Breast Tumor Detection and Classification Using Intravoxel...

Research ArticleBreast Tumor Detection and Classification Using IntravoxelIncoherent Motion Hyperspectral Imaging Techniques

Si-Wa Chan123 Yung-Chieh Chang4 Po-Wen Huang4 Yen-Chieh Ouyang 4

Yu-Tzu Chang4 Ruey-Feng Chang15 Jyh-Wen Chai6 Clayton Chi-Chang Chen6

Hsian-Min Chen7 Chein-I Chang89 and Chin-Yao Lin1011

1 Graduate Institute of Biomedical Electronics and Bioinformatics National Taiwan University Taipei Taiwan2 Department of Medical Imaging Taichung Tzu Chi Hospital Buddhist Tzu Chi Medical Foundation Taichung Taiwan3 Department of Radiology School of Medicine Tzu Chi University Hualien Taiwan4 Department of Electrical Engineering National Chung Hsing University Taichung Taiwan5 Graduate Institute of Networking and Multimedia National Taiwan University Taipei Taiwan6 Department of Radiology Taichung Veterans General Hospital Taichung Taiwan7 Center for Quantitative Imaging in Medicine Department of Medical Research Taichung Veterans General HospitalTaichung Taiwan

8 Remote Sensing Signal and Image Processing Laboratory Department of Computer Science and Electrical EngineeringUniversity of Maryland Baltimore County Baltimore MD USA

9 Department of Electrical Engineering National Taipei University of Technology Taipei Taiwan10Department of Breast Medical Centre Taichung Tzu Chi Hospital Buddhist Tzu Chi Medical Foundation Taichung Taiwan11Department of General Surgery Taichung Tzu Chi Hospital Buddhist Tzu Chi Medical Foundation Taichung Taiwan

Correspondence should be addressed to Yen-Chieh Ouyang ycouyangnchuedutw

Received 8 December 2018 Revised 18 June 2019 Accepted 7 July 2019 Published 28 July 2019

Academic Editor Nicola Galea

Copyright copy 2019 Si-Wa Chan et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Breast cancer is a main cause of disease and death for women globally Because of the limitations of traditional mammography andultrasonography magnetic resonance imaging (MRI) has gradually become an important radiological method for breast cancerassessment over the past decades MRI is free of the problems related to radiation exposure and provides excellent image resolutionand contrast However a disadvantage is the injection of contrast agent which is toxic for some patients (such as patients withchronic renal disease or pregnant and lactating women) Recent findings of gadolinium deposits in the brain are also a concern Toaddress these issues this paper develops an intravoxel incoherentmotion- (IVIM-)MRI-based histogram analysis approach whichtakes advantage of several hyperspectral techniques such as the band expansion process (BEP) to expand a multispectral imageto hyperspectral images and create an automatic target generation process (ATGP) After automatically finding suspected targetsfurther detection was attained by using kernel constrained energy minimization (KCEM) A decision tree and histogram analysiswere applied to classify breast tissue via quantitative analysis for detected lesions which were used to distinguish between threecategories of breast tissue malignant tumors (ie central and peripheral zone) cysts and normal breast tissues The experimentalresults demonstrated that the proposed IVIM-MRI-based histogram analysis approach can effectively differentiate between thesethree breast tissue types

1 Introduction

Breast cancer is the most prevalent cancer and a majorcause of death in women worldwide [1 2] Early diagnosis iscritical for increasing the survival rate among women with

breast cancer However current basic imaging approaches(eg mammography and ultrasonography [US]) have theirlimitations especially in patients with implants postoper-ative scarring and dense breast parenchyma These factorsoccasionally result in improper management of breast cancer

HindawiBioMed Research InternationalVolume 2019 Article ID 3843295 15 pageshttpsdoiorg10115520193843295

2 BioMed Research International

Central Area of the Tumor

Peripheral Area of the Tumor

Normal Breast Tissue

Figure 1 Illustration of the breast tissue regions

[3ndash7] Recent studies demonstrate that dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) tech-niques are the most promising methods for detecting tumorsand assessing therapeutic response [7ndash9] Furthermore theadditional diagnostic value of breast MRI is that it detectsmore multifocal multicentric and contralateral diseases[10 11] The sensitivity of DCE-MRI in the diagnosis ofbreast cancer is as high as 88-100 [12 13] However thespecificity of DCE-MRI varies from 50 to 97 [11 14ndash16] This variability may result from different lesion criteriaand tumor heterogeneity used among studies [15ndash17] Thusdifferentiating a benign lesion from a malignant lesion usingDCE-MRI remains a challenge To increase breast MRIspecificity one of the latest advancements inMRI technologythe apparent diffusion coefficient (ADC) which is obtainedfrom diffusion-weighted imaging (DWI) was developed todifferentiate between benign and malignant breast lesionswith 915 sensitivity and 865 specificity [18 19] A typicalillustration of the regions of the breast tissues is shown inFigure 1

DWI is a nocontrast-enhancedMRI technique It is usefulfor characterizing cancerous tissue heterogeneity by directlyreflecting the Brownian motion of water molecules in bodytissues particularly in breast cancer [19ndash21] It can obtainthe physiological characteristics of breast tissue with theassumption of the monoexponential decay to quantitate theanalysis of the water moleculesrsquo ADC [22] In malignanttumors the ADC value is usually lower than that of thenormal tissue because of the restriction of water diffusionresulting from increased cellularity of the tissue and reducedextracellular space in the tumor Based on the findings ofa meta-analysis [23] the specificity of DWI is higher thancontrast-enhanced MRI although there is a wide range inthreshold values from 090 mm2s to 160 x 10minus3 mm2s toaccount for the use of different b values The specificity ofDWI and DCE-MRI is 84 and 72 [23]

In vivo microscopic motion of water molecules detectedby DWI is influenced by the diffusion of water moleculesin the tissue structure and by the microcirculation of blood

in the capillary network [24] Signal attenuation on mono-exponential DWI therefore presents a linear relationshipbut does not truly reflect the microstructure changes inorganization The reason for this finding is that the ADCvalue is overestimated due to the fact that it contains twotypes of information (1) microcirculation perfusion of bloodin a capillary network and (2) water molecule diffusion ofthe tissue structure Intravoxel incoherent motion (IVIM)which was first described by Li Bihan et al in 1986 [25]separates microcirculation perfusion and water moleculediffusion by using a biexponential model [22 25 26] Usingthis biexponential model DWI is able to use signal intensityproduced by multiple b-values to analyze watertissue dif-fusitvity At low b-values the signal intensity reflects waterdiffusion in tissues as well as microcirculation in a capillarynetwork At high b-values the signal intensity reflects tissuediffusivity Through IVIM biomarkers have been resolvedas the perfusion fraction (PF) water molecule diffusivity(D) and pseudodiffusion (Dlowast) Recent studies demonstratethat the IVIM model is helpful for the differential diagnosisof breast lesions [27ndash31] In general a MR image can beviewed as a multispectral remote sensing image in whicheach spectral band image can be considered as an imageacquired by a particular pulse sequence [32] Hyperspectralimaging has become an emerging remote sensing technologythat improves traditional multispectral imaging by usinghundreds of contiguous spectral bands With such finespectral resolution many unknown subtle substances canbe revealed for subpixel detection and mixed pixel analysis[33] The application of hyperspectral imaging techniquesin MR brain imaging has also been investigated [33ndash35]Many medical imaging techniques are generally supervisedand require training samples provided a priori such as max-imum likelihood classification (MLC) K-nearest-neighbor(KNN) classification [36] neural networks [36] supportvector machine (SVM) [36] and Fisherrsquos linear discrimi-nant analysis (FLDA) [37] To avoid this issue the trainingsamples should be obtained directly from the data to beprocessed in an unsupervised method In this instance wetake advantage of the automatic target generation process(ATGP) developed by Ren and Chang [38] in conjunctionwith the spectral angle mapper (SAM) method which allowsusers to select training samples automatically This procedureis then followed by a supervised hyperspectral subpixel targetdetection technique called kernel-based constrained energyminimization (KCEM) developed by Jiao and Chang [39] todetect breast lesion tissues for quantitative analysis for IVIM-DWI This technique considers MR image slices acquired bypulse sequences and DWI with weighting factors of 12 b-values as a 12-band multispectral image cube in which eachsequence-acquired image is viewed as a spectral band imageThe data set to be used for experiments was collected from 25patients for IVIM imaging examinations

The purpose of this study was three folds (1) to measureIVIM parametric maps (ie tissue diffusivity [D] perfusionfraction [PF] pseudodiffusion coefficient [Dlowast] and slope)obtained from multiple b values of normal breast tissuecysts and malignant tumors (2) to determine whetherthese parameters can be used to distinguish between benign

BioMed Research International 3

and malignant breast lesions and (3) to calculate advanceddiffusion MRI metrics from histogram analysis (ie meanand median) and heterogeneity

2 Background

21 Diffusion-Weighted Imaging Diffusion-weighted imag-ing (DWI) is sensitive to the thermally driven randommotionof water molecules Molecular diffusion is the randommotion of molecules called Brownian motion [25 40] Themotion is modified by the local tissue environment and thecell membranes The motion of water molecules is morerestricted in tissues with a high cellular density than in tissueswith low cellularity Signal intensity in DWI is inverselyproportional to the degree of water molecule diffusion Thehistologic structure is influenced by diffusion The changeof water diffusion in tissues is useful for MR imaging andcan be used as a multifaceted tool to characterize tissuestructures and identify and differentiate disease processesFurthermore DWI can be used to calculate the ADC whichcan quantitatively measure the degree of tissue diffusivity

Spin-echo echo-planar DWI is the most popular clinicaltechnique to produce diffusion-weighted images in whichdiffusion is described by the following monoexponentialequation [25]

ln(1198781198871198780) = minus119887 (ADC) lArrrArr ADC = minus ln (1198781198871198780)119887 (1)

with

119887 = 120574211989221205752 (Δ minus 1205753) (2)

in which S119887 is DWI with diffusion attenuation at the b valueand 1198780 isDWIwithout diffusion attenuation119892 is the diffusiongradient 120574 is the proton gyromagnetic ratio and 120575 is thediffusion gradient duration Equation (2) presents b value(expressed in seconds per square millimeter) and representsthe strength of diffusion weighting in which S119887 is the signalintensity with the diffusion gradient S0 is the signal intensitywithout the diffusion gradient The monoexponential modelis traditionally used to calculate the decay of ADC in (2) andresults from two b values b1 and b2 as follows

