decoding intratumoral heterogeneity of breast cancer by ... · published onlinefirst july 27, 2016;...

Tumor and Stem Cell Biology

Decoding Intratumoral Heterogeneity of BreastCancer by Multiparametric In Vivo Imaging:A Translational StudyJennifer Schmitz1, Julian Schwab1, Johannes Schwenck1,2, Qian Chen3,Leticia Quintanilla-Martinez4, Markus Hahn5, Beate Wietek6, Nina Schwenzer6,Annette Staebler4, Ursula Kohlhofer4, Olulanu H. Aina3, Neil E. Hubbard3, Gerald Reischl1,Alexander D. Borowsky3, Sara Brucker5, Konstantin Nikolaou6,7, Christian la Foug�ere2,7,Robert D. Cardiff3, Bernd J. Pichler1,7, and Andreas M. Schmid1

Abstract

Differential diagnosis and therapy of heterogeneous breasttumors poses a major clinical challenge. To address the need fora comprehensive, noninvasive strategy to define the molecularand functional profiles of tumors in vivo, we investigated a novelcombination of metabolic PET and diffusion-weighted (DW)-MRI in the polyoma virus middle T antigen transgenic mouse modelof breast cancer. The implementation of a voxelwise analysisfor the clustering of intra- and intertumoral heterogeneity inthis model resulted in a multiparametric profile based on[18F]Fluorodeoxyglucose ([18F]FDG)-PET and DW-MRI, whichidentified three distinct tumor phenotypes in vivo, including solidacinar, and solid nodular malignancies as well as cystic hyper-plasia. To evaluate the feasibility of this approach for clinical use,

we examined estrogen receptor-positive and progesterone recep-tor-positive breast tumors from five patient cases using DW-MRIand [18F]FDG-PET in a simultaneous PET/MRI system. The post-surgical in vivo PET/MRI data were correlated to whole-slidehistology using the latter traditional diagnostic standard to definephenotype. By this approach, we showed how molecular, struc-tural (microscopic, anatomic), and functional information couldbe simultaneously obtained noninvasively to identify precancer-ous andmalignant subtypes within heterogeneous tumors. Com-bined with an automatized analysis, our results suggest thatmultiparametric molecular and functional imaging may be capa-ble of providing comprehensive tumor profiling for noninvasivecancer diagnostics. Cancer Res; 76(18); 5512–22. �2016 AACR.

IntroductionBeyond individual variances between different tumors (inter-

tumoral heterogeneity), intratumoral heterogeneity has become amajor focus in oncology (1–4). Defined by the appearance ofdifferent histologic, genetic, and molecular subpopulations with-in single tumors, intratumoral heterogeneity creates a new chal-lenge in cancer classification and drives neoplastic progressionand therapeutic resistance (5–7). To date, 50% to 80% of malig-

nant breast cancers do not display sufficient characteristics of anyspecial histologic type, but sometimes show high grades ofintratumoral heterogeneity (1, 2) and are subsequently classifiedas "invasive breast carcinoma of no special type" (IBC-NST;Supplementary Table S1 contains a list of abbreviations; ref. 8).

Pathologists, therefore, analyze tissue sections from differentregions of the tumor, reporting the highest observed grade (9).However, tumor classification and subsequent diagnosis in theclinic remainmostly based on small tissue samples that representa single point of information rather than the status of the wholelesion.

Molecular imaging modalities, such as PET and MRI, have thepotential to provide whole-lesion information noninvasively.PET shows great benefits in staging and therapy response evalu-ation of lung cancer and lymphoma (10–12) but has not beentransferred into routine clinical practice for breast cancer patients.

Routinely performed PET imaging analyses of tumors are basedon mean or hot spot values of tracer uptake, such as maximalstandardized uptake value (SUVmax) or "metabolic active tumorregion" evaluation (13, 14). Wahl and colleagues described thebeneficial value of the "PET response criteria in solid tumors"(PERCIST) and discussed the great advantages of hot spot analysisprocedures over the mean value analysis (15). However, all ofthese procedures neglect a substantial part of the information, asonly a small fraction of the tumor is considered. More sophisti-cated approaches such as radiomics, extracting large numbers ofquantitative imaging features for tissue characterization (16, 17),

