the flrst tests taking
place for the classiflcation of the disease. The test is short, low dose for
the patient and cheap. With this analysis we can identify lung lesions
but it is not possible to indicate their nature. However, as mentioned in
paragraph 1.3, mass screening programs based on periodic repetition of
standard chest radiography, with or without cytological analysys of sputum
specimens, have shown no reduction in lung-cancer mortality [12]. This test
isnotverysuitableforsmalllesionsbecauseitisinsu–cientlyclearforsmall-
size regions, especially if they are located in the "dark areas" (behind the
heart, behind the clavicle or near the mediastinum). Therefore, the advent
of CT in the nineties, and advances in multidetector computed tomography,
were undoubtedly an innovation in medical imaging: they have made high-
resolution volumetric imaging possible in a single breath hold at acceptable
levels of radiation exposure [13]. Furthermore, the tomographic technique
has many advantages over traditional 2D medical radiography, such as a
reduction of the cardiac and respiratory motion artifacts, the elimination of
the superimposition of images of structures outside the area of interest, an
improved contrast resolution that makes possible to distinguish difierences
between tissues that difier in physical density by less than 1% and, through
the use of iodinated contrast medium, a better accuracy in the evaluation of
lung parenchyma, hilar and mediastinal lymph nodes, vascular component
and of all other mediastinal structures. However CT is a technique in which
thepatientissubjectedtoasigniflcantdoseofionizingradiation. Itisobvious
thatCTexaminationsshouldbeperformedonlywhenstrictlynecessary,that
is,theradiationdosedeliveredtothepatientmustbejustifledbyanadequate
beneflt not only of diagnostic nature but also of therapeutic one. This is
1.4 Computed Tomography 12
the justiflcation principle: no unnecessary use of radiation is permitted,
which means that the advantages must outweigh the disadvantages. The
possibility of radiation-induced cancer, flrst described in 1898 soon after the
developmentofroentgenograms,makescomputedtomographyinadequatefor
mass screening programs if made out on a population that is mostly healthy
and for a long period of time. In this context, however, CT may become a
rather important tool if low doses are used for a randomized study focused
exclusively on high-cancer-risk persons.
An important issue within radiology today is indeed how to reduce the
radiation dose during CT examinations without compromising the image
quality. In general, higher radiation doses result in higher-resolution images,
while lower doses lead to increased image noise and unsharp images. Several
methodsthatcanreducetheexposuretoionizingradiationduringaCTscan
exist:
† newsoftwaretechnologycansigniflcantlyreducetherequiredradiation;
† individualize the examination and adjust the radiation dose to the
body type and body organ examined. Difierent body types and organs
require difierent amounts of radiation;
† prior to every CT examination, evaluate the appropriateness of the
exam whether it is motivated or if another type of examination is more
suitable. Higherresolutionisnotalwayssuitableforanygivenscenario,
such as detection of small pulmonary masses [14].
Low-dose helical computed tomography is considered, at present, as one
of the best imaging techniques for the detection of pulmonary nodules of
sizes even smaller than 5 mm [15], deflning optimal pleural proflles.
1.4.1 Principles of operation and evolution of CT 13
1.4.1 Principles of operation and evolution of CT
Tomography (whose name derives from the greek t omos which means "slice"
or "section" and graphein which means "to write") is used to obtain images
ofsections(slices) oftheobjectunderexamination. It isbased on the Radon
transform theory (see section 1.5), designed in 1917, although the flrst to
employ it in the medical fleld was the American physicist Allan Cormack
in 1963. Medical CT produces cross-sectional imaging data of internal
structures of the body based on their ability to block an X-ray beam. Single
or multiple X-ray tubes rotate around the patient with an opposed array
of detectors picking up the transmitted radiation. The digitised data are
thenusedtocalculatetheradiologicalabsorptioncharacteristicsofindividual
volume elements (voxels) of the body parts scanned. These can then be used
togenerateimageswitheachvoxeldisplayedasatwo-dimensionalpixel. The
usualimagesarecross-sectional(axial), butcanalsobereformattedinnewer
scanners to provide coronal, sagittal or three-dimensional images. The flrst
commercially viable CT scanner was invented by Sir Godfrey Hounsfleld in
Hayes,UnitedKingdom,atEMICentralResearchLaboratoriesusingX-rays.
Inthemid1960sHounsfleldwasworkingonthepatternrecognitionofletters
when he began to consider whether he could reconstruct a three-dimensional
representation of the contents of a box from a set of readings taken through
the box at randomly selected directions. He found that by considering the
three-dimensional object within the box as a series of slices, reconstruction
was easier than treating the content as a volume. He tested the theoretical
principal by working with a matrix of numbers whose values were all set to
zero but a square in the middle where values were set to 1000. He entered
thesedataintoacomputerprogrammetogetsimulatedabsorptionvaluesand
thenreconstructedthepictureusinganotherprogramme. Hounsfleldrecalled
his surprise at how accurate the result was [16]. He flnally conceived his idea
1.4.1 Principles of operation and evolution of CT 14
in 1967. The flrst EMI-Scanner was installed in Atkinson Morley Hospital in
Wimbledon,England,andtheflrstpatientbrain-scanwasdoneon1October
1971. It was publicly announced in 1972. The original 1971 prototype took
160parallelreadingsthrough180angles,each1
–
apart,witheachscantaking
a little over 5 minutes. The images from these scans took 2.5 hours to be
processedbyalgebraicreconstructiontechniquesonacomputer. Thescanner
had a single photomultiplier detector, and operated on the Translate/Rotate
principle [16].
The CT devices have undergone incessant process of evolution from
their initial marketing in the 70s of last century. However, the general
conflguration is unchanged. In fact, all CT devices consist of three main
parts:
Figure 1.3: A CT gantry and bed.
† the gantry (flg. 1.3), which contains the source of radiation (X-
ray tube), collimators, the cooling systems, and the detection system
(detectors);