D = minus 11198871 minus 1198872 ln(119878 (1198871)119878 (1198872)) (3)

22 Intravoxel Incoherent Motion Imaging Using the valueof ADC in (3) does not sufficiently describe tissue behaviorTo address this issue Le Bihan et al [25] proposed a novelapproach called IVIM MRI which demonstrated that theADC values of data samples obtained at low b values can beused to measure two IVIM parameters (1) pseudodiffusion(Dlowast) and (2) the perfusion fraction (PF) of pure moleculardiffusion and microcirculation or blood perfusion [25 26]In particular Dlowast and PF can be used as quantitative biomark-ers to describe changes in diffusivity and microcapillaryperfusion of tissues Increasing the b-value can change the

values of Dlowast and PF and therefore can reflect the diffusivityof different tissues and tissue microcapillary perfusion

DWI is influenced by the diffusion of water moleculeswhich is affected by microcapillary blood in the capillarynetwork in tissue structures Both processes cause phasedispersion in DWI and consequently signal attenuationLe Bihan et al [25] described the behavior of protonsthat displayed signal attenuation in DWI as IVIM imagingwhich does not represent a linear relationship as shown inFigure 2 The biexponential model involves curve fitting forIVIM imaging to separate the estimation of tissue diffusivityfrom perfusion by using multiple b values in the followingequation

1198781198871198780 = (1 minus PF) exp (minus119887D) + PF exp (minus119887Dlowast) (4)

where D is the pure diffusion coefficient S119887 is the signalintensity in the pixel with diffusion gradient and S0 is thesignal intensity in the pixel without a diffusion gradient Thisbiexponential analysis based on the IVIM sequence canbe used to describe the mixture of perfusion and diffusionresulting from DWI in three parameters pure diffusioncoefficient [D] perfusion fraction [PF] and pseudodiffusioncoefficient [Dlowast]23 Band Expansion Process Intravoxel incoherent imagesdo not provide sufficient spectral information to be pro-cessed as hyperspectral images therefore the band expan-sion process (BEP) developed by Ren and Chang [41] isused to resolve this issue for IVIM analysis This processproduces new images from the original IVIM images vianonlinear functions Its premise is similar to the momentgenerating function used by a random process that makesuse of the statistics of all orders to describe probabilisticbehaviors of a random process In the BEP only the statisticsof the second order of IVIM images autocorrelation andcross-correlation functions are used for this purpose Suchnonlinearly generated images provide nonlinear spectralinformation contained in IVIM images and can help improvedata analysis

The BEP presented in this section is a nonlinear processusing correlation functions to generate new band imagesfrom the original set of multispectral images

The BEP for IVIM images is as follows

Step 1 The 1st-order IVIM images are obtained as follows119861119897119871119897=1 = the set of original IVIM images

Step 2 The 2nd-order correlated IVIM images are obtained asfollows

(i) B2119897 119871119897=1= the set of autocorrelated IVIM images

(ii) B119896B119897119871119871119896=1119897=1119896=119897 = the set of cross-correlated IVIMimages

In case a rescaling is needed auto- or cross-correlatedIVIM images can be normalized by variances in the IVIMimages such as (1205902119861119897)minus11198612119897 and (120590119861119896120590119861119897)minus1119861119896119861119897


Pseudodiffusion (Dlowast)ADC(low) is perfusion sensitive but also affected by diffusion effects in tissues

Tissue Diffusion (D)D asymp ADC

105

1

095

09

085

08

075

100 200 300 400 500 600 700 800 900 1000

inde

x

b value (sＧＧ2)

Sb

S0

PF = (S0 minus Fit(0))S0

b valueADCIVIM

Figure 2 The graph of curve fitting of IVIM imaging used the biexponential model to separate the estimation of tissue diffusivity andperfusion based on multiple 119887 values

Step 3 The 3rd order correlated IVIM images are obtained asfollows

(i) B3119897 119871119897=1 = the set of autocorrelated IVIM images(ii) B2119896B119897119871119871119896=1119897=1119897=119896 = the set of two cross-correlated IVIM

images(iii) B119896B119897B119898119871119871119871119896=1119897=1119898=1119896=119897 =119898= the set of three cross-cor-

related IVIM images

As in Step 2 auto- or cross-correlated IVIM images canbe similar Therefore it may require normalization by thevariances in the IVIM images For example

(1205903B119897)minus1 B3119897 (1205902B119896120590B119897)minus1 B2119896B119897

(120590B119896120590B119897120590B119898)minus1 B119896B119897B119898 (5)

Step 4 Nonlinear correlated IVIM images are obtained asfollows

(i) radicB119897119871119897=1 = the set of IVIM images stretched out by thesquare root

(ii) logradicB119897119871119897=1 = the set of IVIM images stretched out bythe logarithmic function

Figure 3 shows an example of IVIM images (a) groundtruth (b) compared to the one in (c) without using BEP

24 Automatic Target Generation Process Once IVIM imagesare nonlinearly expanded they are then processed by anunsupervised target detection algorithm called automatictarget detection algorithm (ATGP) to detect potential

unknown targets without prior knowledge The procedure isbriefly described as follows

Let X be a data matrix formed by data samplevectorsr119894119873119894=1 (ie X = [r1r2 sdot sdot sdot r119873])The norm of thedata matrix X is defined by the equation as follows

X = max1le119894le119873

1003817100381710038171003817r1198941003817100381710038171003817 (6)

where r119894 is the length of the vector r119894 = (1199031198941 1199031198942 sdot sdot sdot 119903119894119871)119879as defined by r1198942 = sum119871119897=1 1199032119894119897 and assuming that 119894lowast =arg max1le119894le119873r119894 The norm of data matrix X in (6) canfurther be expressed by

X = 1003817100381710038171003817r119894lowast1003817100381710038171003817 (7)

which is exactly the brightest pixel r119894lowast and its normhas themaximum vector length The maximum l2 norm defined by(7) is indeed themaximum pixel vector length correspondingto the brightest pixel vector in the data set

Using (7) the ATGP produces a sequence of orthogonalsubspace projections (OSPs) as follows

119875perpU = I minus UU (8)

Thus if X is the original hyperspectral image cube the ATGPfirst selects an initial target pixel 1199051198601198791198661198750 that yields the normof the space X denoted by tATGP0 = X via (6) It thenprojects the space X into a subspace orthogonal to ⟨tATGP0 ⟩via 119875perpU0 with U0 = [tATGP0 ] The resulting subspace is denotedby X1 = ⟨tATGP0 ⟩perp The ATGP selects a first target pixel thatyields the norm of space X1 denoted by tATGP1 = X1 via(7) and then projects space X into a subspace orthogonal to⟨tATGP0 tATGP1 ⟩ via 119875perpU1 with U1 = [tATGP0 tATGP1 ] The resultingsubspace is denoted by X2 = ⟨tATGP0 tATGP1 ⟩perp The same


(a) (b) (c)

Figure 3 An example of using BEP to generate left breast malignant tissue in (a) ground truth (b) an image compared to image (c) and (c)not using BEP

procedure is repeated until the stopping rule is satisfied (iethe number of target pixels for the ATGP required to extract)The details of its algorithmic implementation are as follows

(1) Initial condition the variable 120576 is the prescribed errorthreshold and t0 is a pixel with the brightest intensity value(ie the maximal gray level value) Set k=0

(2) For 119896 larr997888 119896+1119875perpt0 is applied via (8) withU=[t0] to allimage pixels r in the image The 119896th target t119896 is generated atthe 119896th stage which has the maximum orthogonal projectionas follows

t119896 = arg maxr[(119875perp[U119896minus1t119896]r)119879 (119875perp[U119896minus1t119896]r)] (9)

If 119898(t119896minus1 t119896) gt 120576 in which 119898(sdot sdot) can be any target discrim-ination measure then Step 2 is repeated The algorithm isotherwise terminated At this point all generated target pixelst0 t1 sdot sdot sdot t119896minus1 are the desired targets

25 Spectral Angle Mapper The targets generated by ATGPare single targets To find other targets that have spectralsignatures similar to those of the ATGP-detected targets weused the SAM [42] as a spectral measure to calculate thesimilarity between two vectors by finding their angles [33]In particular s and twere two vectorsThe SAM is defined by

SAM (s t) = cosminus1 ( sTts12 t12) in radians (10)

In general it is used to compare a spectral vector against areference vector in terms of their between angle Thereforethe larger the angle is in (10) the less is the similarity

26 Kernel-Based Constrained Energy Minimization Oncetargets were determined and identified by using SAM atarget detector was further designed for target detectionIn this section a finite impulse response (FIR) linear filtercalled constrained energy minimization (CEM) developed inseveral studies [42ndash44] was used for this purpose

More specifically CEM is derived from the linearlyconstrained minimum variance originally proposed by Frostfor adaptive beam forming [45] If a hyperspectral image is

represented by a collection of image pixel vectors denotedby r1 r2 sdot sdot sdot r119873 in which r119894 = (1199031198941 1199031198942 sdot sdot sdot 119903119894119871)119879 for 1 le119894 le 119873 is an L-dimensional pixel vector then N is the totalnumber of pixels in the image and L is the total numberof spectral channels Furthermore d = (1198891 1198892 119889119871)119879isspecified by a desired signature of interest to be used for targetdetection The goal is to find a target detector that can detectdata samples specified by the desired target signal d via aFIR filter with L filter coefficients 1199081 1199082 sdot sdot sdot 119908119871 denotedby an L-dimensional vector w = (1199081 1199082 sdot sdot sdot 119908119871)119879 whichminimizes the filter output energy subject to the constraintd119879w = w119879d = 1 More specifically if y119894 denotes the output ofthe designed FIR filter resulting from the input ri then y119894 canbe expressed by the following equation

119910119894 = 119871sum119897=1

119908119897119903119894119897 = (w)119879 r119894 = r119879119894 w (11)

The average energy of the filter output is given by (12) inwhich (13) provides the autocorrelation sample matrix of theimage

( 1119873) 119873sum119894=1

1199102119894 = ( 1119873) 119873sum119894=1

(r119879119894 w)2

= w119879 [( 1119873) 119873sum119894=1

r119894r119879

119894 ]w = w119879Rw

(12)

R = ( 1119873) 119873sum119894=1

r119894r119879

119894 (13)