1Department of Preclinical Imaging and Radiopharmacy, WernerSiemens Imaging Center, Eberhard Karls University Tuebingen, Tue-bingen, Germany. 2Department of Nuclear Medicine and ClinicalMolecular Imaging, Eberhard Karls University Tuebingen, Tuebingen,Germany. 3Center for Comparative Medicine, University of California,Davis,California. 4DepartmentofPathology,EberhardKarlsUniversityTuebingen, Tuebingen, Germany. 5Department of Women's Health,Eberhard Karls University Tuebingen, Tuebingen, Germany. 6Depart-ment of Diagnostic and Interventional Radiology, Eberhard KarlsUniversity, Tuebingen, Germany. 7German Cancer Consortium, Ger-man Cancer Research Center, Tuebingen, Germany.

Note: Supplementary data for this article are available at Cancer ResearchOnline (http://cancerres.aacrjournals.org/).

Corresponding Author: Andreas M. Schmid, Eberhard Karls University Tuebin-gen, Roentgenweg 13, 72076 Tuebingen, Germany. Phone: 70712987510; Fax:70712925111; E-mail: [email protected]

doi: 10.1158/0008-5472.CAN-15-0642

�2016 American Association for Cancer Research.

CancerResearch

Cancer Res; 76(18) September 15, 20165512

on December 21, 2020. © 2016 American Association for Cancer Research. cancerres.aacrjournals.org Downloaded from

Published OnlineFirst July 27, 2016; DOI: 10.1158/0008-5472.CAN-15-0642

http://cancerres.aacrjournals.org/

or other types of voxelwise analyses generating parametric mapsof functional parameters (18, 19) are rarely used in clinicalroutine PET.

We hypothesized that hybrid PET/MRI can provide a clinicallyrelevant tumor profiling and phenotyping approach of breastcancer. Thus, in this study, clinically applied PET and MRI meth-ods using [18F]Fluorodeoxyglucose ([18F]FDG)-PET and diffu-sion-weighted (DW)-MRI were combined with a uniqueapproach of voxelwise analysis, which identified different phe-notypes within single tumors. To mimic the heterogeneousnature of breast cancer, polyoma virus middle T antigen transgenic[tg(PyV-mT)] mice, that develop malignant tumors, passingthrough distinct stages of tumor progression, from atypical hyper-plasia to mammary intraepithelial neoplasia and invasive carci-noma (20, 21),were investigated. Furthermore, the clinical poten-tial of the method was proven, including a small cohort of fiveIBC-NST patients. Our clustering approach provides ameasure forthe number of clusters and corresponding thresholds for thesubsequent identification of tumor phenotypes and allows aneasy translation to other oncological entities.

Materials and MethodsMice

All experiments were performed with female tg(PyV-mT) miceon a C57BL/6 background. All animals were housed understandardized environmental conditions with free access to foodand water. The tg(PyV-mT) mice developed an average of fivepalpable tumors, composed of a mixture of glandular and cysticdysplasias and invasive carcinomas (21, 22).

Preclinical in vivo imagingPreclinical in vivo studies were performed in a sequential PET/

MRI setup using a dedicated 7 T small-animal scanner [ClinScan(7T-MRI); Bruker BioSpin GmbH] and a small-animal PET scan-ner [Inveon dedicated PET (DPET); Siemens Healthcare]. Fiveanesthetized animals (1–2% isoflurane evaporated in breathingair) at the age of 19 to 22 weeks, bearing together 26 tumors weremeasured on temperature-controlled animal beds. Apparent dif-fusion coefficient (ADC)MRI measurements using a half Fourier-acquired single shot turbo spin echo sequence [HASTE; repetitiontime (TR) 5,000 milliseconds, echo time (TE) 112 milliseconds,pixel size 0.21 � 0.21 mm2 in plane, 1 mm slice thickness,b-values 150, 600, 1000 s/mm2] and sequential [18F]FDG-PET/MRI [intravenously injection of 10 � 2 MBq of [18F]FDG; 50minutes sleeping uptake, 10 minutes emission scan, 13 minutestransmission scan] were performed on two consecutive days. PETdata were reconstructed using the ordered-subsets expectationmaximization algorithmwith 3D post-reconstruction (OSEM3D;Inveon Acquisition Workplace, Siemens Healthcare). The ana-tomical MRI was measured directly before the PET scans or theADC measurements with a 3D T2-weighted turbo spin echosequence (TSE; TR 2500milliseconds, TE 202milliseconds, voxelsize 0.27 � 0.27 � 0.27 mm3).