The goal was to solve the following linearly constrainedoptimization problem

minw w119879Rwsubject to d119879w = w119879d = 1 (14)

in which w119879RW is the variance resulting from signals notpassing through the filter The optimal solution to (14) asdescribed in [17ndash19] is presented in (15) and (16) as follows


wCEM = (d119879Rminus1d)minus1 Rminus1d (15)

min119908

wRminus1w = (wCEM)119879 Rminus1wCEM = 1d119879Rminus1d

(16)

which is the CEM error considered as the least energyresulting from unwanted signal sources impinging on anarray of sensors With the optimal weight wCEM specified by(15) a filter called CEM and denoted by 120575CEM(r) was derivedfrom a previous study [43] and can be specified by

120575CEM (r) = (wCEM)119879 r = (d119879Rminus1d)minus1 (Rminus1d)119879 r (17)

Constrained energy minimization is a linear filter and IVIMimages are nonlinearly generatedmultispectral images there-fore nonlinear separability may present an issue for CEMTo mitigate this problem CEM was further expanded to akernel version of CEM called kernel CEM (KCEM) whichwas derived as follows [39] Φ R119873 997888rarr 119865 by r 997891997888rarr Φ(r)TheR in (13) can be expanded in feature space F as presentedin

RΦ= 1119873

119873sum119894=1

Φ119879 (r119894)Φ (r119894) (18)

The CEM derived from (17) can be extended to a kernelversion of CEM (KCEM) given by

120575K-CEM (r) = Φ119879 (d)Rminus1Φ Φ (r)Φ119879 (d)Rminus1

ΦΦ (d) (19)

By virtue of (19) the KCEM can be calculated in featurespace F via the kernel trick without mapping the originaldata sample vectors into the feature space Several commonlyused kernel functions can be used for (19) In this paper theGaussian-based radial basis function (RBF) kernel is given by

119896 (x y) = exp(minus1003817100381710038171003817x minus y1003817100381710038171003817221205902 ) for 120590 isin R (20)

which was used for experiments to implement KCEMThereare two reasons for choosing Gaussian-based RBF kernel(1) it is a translation invariant kernel and (2) its associatednonlinear map is smooth Even when the spectral signaturesof a given hyperspectral data set are subject to irregularillumination the translationally invariant kernels typicallyprovide robust detection This finding is because it dependsonly on the difference between x and y not the absolutepositions of a single spectral vector The smooth nonlinearmapping associated with the Gaussian RBF kernel meansthat after nonlinear mapping the topographic ordering ofthe data in the input space is preserved in the feature spaceThe mapped data sample vectors in the feature space alsooccupy the small subspace of the feature space in which databelonging to different classes can be separated to a largerextent than in the input space As a result the Gaussian RBFkernel-mapped feature space in the target detection task ismore effective

2713resholding TheKCEM-generated detectionmap is realvalued therefore it requires a thresholding technique toproduce a binary image to show detected lesion tissues Inthe literature many thresholding techniques have been pro-posed such as Otsursquos method [46] two entropic thresholdingmethods local entropy and joint entropy methods [45] andthree relative entropy methods local relative entropy jointrelative entropy and global relative entropy methods in [4748] Based on our extensive experiments the local entropymethod [45] was the best among entropy-based methodsand outperformed the most widely used Otsursquos methodThe local entropy method was consequently selected as thethresholding technique to segment lesions from the KCEM-produced detection maps

28 Decision Tree A decision tree is a nonparametric super-vised learning method used for classification and regressionIt builds classification or regression models in the form ofa tree structure in which each internal node denotes a teston an attribute each branch represents an outcome of thetest and leaf nodes represent classes or class distributionsIt breaks a dataset into smaller and smaller subsets whileincrementally developing an associated decision tree at thesame time

Decision trees can be used for categorical and numericaldata The goal is to create a model that can be used to predictthe value of a target variable by learning simple decision rulesinferred from the data features

The lesions detected by the local entropy-thresholderKCEM were further used as the inputs of a decision treeIn order to perform breast lesion classification the decisiontree is specifically developed by a nonparametric supervisedlearning method which is generally used for classificationand regression It builds classification or regression modelsin the form of a tree structure in which each internal nodedenotes a test based on an attribute Each branch representsan outcome of the test and leaf nodes represent classes orclass distributions Categorical and numerical data can beused in such decision trees The goal was to create a modelthat could be used to predict the value of a target variable bylearning simple decision rules inferred from the data features

3 Materials and Methods

31 Patient Selection After obtaining institutional reviewboard approval to conduct this prospective study 25 patientswith breast cancer (20 malignant lesions and 5 cysts) under-went MR examination in our institution from May 2014to June 2015 Written informed consent was obtained fromall patients The patient inclusion criteria were as follows(1) newly diagnosed with breast cancer confirmed by nee-dle biopsy (2) the patient was not receiving neoadjuvantchemotherapy or hormonal treatment at the time of imaging(3) tumor size (solid portion referring to the T2-weightedand contrast-enhanced T1-weighted images)gt 1 cm in DWimaging (4) unilateral breast cancer (5) no previous breastsurgery history and (6) no motion artifact The definitivediagnosis of cyst was obtained by means of ultrasonographic(US) and breast MR imaging findings The simple cyst shows


thin wall circumscribed margin and anechoic at US lowT1 signal intensity high T2 signal intensity at fat-suppressedimages and no enhancement at postcontrast fat-suppressedimages All MRI studies were reviewed by an experiencedradiologist (SWC) who had access to all patient information

32 Experimental Materials MRIwas conducted using a 15Tsystem (Aera Siemens Erlangen Germany) using a bodycoil as the transmitter and a dedicated 16-channel receivercoil The patient was prone and head first Conventional T1-and T2-weighted images were acquired with and without fatsuppression The axial IVIM images were obtained by usingsingle-shot spin-echo echo-planar imaging (EPI) The axialIVIM images with bilateral breast coverage were acquired(TRTE 580068 ms FOV 320 mm2 x 264 mm2 matrix 132x 160 x 30 reconstructed voxel size 2 mm3times2 mm3times3 mm3with spectral presaturation inversion recovery and diffusionsensitization in the anterior-posterior direction applied withweighting factors of b = 0 15 30 45 60 100 250 400 550700 850 1000 secmm2 Total scan time required by an IVIMimaging scan was 63 minutes Axial T1-weighted DCE MRI(TRTE 4518 ms FA 12 FOV 320mm2 x 320mm2 matrixsize 512 x 512 slice thickness 15 mm) was acquired using thegradient echo sequence The contrast media were capturedafter precontrast and four consecutive time points (ie 60-second interval time) after administrating of gadolinium(Gadovist 10 mmolmL) by using a power injector at a flowrate of 20 mLs followed by 20 cc normal saline flush

Sixteen sets of MR images were used for the experimentswhich were acquired by spin-lattice relaxation time (T1)spin-spin relaxation time (T2) proton density (PD) and T1-DCE along with IVIM based DW images for which T1-DCEwas involved using contrast agents at different times TheseIVIM-DWI images were considered as the multispectralimages

Image analysis was conducted on a personal computerusing in-house software written in MATLAB (Math WorksNatick MA) The IVIM parameters were calculated fromthe DW images at all b-values by using the biexponentialmodel described by Le Bihan et al [25 26] At first theassumption parameters derived from the IVIM image usingall b-values were compared with DCE MRI to obtain reliableresults A radiologist with 13 years of experience read all MRIbreast images to determine the tumor region of interest (ROI)and exclude areas of necrosis bleeding and edema Afterparameters were mapped again onto the DCE MRI as theground truth There is no need of ROI in our study becauseentire images were processed pixel by pixel Thereforeinterobserver variability was limited

33 Preprocessing Before processing the data a techniquecalled nonparametric nonuniform intensity normalization(N3)was used to adjust the nonuniformity of theMR imagesMultiple b-values images of IVIM displacement caused byrespiration must be corrected therefore an alignment wasused to register all pixels Once all images were aligned anautomatic method to extract the breast region was developedas follows Automatic process to extract the breast region

Step 1 A high pass filter was used to enhance theboundary of the breast image

Step 2 A Sobel edge detector was used to extract thecontour of the breast image

Step 3 The hit-or-miss transformation function wasapplied to the flat area between two breasts to extractthe sternum

Step 4 Otsursquos method was implemented to produce abinary image

Step 5 A morphological opening using a structuralelement with a window (size 2x2 pixels) was usedto detach the connected components in the binaryimage

Step 6 A morphological opening was used to isolatethe largest connected component

Step 7 A Sobel edge detector was applied again toobtain the silhouette of the breast image

34 Breast Lesion Tissue Detection The BEP was used to cre-ate additional IVIM images for the IVIM-MRI classificationAfter BEP the original images were expanded to 90 bandimages Since KCEM requires the knowledge of the desiredsignature d therefore we first used ATGP to determine theinitial target pixels These pixels were further used to findtarget pixels with similar spectral signatures measured bySAM as the training samples that could be used as the desiredd for KCEM The detected region was further verified byan experienced breast radiologist and by tumor biopsy Thedetected part of each breast lesion tissue on an ADCmap wascalculated by all b values by using (3) For all lesion tissues theDDlowast PF and slope values were further calculated on a pixel-by-pixel basis by using (4) The lesion tissues of 25 individualpatients were tested and the statistical values of all parameterswere used as the standard for the decision tree

35 Quantitative Analysis This study was conducted onIVIM-DWI using 12 b values b=0 15 30 45 60 100 250400 550 700 850 1000 smm2 The biexponential model wasproposed by Le Bihan et al [25] microcapillary perfusion ofthe blood has no specific orientation which depends on thevelocity of flowing blood and the vascular architecture Theeffect of pseudodiffusion on signal attenuation in an IVIM-DWI voxel is also b value dependent The biexponentialanalysis to describe DWI comprises a mixture of perfusionand diffusion

The analytic model of IVIM is a biexponential andmonoexponential model demonstrated in (4) The monoex-ponential model of IVIM is expressed in (1) The ADC wasobtained by using all b-values (0-1000 smm2) and then fittedto the (4) Four parameters D (puremolecular diffusion) Dlowast(perfusion-related diffusion) PF (perfusion fraction) andADC (apparent diffusion coefficient) were determined afterundergoing processing by image preprocessing detectionand analysis of the lesion Through these parameters thelesion area could be further compared


lowast1 lowast1 lowastweight

OriginalIVIM Image

Mono-exponentialfitting

Bi-exponentialfitting

Signal intensityfitting

ADC Dlowast D PF Slope

Analyze breast tissuewith D Dlowast and PF

Decision tree of all parameters result

Non-Mass Normal Mass

Non-Tumor Tumor

Cyst

Index Difference

Figure 4 The flow chart shows the decision tree structure for breast tissue classification using ADC D Dlowast PF and slope parameters

36 Histogram Analysis In this section a quantitative anal-ysis was conducted on the IVIM-DWI images obtained byusing 12 diffusion-weighted b values 0 15 30 45 60 100250 400 550 700 850 1000 smm2 The faster a moleculediffuses the greater is the attenuation and the weaker isthe corresponding pixel signal intensity on a DWI imageThe biexponential analysis used in (4) to describe DWIcomprises a mixture of perfusion and diffusion whereas asimple monoexponential analysis only uses the b-values of 0smm2 and 1000 smm2 via (3)

The ADC was obtained by the monoexponential modelin (3) using b-values ranging from 0 to 1000 smm2 and thenfitted to the biexponential model in (4) Four parametersresult from the series of processes preprocessing detectionand analysis of the lesion tissue D (puremolecular diffusion)Dlowast (perfusion-related diffusion) PF (perfusion fraction)and slope Figure 4 shows the flowchart for IVIM imagingprocessing

Details of step-by-step implementations are as follows

Step 1 TheADC D Dlowast PF and slope values were calculatedto find the threshold values for central tumor peripheraltumor normal tissue and cyst

Step 2 The ADC slope D Dlowast and PF values of each pixelwere used to run the decision tree to determine to which ofthe three categories a pixel belongs to tumor normal tissueor cyst

Step 3 If a pixel indicated a tumor then histogram analysiswas used to determine whether the pixel indicated a centraltumor or a peripheral tumor

Step 4 Otherwise the pixel indicated a nontumor lesionThedecision tree was then run again to determine whether thepixel was normal tissue cyst or other

Histogram analysis can provide information beyond theIVIM parameters in the spatial distribution histogram suchas the skewness and kurtosis of the parameter distributionsSkewness is a measure of symmetry where kurtosis is ameasure of whether the data are peaked or flat relative to anormal distribution

4 Results and Discussion

Figure 5 shows a case of breast cancer on IVIM images atdifferent b values (a) b = 0 (b) b =15 (c) b = 30 (d) b = 45


(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

(m) (n)

Figure 5 The IVIM-DWI images of left breast cancer at different b values (a) b = 0 (b) b =15 (c) b = 30 (d) b = 45 (e) b = 60 (f) b = 100 (g) b= 250 (h) b = 400 (i) b = 550 (j) b = 700 (k) b = 850 (l) b = 1000 smm2 (m) The dynamic T1-DCE MR image (n) Lesion mapping with redand green pixels indicating the central and peripheral tumor regions respectively


Table 1 Monoexponential analysis (ie ADC) IVIM parameters (ie D Dlowast and PF) and slope in one case of malignant breast cancer cases

Slice ADC(10minus3 mm2s) slope D

(10minus3 mm2s)Dlowast

(10minus3 mm2s)PF()

10 134 -029 0743 921 2611 124 -0271 0737 88 2312 121 -0258 0732 857 2213 12 -0245 0718 988 2214 125 -0243 0741 1018 2315 129 -0251 0771 1015 2316 131 -0253 0778 1009 2417 126 -0248 0761 1015 2218 118 -0242 0734 1015 2019 115 -0239 0707 1015 2020 115 -0239 068 1013 2121 122 -0248 0696 1031 2322 121 -0252 0702 1005 23Average 123 -0252 0731 1008 22

(e) b = 60 (f) b = 100 (g) b = 250 (h) b = 400 (i) b = 550 (j) b =700 (k) b = 850 (l) b = 1000 smm2 (m)The dynamic T1-DCEMR image The lesion mappings are shown in image (n) Thered and green pixels indicate the central tumor region and theperipheral tumor region respectively

Figure 6 shows the ADC D Dlowast PF and slope values ofthe histogram analysis in the range of the tumor The red andgreen pixels show the central tumor and peripheral tumorregion respectively

To automatically find potential lesion pixels ATGP wasused to find the brightest point and then the SAM algorithmwas applied to locate the points with spectral signaturessimilar to the ATGP-found point that could be used asdesired target signatures for KCEM to detect a suspectedtumor range However the signal strength was insufficientlystrong to be curbed and the tumor area could not be fullydetectedTherefore a local entropic thresholdingmethodwasapplied to KCEM for suspected tumor range improvementFigure 7(a) shows the results of tumor detection usingKCEM

The local entropy method was used to find an appro-priate threshold value Threshold results are presented inFigure 7(b) in which areas with low intensity values couldstill be detected by KCEM However using the local entropymethod to adjust threshold values could only capture areasthat are different from the background In addition in thistechnique the threshold value needs to be manually adjustedslice by slice thus it is also subjective and time consumingTo address this issue a histogram analysis using the fiveparameters of ADC D Dlowast PF and slope was further usedto determine the true scale of the suspected tumor area As aresult some dense parts of the tissues could also be detectedas shown in Figure 7(c)

The monoexponential and biexponential models [ie (3)and (4) respectively] were then applied to the IVIM-DWimages in which the relative signal intensity decays of centraltumor peripheral tumor cyst and normal tissue are specifiedby the five parameters of ADC D Dlowast PF and slope We

first obtained MRI-IVIM images and then obtained contrast-enhanced MR images as the ground truth We used a patientwith breast cancer to demonstrate how these five parameterscould be used to classify breast tissues In Figure 7(b) thetumor area was marked by red pixels and was detected bythe KCEM method using the local entropy method For thepixels in the red area the calculated values of Dlowast were largerthan those of other areas without tumor The D and ADCvalues in the tumor area were both lower than those in thenormal breast tissue area The D PF and Dlowast values obtainedfrom IVIM biexponential model fitting and the ADC valueobtained frommonoexponential model fitting were also usedto classify breast tissues into three categories the tumor(central and peripheral tumor) cyst and normal tissuesThese values are tabulated in Tables 1 and 2

In particular Table 2 tabulates the average results of ADCD Dlowast PF and slope parameters of the total cases (ie testresults of 25 patients) based on mono- and biexponentialmodel analysis For all images each slice contained 12 IVIMimages from b=0 mm2s to b=1000 mm2s Each case maycontain 20 IVIM slices and more than 40 T1 T2 PD and T1-delay slices

A boxplot is a convenient way of graphically depictinggroups of numerical data through their quartiles Boxplotscan show the minimum (Q1) the medium (Q3) interquartilerange (interquartile range = Q3 ndash Q1) and maximum valuesThe boxplot in Figure 8 illustrates the distribution for all25 patients of biexponential median values for (a) ADC (b)slope (c) D (d) Dlowast and (e) PF

In practice the histological features of parameter D inIVIM are similar to those of parameter ADC in DWI How-ever it is better to use parameter D than to use the parameterADC because it can accurately demonstrate true diffusionwithout being affected by perfusion-related diffusion Inthis study we also found that parameter D has significantdifferences between malignant tumors and cysts so the Dvalue in IVIM can effectively complement existing traditional


02468

1012141618

0 05 1 15 2 25 3 35

amou

nt

ADC

05

10152025303540

0 05 1 15 2 25am

ount

D

0

5

10

15

20

25

30

0 01 02 03 04 05 06 07 08 09 1

amou

nt

PF

0

5

10

15

20

25

minus6 minus5 minus4 minus3 minus2 minus1 0

amou

nt

Signal Decay Slope

05

101520253035

0 10 20 30 40 50 60 70 80 90 100

amou

nt

＄lowast

ADC(10minus3

ＧＧ2Ｍ)

D(10minus3

ＧＧ2Ｍ)

Dlowast(10minus3

ＧＧ2Ｍ)

PF ()

Signal decay slope(10minus4)

ADC

D

Dlowast

PF

Slope

Figure 6 The ADC D Dlowast PF and slope parameters of the histogram analysis in the range of the tumor The red and green pixels indicatedthe central and peripheral tumor regions respectively


(a) (b) (c)

Figure 7 (a) Lesions detected by KCEM (b) lesions detected by thresholding the image in (a) by using local entropy method (c) lesiondetection by histogram analysis using decision tree

Mass Nonmass Cyst

112141618

222242628

3times10

minus3

(a) ADC

minus5

minus4

minus3

minus2times10

minus4

Mass Nonmass Cyst

(b) Slope

06

08

1

12

14

16

18

2

Mass Nonmass Cyst

times10minus3

(c) D

2

4

6

8

10

12

14

Mass Nonmass Cyst

times10minus3

(d) Dlowast

15

2

25

3

35

4

45

Mass Nonmass Cyst

(e) PF

Figure 8 Boxplots of the median values of different types of breast tissue based on (a) ADC (b) slope (c) D (d) Dlowast and (e) PF Thehorizontal solid red lines within each box represent the median value

Table 2The average values of the monoexponential model (ie ADC) IVIM parameters (ie D Dlowast and PF) and slope analysis of differentbreast tissues in our experimental case

ADC(10minus3 mm2s) slope D


(10minus3 mm2s)PF()

Central Area of the Tumor 123 (plusmn09) -024(plusmn02) 076(plusmn06) 938

(plusmn11) 215(plusmn1)

Peripheral Area of the Tumor 15 (plusmn013) -03(plusmn03) 093

(plusmn05) 902(plusmn103) 253

(plusmn3)Cyst 254 (plusmn025) -045



Normal Tissue 186 (plusmn025) -037(plusmn03) 114


(plusmn28)


DCE-MRI and DWI to distinguish between malignant breasttumors and cysts The mean D value of cysts is higher thanthe mean D value of malignant tumors close to the result ofMa et al [31] Inconsistent parameter results may be due todifferences in the number and distribution of low and highb values used in IVIM From the experimental results wealso found that the Dlowast and f values of benign and malignantbreast lesions have relatively low effects so the D parameteris given a higher weight in the part of the detection algorithmdecision tree One of the main limitations of our study is thatthe number of patients is not large enough including varietyof diseases Second the selection and appropriate number ofb values for breast IVIM are still unknown

Currently DCE-MRI is still one of the most importantand mature methods in breast lesion imaging AlthoughDCE-MRI has high sensitivity for detecting breast cancer itsdisadvantage is that it requires injection of contrast agentsHowever the significant association between gadolinium-based contrast agents and the incidence of nephrogenicsystemic fibrosis (NSF) in patients with advanced renaldisease has been reported DWI is a noninvasive method thatuses magnetic resonance imaging to observe the diffusion ofwater molecules in living tissue Current results show thatIVIM-DWI helps to understand tissue characteristics anddistinguish between benign andmalignant lesionsThereforeDWI may have a role as an alternative diagnostic techniquefor detecting breast lesion without the need of contrast agent