Although not included into this report, four further PET/MRImeasurements, investigating the PET tracers [68Ga]NODAGA-RGD, [11C]methionine, [11C]choline, and [18F]FMISO were per-formed with the same animals previously within 1 week.

An additional group of four tg(PyV-mT) mice at the age of 20weeks was measured in a sequential PET/MRI setup includinganatomical and DW-MRI and [18F]FDG-PET. The mice were

injected intravenously with 10 � 2 MBq of [18F]FDG. During theuptake time of 60 minutes, MRI measurements were performed,directly followed by the PET scans.

All experiments were approved by the RegierungspraesidiumTuebingen.

Preclinical ex vivo methodsFollowing the last in vivo measurement, the mice were sacri-

ficed, and tumors were excised and prepared for hematoxylinand eosin (H&E) histology and/or autoradiography followingstandard procedures (Supplementary Materials and Methods).

Clinical imagingFive patients at the age of 66 � 10 years with biopsy-proven

breast cancer lesions [estrogen receptor- and progesteronereceptor-positive (ERþ/PRþ) IBC-NST, G2–G3] of �1.5 cm insize were examined using a 3T-PET/MR scanner (BiographmMR; Siemens Healthcare). After 60 minutes of [18F]FDGuptake, the simultaneous PET/MRI acquisition was performed(258 � 4 MBq, 20 minutes PET acquisition, 4.3 mm maximalachievable resolution; ref. 23). The PET data were reconstructedusing an OSEM3D algorithm, applying an MR-based attenua-tion map from segmented MR images provided by the vendor.For the MRI measurements, a dedicated four-channel breastMRI coil (Siemens Healthcare), that was installed during allPET and MRI measurements, was used to acquire the followingMRI sequences within 20 minutes: short-time inversion recov-ery (STIR), DW echo planar imaging (EPI, TR 11,000 milli-seconds, TE 76 milliseconds, pixel size 1.8 � 1.8 mm2 in plane,4 mm slice thickness, b-values 50 and 800 s/mm2), contrastagent (Gadovist; Bayer Vital GmbH) enhanced (CE) dynamicimaging [T1 fast low-angle shot (FLASH) 3D] and post contrastagent T1 FLASH 3D fat saturated (TR 8.75 milliseconds, TE 4.21milliseconds, voxel size 0.9 � 0.9 � 0.9 mm3). ADC maps werecalculated from the DW EPI.

All patients gave written informed consent for the purpose ofanonymized evaluation andpublication of their data. All reportedinvestigations were conducted in accordance with the HelsinkiDeclaration and with national regulations.

Clinical ex vivo methodsPostsurgery, whole slide histology of the tumorswas performed

and correlated to the imaging data. The excised tissues were cut in0.5 cm slices for macroscopic analysis and fixed in formalin andembedded in paraffin. Tissue samples were taken for routinemicroscopic evaluation in 4 � 4 cm2 sections. Subsequently, 3mm H&E sections were prepared and digitized (Axioscope; LeicaMicrosystems).

Data analysisInveon Research Workplace (IRW; Siemens Healthcare) was

used for in vivo image analysis. The corresponding preclinical PETandMR images were fused and coregistered using capillaries filledwith the tracer, which were placed on the animal beds as markersfor the PET and MRI data fusion. The simultaneously acquiredclinical datawere coregistered by the scanner software. Volumesofinterest (VOI) covering the entire regions (whole breasts orlesions, respectively) were drawn on the basis of the MRI data(T2-weighted TSE for preclinical data; postcontrast agent T1FLASH for the clinical data).

Decoding Intratumoral Heterogeneity

www.aacrjournals.org Cancer Res; 76(18) September 15, 2016 5513




Beyond the tumor-to-muscle ratios (TMR, preclinical data) andstandardized uptake values (SUV, clinical data), a new voxel-based approach was developed for [18F]FDG-PET and DW-MRIusing a Gaussian mixture model (GMM) in MATLAB (Math-Works), fitting multiple Gaussian distributions to the collectivevoxel data (24). For every set of data, several different mixturemodels with up to seven Gaussian distributions were fitted andcompared using the Akaike information criterion (AIC) andBayesian information criterion (BIC), as implemented in theMATLAB package. The intersections formed by the descendingpart of one Gaussian distribution and the ascending part of thefollowing distribution were used as thresholds to separate thesubpopulations. Every Gaussian distribution within the thresh-olds defined a distinct biomarker population. To illustrate theresults, histograms of the voxel values were calculated and over-laid with the GMM fitting results.