5 Conclusions

In this paper hyperspectral imaging techniquewas developedto detect and classify breast tissue lesions which can beimplemented in two stages The first stage is the detection ofthe breast tissue followed by the classification in the secondstage This study makes several contributions to the breastimaging process First the most important contributionis to process IVIM-DW images as multispectral imageswhich can be further extended to hyperspectral images byBEP Second an unsupervised hyperspectral target detectionalgorithm ATGP was applied to the expanded IVIM-DWimages to determine potential lesion pixels which can beused as the desired target knowledge for the follow-up targetdetection algorithm KCEM to locate suspected areas ofbreast cancer lesion tissues Third a thresholding techniquewas used to extract the lesion areas Fourth the detectedlesion pixels were used to calculate five parameters ADCD Dlowast PF and slope via the biexponential model specifiedby (4) The calculated values of ADC D Dlowast PF and slopewere ultimately input into a decision tree using histogramanalysis to classify the detected breast lesion tissues intothree categories malignant tumor (central and peripheraltumor) cysts and normal tissue It is our belief that the workpresented in this paper is the first to process breast IVIM-DW images as hyperspectral images Our findings show thathyperspectral imaging techniques could be used to detectbreast lesion tissues We also believe that this work is thefirst to take advantage of a biexponential model to classifybreast lesion tissue by using the five parameters of ADC DDlowast PF and slope This study was limited by its small sample

size and it did not employ multiple comparison correlationsTherefore the usefulness of the IVIM parameters for lesiondifferentiation requires further investigation

Data Availability

The data used to support the findings of this study areincluded within the article

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

References

[1] L A Torre R L Siegel E M Ward and A Jemal ldquoGlobalcancer incidence and mortality rates and trendsmdashan updaterdquoCancer Epidemiology Biomarkers amp Prevention vol 25 no 1pp 16ndash27 2016

[2] WorldCancerResearch Fund InternationalBreast cancer statis-tics 2018 httpswwwwcrforgintcancer-facts-figuresdata-specific-cancersbreast-cancer-statistics

[3] MAHelvie L K Joynt R L Cody L J PierceDD Adler andS DMerajver ldquoLocally advanced breast carcinoma accuracy ofmammography versus clinical examination in the prediction ofresidual disease after chemotherapyrdquo Radiology vol 198 no 2pp 327ndash332 1996

[4] B E Adrada L Huo D L Lane E M Arribas E ResetkovaandW Yang ldquoHistopathologic correlation of residual mammo-graphicmicrocalcifications after neoadjuvant chemotherapy forlocally advanced breast cancerrdquoAnnals of Surgical Oncology vol22 no 4 pp 1111ndash1117 2015

[5] W A Berg and P L Gilbreath ldquoMulticentric and multifocalcancer whole-breast us in preoperative evaluationrdquo Radiologyvol 214 no 1 pp 59ndash66 2000

[6] A A Tardivon L Ollivier C El Khoury and FThibault ldquoMon-itoring therapeutic efficacy in breast carcinomasrdquo EuropeanRadiology vol 16 no 11 pp 2549ndash2558 2006

[7] R Croshaw H Shapiro-Wright E Svensson K Erb and TJulian ldquoAccuracy of clinical examination digital mammogramultrasound and mri in determining postneoadjuvant patho-logic tumor response in operable breast cancer patientsrdquoAnnalsof Surgical Oncology vol 18 no 11 pp 3160ndash3163 2011

[8] F Pediconi C Catalano A Roselli et al ldquoThe challenge ofimaging dense breast parenchyma ismagnetic resonancemam-mography the technique of choiceA comparative study with x-ray mammography and whole-breast ultrasoundrdquo InvestigativeRadiology vol 44 no 7 pp 412ndash421 2009

[9] K Vassiou T Kanavou M Vlychou et al ldquoCharacterizationof breast lesions with CE-MR multimodal morphological andkinetic analysis Comparisonwith conventionalmammographyandhigh-resolutionultrasoundrdquoEuropean Journal of Radiologyvol 70 no 1 pp 69ndash76 2009

[10] M E Starver C E Loo E J Rutgers et al ldquoMRI-modelto guide the surgical treatment in breast cancerpatients afterneoadjuvant chemotherapyrdquo Annals of Surgery vol 251 no 4pp 701ndash707 2010

[11] G L G Menezes F M Knuttel B L Stehouwer R MPijnappel andM A A J Van Den Bosch ldquoMagnetic resonance


imaging in breast cancer A literature review and future per-spectivesrdquo World Journal of Clinical Oncology vol 5 no 2 pp61ndash70 2014

[12] W Huang P R Fisher K Dulaimy L A Tudorica B OrsquoHeaand T M Button ldquoDetection of breast malignancy diagnosticMR protocol for improved specificityrdquo Radiology vol 232 no2 pp 585ndash591 2004

[13] R M L Warren L Pointon D Thompson et al ldquoReadingprotocol for dynamic contrast-enhanced MR images of thebreast sensitivity and specificity analysisrdquo Radiology vol 236no 3 pp 779ndash788 2005

[14] W A Berg L Gutierrez M S NessAiver et al ldquoDiagnosticaccuracy of mammography clinical examination US and MRimaging in preoperative assessment of breast cancerrdquoRadiologyvol 233 no 3 pp 830ndash849 2004

[15] N H G M Peters I HM Borel Rinkes N P A ZuithoffW PTMMali KGMMoons andPHM Peeters ldquoMeta-analysisofMR imaging in the diagnosis of breast lesionsrdquoRadiology vol246 no 1 pp 116ndash124 2008

[16] M B I Lobbes R PrevosM Smidt et al ldquoThe role ofmagneticresonance imaging in assessing residual disease and pathologiccomplete response in breast cancer patients receiving neoadju-vant chemotherapy A systematic reviewrdquo Insights into Imagingvol 4 no 2 pp 163ndash175 2013

[17] C E Meacham and S J Morrison ldquoTumour heterogeneity andcancer cell plasticityrdquo Nature vol 501 no 7467 pp 328ndash3372013

[18] S Kul A Cansu E Alhan H Dinc G Gunes and AReis ldquoContribution of diffusion-weighted imaging to dynamiccontrast-enhanced MRI in the characterization of breasttumorsrdquo American Journal of Roentgenology vol 196 no 1 pp210ndash217 2011

[19] S C Partridge H Rahbar R Murthy et al ldquoImproved diag-nostic accuracy of breast MRI through combined apparentdiffusion coefficients and dynamic contrast-enhanced kineticsrdquoMagnetic Resonance in Medicine vol 65 no 6 pp 1759ndash17672011

[20] S Sinha F A Lucas-Quesada U Sinha N DeBruhl and L WBassett ldquoIn vivo diffusion-weightedMRI of the breast Potentialfor lesion characterizationrdquo Journal of Magnetic ResonanceImaging vol 15 no 6 pp 693ndash704 2002

[21] M Iima K Yano M Kataoka et al ldquoQuantitative non-gaussiandiffusion and intravoxel incoherentmotionmagnetic resonanceimaging differentiation ofmalignant andbenignbreast lesionsrdquoInvestigative Radiology vol 50 no 4 pp 205ndash211 2015

[22] D-M Koh D J Collins and M R Orton ldquoIntravoxel inco-herent motion in body diffusion-weighted MRI Reality andchallengesrdquo American Journal of Roentgenology vol 196 no 6pp 1351ndash1361 2011

[23] X Chen W Li Y Zhang Q Wu Y Guo and Z Bai ldquoMeta-analysis of quantitative diffusion-weighted MR imaging in thedifferential diagnosis of breast lesionsrdquo BMC Cancer vol 10article 693 2010

[24] Y Mazaheri A Afaq D B Rowe Y Lu A Shukla-Dave andJ Grover ldquoDiffusion-weighted magnetic resonance imaging ofthe prostaterdquo Journal of Computer Assisted Tomography vol 36no 6 pp 695ndash703 2012

[25] D Le Bihan E Breton D Lallemand et al ldquoMR imagingof intravoxel incoherent motions application to diffusion andperfusion in neurologic disordersrdquo Radiology vol 161 no 2 pp401ndash407 1986

[26] D Le Bihan E Breton D Lallemand M-L Aubin J VignaudandM Laval-Jeantet ldquoSeparation of diffusion and perfusion inintravoxel incoherentmotion MR imagingrdquo Radiology vol 168no 2 pp 497ndash505 1988

[27] L Bokacheva J B Kaplan D D Giri et al ldquoIntravoxelincoherent motion diffusion-weighted MRI at 30 T differen-tiates malignant breast lesions from benign lesions and breastparenchymardquo Journal of Magnetic Resonance Imaging vol 40no 4 pp 813ndash823 2014

[28] C Liu C Liang Z Liu S Zhang and B Huang ldquoIntravoxelincoherentmotion (IVIM) in evaluation of breast lesions com-parison with conventional dwirdquo European Journal of Radiologyvol 82 no 12 pp e782ndashe789 2013

[29] E E Sigmund G Y Cho S Kim et al ldquoIntravoxel inco-herent motion imaging of tumor microenvironment in locallyadvanced breast cancerrdquo Magnetic Resonance in Medicine vol65 no 5 pp 1437ndash1447 2011

[30] X Mao X Zou N Yu X Jiang and J Du ldquoQuantitativeevaluation of intravoxel incoherent motion diffusion-weightedimaging (IVIM) for differential diagnosis and grading predic-tion of benign and malignant breast lesionsrdquo Medicine vol 97no 26 p e11109 2018

[31] D Ma F Lu X Zou et al ldquoIntravoxel incoherent motiondiffusion-weighted imaging as an adjunct to dynamic contrast-enhancedMRI to improve accuracy of the differential diagnosisof benign and malignant breast lesionsrdquo Magnetic ResonanceImaging vol 36 pp 175ndash179 2017

[32] M W Vannier T K Pilgram C M Speidel L R NeumannD L Rickman and L D Schertz ldquoValidation of MagneticResonance Imaging (MRI) multispectral tissue classificationrdquoComputerized Medical Imaging and Graphics vol 15 no 4 pp217ndash223 1991

[33] C I Chang Hyperspectral Imaging Techniques for SpectralDetection and Classification Springer Science and BusinessBerlinHeidelberg Germany 2003