For ADCandPETdata, parametricmapswere computed, whereeach identified population was assigned to a specific color.Subsequently, for the sequentially acquired cohort multipara-metric [18F]FDG/ADCmapswere computed for the entire tumors.

Autoradiography correlationTo confirm the preclinical in vivo [18F]FDG results, the GMM

approach was transferred to the autoradiography data of eighttumors. Following a background correction, TMR maps werecomputed and pixel values were extracted. The collective data setof all tumor pixels within the eight tumors was fitted. Using thedetermined thresholds, parametric autoradiography maps werecomputed.

ResultsConventional image analysis hampers heterogeneous tumorcharacterization

First results from the tg(PyV-mT) studies revealed that entirelesions (n ¼ 26 tumors, 53–769 mm3 in size) could only bedetected on the basis of anatomical MRI, as intratumoral[18F]FDG uptake and the ADC were very heterogeneous. Becauseof the extensive intratumoral heterogeneity, phenotypic identifi-

cationofwhole lesions in termsof intertumoral comparisonusingsingle value analysis, such as mean and maximal value analyses,was not feasible, as bothmethods omitted important information(Supplementary Fig. S1).

Identification of intratumoral uptake populations using GMMfor a voxel-based analysis in tg(PyV-mT) miceDistinct [18F]FDG uptake populations correlated with phenotypicoutcome. To decode the heterogeneous tumor uptake pattern thatis shown for the [18F]FDG uptake of four representative tumors(Fig. 1A), a voxel-based analysis approach for themultiparametricimaging procedure was developed. Three mice at the age of 22weeks bearing 18 tumorswere analyzed by a voxelwise calculationof the TMR for every tumor, representing the training data set thatwas collectively fitted using the GMM.

Evaluating the AIC and BIC, a mixture of four uptake popula-tionswas determined as thebestfit for the [18F]FDGdata (Fig. 1B).The intersections of the four distributions at TMR ¼ 2.3, 5.1, and9.6 represented the thresholds between the uptake populations(Fig. 1C). Parametric color maps of the tumors were created byapplying these thresholds on the image data (Fig. 1E). Comparingto histology, these maps showed a correlation of the two popula-tions with low [18F]FDG (blue, TMR < 5.1) uptake with cystictumor regions, whereas solid tumor regions correlated very wellwith the population with the highest uptake (red, TMR > 9.6; Fig.1D and E). However, not every Gaussian distribution directlycorrelated to a specific histologic pattern. Because of the limitedspatial resolutionof thePET scanner (1.6mmmaximal achievableresolution; ref. 25), the population with intermediate uptake,representing TMR values between 5.1 and 9.6, indicated mixedregions in which more than one phenotype was present withinsingle voxels.

To confirm this phenotypic correlation, the identified thresh-olds from the training data set were applied to a test data set ofeight tumors (two mice at the age of 19 weeks) without using theGMM analysis again and correlated with ex vivo autoradiographyresults (Fig. 2). Therefore, the TMR maps of the autoradiographydata of these eight tumors were calculated and analyzed using theGMM algorithm. A total of five Gaussian distributions were

Figure 1.

Voxelwise analysis approach for[18F]FDG-PET. A, [18F]FDG-PET/MRIserial slides of four tumors. Voxel dataof 18 tumors was examined. A fourGaussian distribution was the best fitaccording to AIC and BIC (B) and wasapplied to the summed histogramwith the thresholds TMR ¼ 2.3, 5.1, and9.6 (C). H&E histology (D) comparedto a parametric color map of the in vivodata (E).

Schmitz et al.