[34] C I Chang Hyperspectral Data Processing Algorithm DesignandAnalysis JohnWileyamp SonsHobokenNJUSA 1st edition2013

[35] Y C Ouyang H M Chen J W Chai C Chen et al ldquoBandExpansion-Based Over-Complete Independent ComponentAnalysis for Multispectral Processing of Magnetic ResonanceImagesrdquo IEEE Transactions on Biomedical Engineering vol 55no 6 pp 1666ndash1677 2008

[36] C M Bishop Pattern Recognition and Machine LearningSpringer New York NY USA 2006

[37] R O Duda and P E Hart Pattern Classification and SceneAnalysis John Wiley amp Sons Hoboken NJ USA 1973

[38] RHsuan andCChein-I ldquoAutomatic spectral target recognitionin hyperspectral imageryrdquo IEEE Transactions on Aerospace andElectronic Systems vol 39 no 4 pp 1232ndash1249 2003

[39] X Jiao S S Shen P E Lewis and C Chang ldquoKernel-basedconstrained energy minimization (KCEM)rdquo in Algorithms andTechnologies for Multispectral Hyperspectral and UltraspectralImagery XIV vol 6966 p 69661S International Society forOptics and Photonics Bellingham WA USA 2008

[40] E M Charles-Edwards and N M de Souza ldquoDiffusion-weighted magnetic resonance imaging and its application tocancerrdquo Cancer Imaging vol 6 no 1 pp 135ndash143 2006

[41] H Ren and C-I Chang ldquoA generalized orthogonal sub-space projection approach to unsupervisedmultispectral imageclassificationrdquo IEEE Transactions on Geoscience and RemoteSensing vol 38 no 6 pp 2515ndash2528


[42] J C Harsanyi Detection and Classification of Subpixel SpectralSignatures in Hyperspectral Image Seq uences University ofMaryland Baltimore County 1993

[43] W H Farrand and J C Harsanyi ldquoMapping the distribution ofmine tailings in the Coeur drsquoAlene River Valley Idaho throughthe use of a constrained energy minimization techniquerdquoRemote Sensing of Environment vol 59 no 1 pp 64ndash76 1997

[44] C Chein-I ldquoTarget signature-constrained mixed pixel clas-sification for hyperspectral imageryrdquo IEEE Transactions onGeoscience and Remote Sensing vol 40 no 5 pp 1065ndash10812002

[45] N R Pal and S K Pal ldquoEntropy a new definition and itsapplicationsrdquo13e Institute of Electrical and Electronics EngineersSystems Man and Cybernetics Society vol 21 no 5 pp 1260ndash1270 1991

[46] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979

[47] C Chang K Chen J Wang and M L Althouse ldquoA relativeentropy-based approach to image thresholdingrdquo Pattern Recog-nition vol 27 no 9 pp 1275ndash1289 1994

[48] C I Chang Y Du J Wang S M Guo and P D Thouin ldquoAsurvey and comparative study of entropic and relative entropicthresholding techniquesrdquo IEE Proceedings Vision Image SignalProcess vol 153 no 6 pp 837-838 2006

Hindawiwwwhindawicom

International Journal of

Volume 2018

Zoology

Hindawiwwwhindawicom Volume 2018

Anatomy Research International

PeptidesInternational Journal of



Journal of Parasitology Research

GenomicsInternational Journal of


Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018


BioinformaticsAdvances in

Marine BiologyJournal of



Neuroscience Journal


BioMed Research International

Cell BiologyInternational Journal of



Biochemistry Research International

ArchaeaHindawiwwwhindawicom Volume 2018


Genetics Research International


Advances in

Virolog y Stem Cells International



Enzyme Research



MicrobiologyHindawiwwwhindawicom

Nucleic AcidsJournal of

Volume 2018

Submit your manuscripts atwwwhindawicom


Central Area of the Tumor

Peripheral Area of the Tumor

Normal Breast Tissue

Figure 1 Illustration of the breast tissue regions

[3ndash7] Recent studies demonstrate that dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) tech-niques are the most promising methods for detecting tumorsand assessing therapeutic response [7ndash9] Furthermore theadditional diagnostic value of breast MRI is that it detectsmore multifocal multicentric and contralateral diseases[10 11] The sensitivity of DCE-MRI in the diagnosis ofbreast cancer is as high as 88-100 [12 13] However thespecificity of DCE-MRI varies from 50 to 97 [11 14ndash16] This variability may result from different lesion criteriaand tumor heterogeneity used among studies [15ndash17] Thusdifferentiating a benign lesion from a malignant lesion usingDCE-MRI remains a challenge To increase breast MRIspecificity one of the latest advancements inMRI technologythe apparent diffusion coefficient (ADC) which is obtainedfrom diffusion-weighted imaging (DWI) was developed todifferentiate between benign and malignant breast lesionswith 915 sensitivity and 865 specificity [18 19] A typicalillustration of the regions of the breast tissues is shown inFigure 1

DWI is a nocontrast-enhancedMRI technique It is usefulfor characterizing cancerous tissue heterogeneity by directlyreflecting the Brownian motion of water molecules in bodytissues particularly in breast cancer [19ndash21] It can obtainthe physiological characteristics of breast tissue with theassumption of the monoexponential decay to quantitate theanalysis of the water moleculesrsquo ADC [22] In malignanttumors the ADC value is usually lower than that of thenormal tissue because of the restriction of water diffusionresulting from increased cellularity of the tissue and reducedextracellular space in the tumor Based on the findings ofa meta-analysis [23] the specificity of DWI is higher thancontrast-enhanced MRI although there is a wide range inthreshold values from 090 mm2s to 160 x 10minus3 mm2s toaccount for the use of different b values The specificity ofDWI and DCE-MRI is 84 and 72 [23]

In vivo microscopic motion of water molecules detectedby DWI is influenced by the diffusion of water moleculesin the tissue structure and by the microcirculation of blood

in the capillary network [24] Signal attenuation on mono-exponential DWI therefore presents a linear relationshipbut does not truly reflect the microstructure changes inorganization The reason for this finding is that the ADCvalue is overestimated due to the fact that it contains twotypes of information (1) microcirculation perfusion of bloodin a capillary network and (2) water molecule diffusion ofthe tissue structure Intravoxel incoherent motion (IVIM)which was first described by Li Bihan et al in 1986 [25]separates microcirculation perfusion and water moleculediffusion by using a biexponential model [22 25 26] Usingthis biexponential model DWI is able to use signal intensityproduced by multiple b-values to analyze watertissue dif-fusitvity At low b-values the signal intensity reflects waterdiffusion in tissues as well as microcirculation in a capillarynetwork At high b-values the signal intensity reflects tissuediffusivity Through IVIM biomarkers have been resolvedas the perfusion fraction (PF) water molecule diffusivity(D) and pseudodiffusion (Dlowast) Recent studies demonstratethat the IVIM model is helpful for the differential diagnosisof breast lesions [27ndash31] In general a MR image can beviewed as a multispectral remote sensing image in whicheach spectral band image can be considered as an imageacquired by a particular pulse sequence [32] Hyperspectralimaging has become an emerging remote sensing technologythat improves traditional multispectral imaging by usinghundreds of contiguous spectral bands With such finespectral resolution many unknown subtle substances canbe revealed for subpixel detection and mixed pixel analysis[33] The application of hyperspectral imaging techniquesin MR brain imaging has also been investigated [33ndash35]Many medical imaging techniques are generally supervisedand require training samples provided a priori such as max-imum likelihood classification (MLC) K-nearest-neighbor(KNN) classification [36] neural networks [36] supportvector machine (SVM) [36] and Fisherrsquos linear discrimi-nant analysis (FLDA) [37] To avoid this issue the trainingsamples should be obtained directly from the data to beprocessed in an unsupervised method In this instance wetake advantage of the automatic target generation process(ATGP) developed by Ren and Chang [38] in conjunctionwith the spectral angle mapper (SAM) method which allowsusers to select training samples automatically This procedureis then followed by a supervised hyperspectral subpixel targetdetection technique called kernel-based constrained energyminimization (KCEM) developed by Jiao and Chang [39] todetect breast lesion tissues for quantitative analysis for IVIM-DWI This technique considers MR image slices acquired bypulse sequences and DWI with weighting factors of 12 b-values as a 12-band multispectral image cube in which eachsequence-acquired image is viewed as a spectral band imageThe data set to be used for experiments was collected from 25patients for IVIM imaging examinations

The purpose of this study was three folds (1) to measureIVIM parametric maps (ie tissue diffusivity [D] perfusionfraction [PF] pseudodiffusion coefficient [Dlowast] and slope)obtained from multiple b values of normal breast tissuecysts and malignant tumors (2) to determine whetherthese parameters can be used to distinguish between benign



2 Background




with

119887 = 120574211989221205752 (Δ minus 1205753) (2)


D = minus 11198871 minus 1198872 ln(119878 (1198871)119878 (1198872)) (3)
















105

1

095

09

085

08

075

100 200 300 400 500 600 700 800 900 1000

inde

x

b value (sＧＧ2)

Sb

S0


b valueADCIVIM





related IVIM images



(120590B119896120590B119897120590B119898)minus1 B119896B119897B119898 (5)








X = max1le119894le119873

1003817100381710038171003817r1198941003817100381710038171003817 (6)


X = 1003817100381710038171003817r119894lowast1003817100381710038171003817 (7)






(a) (b) (c)













119910119894 = 119871sum119897=1

119908119897119903119894119897 = (w)119879 r119894 = r119879119894 w (11)


( 1119873) 119873sum119894=1

1199102119894 = ( 1119873) 119873sum119894=1

(r119879119894 w)2

= w119879 [( 1119873) 119873sum119894=1

r119894r119879

119894 ]w = w119879Rw

(12)

R = ( 1119873) 119873sum119894=1

r119894r119879

119894 (13)






min119908


(16)




RΦ= 1119873

119873sum119894=1

Φ119879 (r119894)Φ (r119894) (18)



ΦΦ (d) (19)




























OriginalIVIM Image








Non-Tumor Tumor

Cyst

Index Difference













(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

(m) (n)






(10minus3 mm2s)PF()

10 134 -029 0743 921 2611 124 -0271 0737 88 2312 121 -0258 0732 857 2213 12 -0245 0718 988 2214 125 -0243 0741 1018 2315 129 -0251 0771 1015 2316 131 -0253 0778 1009 2417 126 -0248 0761 1015 2218 118 -0242 0734 1015 2019 115 -0239 0707 1015 2020 115 -0239 068 1013 2121 122 -0248 0696 1031 2322 121 -0252 0702 1005 23Average 123 -0252 0731 1008 22











02468

1012141618

0 05 1 15 2 25 3 35

amou

nt

ADC

05

10152025303540

0 05 1 15 2 25am

ount

D

0

5

10

15

20

25

30

0 01 02 03 04 05 06 07 08 09 1

amou

nt

PF

0

5

10

15

20

25


amou

nt

Signal Decay Slope

05

101520253035

0 10 20 30 40 50 60 70 80 90 100

amou

nt

＄lowast

ADC(10minus3

ＧＧ2Ｍ)

D(10minus3

ＧＧ2Ｍ)

Dlowast(10minus3

ＧＧ2Ｍ)

PF ()


ADC

D

Dlowast

PF

Slope



(a) (b) (c)


Mass Nonmass Cyst

112141618

222242628

3times10

minus3

(a) ADC

minus5

minus4

minus3

minus2times10

minus4

Mass Nonmass Cyst

(b) Slope

06

08

1

12

14

16

18

2

Mass Nonmass Cyst

times10minus3

(c) D

2

4

6

8

10

12

14

Mass Nonmass Cyst

times10minus3

(d) Dlowast

15

2

25

3

35

4

45

Mass Nonmass Cyst

(e) PF





(10minus3 mm2s)PF()










(plusmn28)




5 Conclusions



Data Availability




References






















































Volume 2018

Zoology











Volume 2018

















Advances in




Enzyme Research





Volume 2018




2 Background




with

119887 = 120574211989221205752 (Δ minus 1205753) (2)


D = minus 11198871 minus 1198872 ln(119878 (1198871)119878 (1198872)) (3)
















105

1

095

09

085

08

075

100 200 300 400 500 600 700 800 900 1000

inde

x

b value (sＧＧ2)

Sb

S0


b valueADCIVIM





related IVIM images



(120590B119896120590B119897120590B119898)minus1 B119896B119897B119898 (5)








X = max1le119894le119873

1003817100381710038171003817r1198941003817100381710038171003817 (6)


X = 1003817100381710038171003817r119894lowast1003817100381710038171003817 (7)






(a) (b) (c)













119910119894 = 119871sum119897=1

119908119897119903119894119897 = (w)119879 r119894 = r119879119894 w (11)


( 1119873) 119873sum119894=1

1199102119894 = ( 1119873) 119873sum119894=1

(r119879119894 w)2

= w119879 [( 1119873) 119873sum119894=1

r119894r119879

119894 ]w = w119879Rw

(12)

R = ( 1119873) 119873sum119894=1

r119894r119879

119894 (13)






min119908


(16)




RΦ= 1119873

119873sum119894=1

Φ119879 (r119894)Φ (r119894) (18)



ΦΦ (d) (19)




























OriginalIVIM Image








Non-Tumor Tumor

Cyst

Index Difference













(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

(m) (n)






(10minus3 mm2s)PF()

10 134 -029 0743 921 2611 124 -0271 0737 88 2312 121 -0258 0732 857 2213 12 -0245 0718 988 2214 125 -0243 0741 1018 2315 129 -0251 0771 1015 2316 131 -0253 0778 1009 2417 126 -0248 0761 1015 2218 118 -0242 0734 1015 2019 115 -0239 0707 1015 2020 115 -0239 068 1013 2121 122 -0248 0696 1031 2322 121 -0252 0702 1005 23Average 123 -0252 0731 1008 22











02468

1012141618

0 05 1 15 2 25 3 35

amou

nt

ADC

05

10152025303540

0 05 1 15 2 25am

ount

D

0

5

10

15

20

25

30

0 01 02 03 04 05 06 07 08 09 1

amou

nt

PF

0

5

10

15

20

25


amou

nt

Signal Decay Slope

05

101520253035

0 10 20 30 40 50 60 70 80 90 100

amou

nt

＄lowast

ADC(10minus3

ＧＧ2Ｍ)

D(10minus3

ＧＧ2Ｍ)

Dlowast(10minus3

ＧＧ2Ｍ)

PF ()


ADC

D

Dlowast

PF

Slope



(a) (b) (c)


Mass Nonmass Cyst

112141618

222242628

3times10

minus3

(a) ADC

minus5

minus4

minus3

minus2times10

minus4

Mass Nonmass Cyst

(b) Slope

06

08

1

12

14

16

18

2

Mass Nonmass Cyst

times10minus3

(c) D

2

4

6

8

10

12

14

Mass Nonmass Cyst

times10minus3

(d) Dlowast

15

2

25

3

35

4

45

Mass Nonmass Cyst

(e) PF





(10minus3 mm2s)PF()










(plusmn28)




5 Conclusions



Data Availability




References






















































Volume 2018

Zoology











Volume 2018

















Advances in




Enzyme Research





Volume 2018





105

1

095

09

085

08

075

100 200 300 400 500 600 700 800 900 1000

inde

x

b value (sＧＧ2)

Sb

S0


b valueADCIVIM





related IVIM images



(120590B119896120590B119897120590B119898)minus1 B119896B119897B119898 (5)








X = max1le119894le119873

1003817100381710038171003817r1198941003817100381710038171003817 (6)


X = 1003817100381710038171003817r119894lowast1003817100381710038171003817 (7)






(a) (b) (c)













119910119894 = 119871sum119897=1

119908119897119903119894119897 = (w)119879 r119894 = r119879119894 w (11)


( 1119873) 119873sum119894=1

1199102119894 = ( 1119873) 119873sum119894=1

(r119879119894 w)2

= w119879 [( 1119873) 119873sum119894=1

r119894r119879

119894 ]w = w119879Rw

(12)

R = ( 1119873) 119873sum119894=1

r119894r119879

119894 (13)






min119908


(16)




RΦ= 1119873

119873sum119894=1

Φ119879 (r119894)Φ (r119894) (18)



ΦΦ (d) (19)




























OriginalIVIM Image








Non-Tumor Tumor

Cyst

Index Difference













(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

(m) (n)






(10minus3 mm2s)PF()

10 134 -029 0743 921 2611 124 -0271 0737 88 2312 121 -0258 0732 857 2213 12 -0245 0718 988 2214 125 -0243 0741 1018 2315 129 -0251 0771 1015 2316 131 -0253 0778 1009 2417 126 -0248 0761 1015 2218 118 -0242 0734 1015 2019 115 -0239 0707 1015 2020 115 -0239 068 1013 2121 122 -0248 0696 1031 2322 121 -0252 0702 1005 23Average 123 -0252 0731 1008 22











02468

1012141618

0 05 1 15 2 25 3 35

amou

nt

ADC

05

10152025303540

0 05 1 15 2 25am

ount

D

0

5

10

15

20

25

30

0 01 02 03 04 05 06 07 08 09 1

amou

nt

PF

0

5

10

15

20

25


amou

nt

Signal Decay Slope

05

101520253035

0 10 20 30 40 50 60 70 80 90 100

amou

nt

＄lowast

ADC(10minus3

ＧＧ2Ｍ)

D(10minus3

ＧＧ2Ｍ)

Dlowast(10minus3

ＧＧ2Ｍ)

PF ()


ADC

D

Dlowast

PF

Slope



(a) (b) (c)


Mass Nonmass Cyst

112141618

222242628

3times10

minus3

(a) ADC

minus5

minus4

minus3

minus2times10

minus4

Mass Nonmass Cyst

(b) Slope

06

08

1

12

14

16

18

2

Mass Nonmass Cyst

times10minus3

(c) D

2

4

6

8

10

12

14

Mass Nonmass Cyst

times10minus3

(d) Dlowast

15

2

25

3

35

4

45

Mass Nonmass Cyst

(e) PF





(10minus3 mm2s)PF()










(plusmn28)




5 Conclusions



Data Availability




References






















































Volume 2018

Zoology











Volume 2018

















Advances in




Enzyme Research





Volume 2018



(a) (b) (c)













119910119894 = 119871sum119897=1

119908119897119903119894119897 = (w)119879 r119894 = r119879119894 w (11)


( 1119873) 119873sum119894=1

1199102119894 = ( 1119873) 119873sum119894=1

(r119879119894 w)2

= w119879 [( 1119873) 119873sum119894=1

r119894r119879

119894 ]w = w119879Rw

(12)

R = ( 1119873) 119873sum119894=1

r119894r119879

119894 (13)






min119908


(16)




RΦ= 1119873

119873sum119894=1

Φ119879 (r119894)Φ (r119894) (18)



ΦΦ (d) (19)




























OriginalIVIM Image








Non-Tumor Tumor

Cyst

Index Difference













(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

(m) (n)






(10minus3 mm2s)PF()

10 134 -029 0743 921 2611 124 -0271 0737 88 2312 121 -0258 0732 857 2213 12 -0245 0718 988 2214 125 -0243 0741 1018 2315 129 -0251 0771 1015 2316 131 -0253 0778 1009 2417 126 -0248 0761 1015 2218 118 -0242 0734 1015 2019 115 -0239 0707 1015 2020 115 -0239 068 1013 2121 122 -0248 0696 1031 2322 121 -0252 0702 1005 23Average 123 -0252 0731 1008 22











02468

1012141618

0 05 1 15 2 25 3 35

amou

nt

ADC

05

10152025303540

0 05 1 15 2 25am

ount

D

0

5

10

15

20

25

30

0 01 02 03 04 05 06 07 08 09 1

amou

nt

PF

0

5

10

15

20

25


amou

nt

Signal Decay Slope

05

101520253035

0 10 20 30 40 50 60 70 80 90 100

amou

nt

＄lowast

ADC(10minus3

ＧＧ2Ｍ)

D(10minus3

ＧＧ2Ｍ)

Dlowast(10minus3

ＧＧ2Ｍ)

PF ()


ADC

D

Dlowast

PF

Slope



(a) (b) (c)


Mass Nonmass Cyst

112141618

222242628

3times10

minus3

(a) ADC

minus5

minus4

minus3

minus2times10

minus4

Mass Nonmass Cyst

(b) Slope

06

08

1

12

14

16

18

2

Mass Nonmass Cyst

times10minus3

(c) D

2

4

6

8

10

12

14

Mass Nonmass Cyst

times10minus3

(d) Dlowast

15

2

25

3

35

4

45

Mass Nonmass Cyst

(e) PF





(10minus3 mm2s)PF()










(plusmn28)




5 Conclusions



Data Availability




References






















































Volume 2018

Zoology











Volume 2018

















Advances in




Enzyme Research





Volume 2018




min119908


(16)




RΦ= 1119873

119873sum119894=1

Φ119879 (r119894)Φ (r119894) (18)



ΦΦ (d) (19)




























OriginalIVIM Image








Non-Tumor Tumor

Cyst

Index Difference













(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

(m) (n)






(10minus3 mm2s)PF()

10 134 -029 0743 921 2611 124 -0271 0737 88 2312 121 -0258 0732 857 2213 12 -0245 0718 988 2214 125 -0243 0741 1018 2315 129 -0251 0771 1015 2316 131 -0253 0778 1009 2417 126 -0248 0761 1015 2218 118 -0242 0734 1015 2019 115 -0239 0707 1015 2020 115 -0239 068 1013 2121 122 -0248 0696 1031 2322 121 -0252 0702 1005 23Average 123 -0252 0731 1008 22











02468

1012141618

0 05 1 15 2 25 3 35

amou

nt

ADC

05

10152025303540

0 05 1 15 2 25am

ount

D

0

5

10

15

20

25

30

0 01 02 03 04 05 06 07 08 09 1

amou

nt

PF

0

5

10

15

20

25


amou

nt

Signal Decay Slope

05

101520253035

0 10 20 30 40 50 60 70 80 90 100

amou

nt

＄lowast

ADC(10minus3

ＧＧ2Ｍ)

D(10minus3

ＧＧ2Ｍ)

Dlowast(10minus3

ＧＧ2Ｍ)

PF ()


ADC

D

Dlowast

PF

Slope



(a) (b) (c)


Mass Nonmass Cyst

112141618

222242628

3times10

minus3

(a) ADC

minus5

minus4

minus3

minus2times10

minus4

Mass Nonmass Cyst

(b) Slope

06

08

1

12

14

16

18

2

Mass Nonmass Cyst

times10minus3

(c) D

2

4

6

8

10

12

14

Mass Nonmass Cyst

times10minus3

(d) Dlowast

15

2

25

3

35

4

45

Mass Nonmass Cyst

(e) PF





(10minus3 mm2s)PF()










(plusmn28)




5 Conclusions



Data Availability




References






















































Volume 2018

Zoology











Volume 2018

















Advances in




Enzyme Research





Volume 2018




















OriginalIVIM Image








Non-Tumor Tumor

Cyst

Index Difference













(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

(m) (n)






(10minus3 mm2s)PF()

10 134 -029 0743 921 2611 124 -0271 0737 88 2312 121 -0258 0732 857 2213 12 -0245 0718 988 2214 125 -0243 0741 1018 2315 129 -0251 0771 1015 2316 131 -0253 0778 1009 2417 126 -0248 0761 1015 2218 118 -0242 0734 1015 2019 115 -0239 0707 1015 2020 115 -0239 068 1013 2121 122 -0248 0696 1031 2322 121 -0252 0702 1005 23Average 123 -0252 0731 1008 22











02468

1012141618

0 05 1 15 2 25 3 35

amou

nt

ADC

05

10152025303540

0 05 1 15 2 25am

ount

D

0

5

10

15

20

25

30

0 01 02 03 04 05 06 07 08 09 1

amou

nt

PF

0

5

10

15

20

25


amou

nt

Signal Decay Slope

05

101520253035

0 10 20 30 40 50 60 70 80 90 100

amou

nt

＄lowast

ADC(10minus3

ＧＧ2Ｍ)

D(10minus3

ＧＧ2Ｍ)

Dlowast(10minus3

ＧＧ2Ｍ)

PF ()


ADC

D

Dlowast

PF

Slope



(a) (b) (c)


Mass Nonmass Cyst

112141618

222242628

3times10

minus3

(a) ADC

minus5

minus4

minus3

minus2times10

minus4

Mass Nonmass Cyst

(b) Slope

06

08

1

12

14

16

18

2

Mass Nonmass Cyst

times10minus3

(c) D

2

4

6

8

10

12

14

Mass Nonmass Cyst

times10minus3

(d) Dlowast

15

2

25

3

35

4

45

Mass Nonmass Cyst

(e) PF





(10minus3 mm2s)PF()










(plusmn28)




5 Conclusions



Data Availability




References






















































Volume 2018

Zoology











Volume 2018

















Advances in




Enzyme Research





Volume 2018



(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

(m) (n)






(10minus3 mm2s)PF()

10 134 -029 0743 921 2611 124 -0271 0737 88 2312 121 -0258 0732 857 2213 12 -0245 0718 988 2214 125 -0243 0741 1018 2315 129 -0251 0771 1015 2316 131 -0253 0778 1009 2417 126 -0248 0761 1015 2218 118 -0242 0734 1015 2019 115 -0239 0707 1015 2020 115 -0239 068 1013 2121 122 -0248 0696 1031 2322 121 -0252 0702 1005 23Average 123 -0252 0731 1008 22











02468

1012141618

0 05 1 15 2 25 3 35

amou

nt

ADC

05

10152025303540

0 05 1 15 2 25am

ount

D

0

5

10

15

20

25

30

0 01 02 03 04 05 06 07 08 09 1

amou

nt

PF

0

5

10

15

20

25


amou

nt

Signal Decay Slope

05

101520253035

0 10 20 30 40 50 60 70 80 90 100

amou

nt

＄lowast

ADC(10minus3

ＧＧ2Ｍ)

D(10minus3

ＧＧ2Ｍ)

Dlowast(10minus3

ＧＧ2Ｍ)

PF ()


ADC

D

Dlowast

PF

Slope



(a) (b) (c)


Mass Nonmass Cyst

112141618

222242628

3times10

minus3

(a) ADC

minus5

minus4

minus3

minus2times10

minus4

Mass Nonmass Cyst

(b) Slope

06

08

1

12

14

16

18

2

Mass Nonmass Cyst

times10minus3

(c) D

2

4

6

8

10

12

14

Mass Nonmass Cyst

times10minus3

(d) Dlowast

15

2

25

3

35

4

45

Mass Nonmass Cyst

(e) PF





(10minus3 mm2s)PF()










(plusmn28)




5 Conclusions



Data Availability




References






















































Volume 2018

Zoology











Volume 2018

















Advances in




Enzyme Research





Volume 2018






(10minus3 mm2s)PF()

10 134 -029 0743 921 2611 124 -0271 0737 88 2312 121 -0258 0732 857 2213 12 -0245 0718 988 2214 125 -0243 0741 1018 2315 129 -0251 0771 1015 2316 131 -0253 0778 1009 2417 126 -0248 0761 1015 2218 118 -0242 0734 1015 2019 115 -0239 0707 1015 2020 115 -0239 068 1013 2121 122 -0248 0696 1031 2322 121 -0252 0702 1005 23Average 123 -0252 0731 1008 22











02468

1012141618

0 05 1 15 2 25 3 35

amou

nt

ADC

05

10152025303540

0 05 1 15 2 25am

ount

D

0

5

10

15

20

25

30

0 01 02 03 04 05 06 07 08 09 1

amou

nt

PF

0

5

10

15

20

25


amou

nt

Signal Decay Slope

05

101520253035

0 10 20 30 40 50 60 70 80 90 100

amou

nt

＄lowast

ADC(10minus3

ＧＧ2Ｍ)

D(10minus3

ＧＧ2Ｍ)

Dlowast(10minus3

ＧＧ2Ｍ)

PF ()


ADC

D

Dlowast

PF

Slope



(a) (b) (c)


Mass Nonmass Cyst

112141618

222242628

3times10

minus3

(a) ADC

minus5

minus4

minus3

minus2times10

minus4

Mass Nonmass Cyst

(b) Slope

06

08

1

12

14

16

18

2

Mass Nonmass Cyst

times10minus3

(c) D

2

4

6

8

10

12

14

Mass Nonmass Cyst

times10minus3

(d) Dlowast

15

2

25

3

35

4

45

Mass Nonmass Cyst

(e) PF





(10minus3 mm2s)PF()










(plusmn28)




5 Conclusions



Data Availability




References






















































Volume 2018

Zoology











Volume 2018

















Advances in




Enzyme Research





Volume 2018



02468

1012141618

0 05 1 15 2 25 3 35

amou

nt

ADC

05

10152025303540

0 05 1 15 2 25am

ount

D

0

5

10

15

20

25

30

0 01 02 03 04 05 06 07 08 09 1

amou

nt

PF

0

5

10

15

20

25


amou

nt

Signal Decay Slope

05

101520253035

0 10 20 30 40 50 60 70 80 90 100

amou

nt

＄lowast

ADC(10minus3

ＧＧ2Ｍ)

D(10minus3

ＧＧ2Ｍ)

Dlowast(10minus3

ＧＧ2Ｍ)

PF ()


ADC

D

Dlowast

PF

Slope



(a) (b) (c)


Mass Nonmass Cyst

112141618

222242628

3times10

minus3

(a) ADC

minus5

minus4

minus3

minus2times10

minus4

Mass Nonmass Cyst

(b) Slope

06

08

1

12

14

16

18

2

Mass Nonmass Cyst

times10minus3

(c) D

2

4

6

8

10

12

14

Mass Nonmass Cyst

times10minus3

(d) Dlowast

15

2

25

3

35

4

45

Mass Nonmass Cyst

(e) PF





(10minus3 mm2s)PF()










(plusmn28)




5 Conclusions



Data Availability




References






















































Volume 2018

Zoology











Volume 2018

















Advances in




Enzyme Research





Volume 2018



(a) (b) (c)


Mass Nonmass Cyst

112141618

222242628

3times10

minus3

(a) ADC

minus5

minus4

minus3

minus2times10

minus4

Mass Nonmass Cyst

(b) Slope

06

08

1

12

14

16

18

2

Mass Nonmass Cyst

times10minus3

(c) D

2

4

6

8

10

12

14

Mass Nonmass Cyst

times10minus3

(d) Dlowast

15

2

25

3

35

4

45

Mass Nonmass Cyst

(e) PF





(10minus3 mm2s)PF()










(plusmn28)




5 Conclusions



Data Availability




References






















































Volume 2018

Zoology











Volume 2018

















Advances in




Enzyme Research





Volume 2018





5 Conclusions



Data Availability




References






















































Volume 2018

Zoology











Volume 2018

















Advances in




Enzyme Research





Volume 2018












































Volume 2018

Zoology











Volume 2018

















Advances in




Enzyme Research





Volume 2018


Breast Tumor Detection and Classification Using Intravoxel...

Documents

Transcript of Breast Tumor Detection and Classification Using Intravoxel...