Cancer Res; 76(18) September 15, 2016 Cancer Research5514




determined as thebestmodel ex vivo (Fig. 2A). The thresholdswerecalculated as TMR ¼ 3.2, 7.7, 16.2, and 41.6 and applied to theautoradiography data to compute parametric color maps. Thenumerical difference in the TMR thresholds between the in vivoand autoradiography measurements may be attributed to differ-ent shrinking behaviors of the tissues during the autoradiographyprocess, including the initial freezing and subsequent dryingduring the exposure time as well as partial volume effects (PVE)in PET. The 41.6 threshold was not considered to be a realthreshold as the two populations with the highest uptake valuesoverlapped significantly and were grouped together, as a securedassignment of individual voxels to either population was notpossible. As the first two blue populations (TMR < 3.2 and TMR3.2–7.7)mainly corresponded to the background and spillover atthe edges of the tumors, the number of dedicated tumor popula-tions was reduced to 2 (TMR 7.7–16.2 and 16.2 < TMR). Thisresult was expected due to the better spatial resolution of theautoradiography plate (0.05 mm pixel diameter) compared withthe PET data, which led to a reduction of the mixed populationthat was observed in vivo, when more than one phenotype waspresent in a single voxel. The correlation of the in vivo histogramsof single tumors (Fig. 2B) and the ex vivo histograms, as well assingle slices of the autoradiography (Fig. 2C) with the PET data(Fig. 2D), revealed a similar uptake pattern as shown for arepresentative tumor. Themixed areas in thePETdataweremainlycorrelated with the population showing positive uptake by auto-radiography, indicating a high amount of solid tissue in themixedareas of the in vivodata thatwas verifiedby the correspondingH&Ehistology. The lower values of these regions in the in vivo datacould be explained by an underestimation of the uptake due toPVE.

ADC complements the [18F]FDG pattern for population analysisTheGMMapproachwas also applied to the ADCdata. All tumorvoxels within the training data set were fitted (Fig. 3A). Accord-ing to the defined fitting criteria, five ADC populations wereidentified as the best model (Fig. 3B). As the first and secondpopulation represented background populations (data notshown), these two populations could be neglected. The thresh-olds of the three remaining populations were determined asADC ¼ 0.6 � 10�3 mm2/s and ADC ¼ 1.0 � 10�3 mm2/s.Mapping these populations to 3D parametric color maps andcorrelating these maps with the H&E histology showed a cor-relation of the high ADC values (ADC > 1.0 � 10�3 mm2/s)with the cystic hyperplastic regions within the lesions, whereasthe low ADC population (ADC < 0.6� 10�3 mm2/s) correlatedwith the solid regions. Similar to the [18F]FDG populations, amixed ADC population was observed (ADC: 0.6–1.0 � 10�3

mm2/s), where more than one phenotype was present withinsingle voxels.

To compare the parameters of [18F]FDG and ADC, both classi-ficationswere applied to each tumor voxelwithin the training dataset, to investigate the tumors' composition according to bothparameters (Fig. 3C). Although the population with high[18F]FDG uptake (TMR > 9.6) correlated with solid regionsshowing high metabolic activity, ADC maps clearly revealedhyperplastic cystic regions (ADC > 1.0 � 10�3 mm2/s), whichcould not be identified by the [18F]FDG uptake due to the lowuptake values < 5.1 TMR. The combination of both methods,[18F]FDG and ADC PET/MRI, provided the opportunity to iden-tify solid and cystic regions and thus, characterized the wholetumor burden within the mice without the need for tissuesections.

Figure 2.

Ex vivo correlation of the [18F]FDGdata with autoradiography. A, thepixel values (TMR) of eight tumors,and aGMM fit was appliedwith a resultof five Gaussian distribution with thethresholds TMR ¼ 3.2, 7.7, 16.2, and41.6. The in vivo histogram of a singletumor (B) and the in vivo parametricmap revealed good correlation to theex vivo [18F]FDG and the H&E-stainedsection (C) with the in vivo PET mapand MRI (D). The mixed population ofthe in vivo datawasmainly covered bythe high uptake population in the exvivo data (white arrow).






Multiparametric PET/MRI decodes intratumoral phenotypicheterogeneity in vivo

On the basis of these results, a multiparametric study wasdesigned that enabled the direct correlation of the voxel data ofboth modalities for in vivo phenotypic clustering. Because of thesequential imaging setup, the ADC and [18F]FDG data of fourmice at the ageof 20weeks could bedirectly fused for analysis. Thethresholds of the training data set were directly transferred to thistest data set without using the GMM fitting again.

Furthermore, the sequential setup allowed the automatizedclustering of the tumors into phenotypic regions based on thecoregistered ADC and [18F]FDG values in every voxel, providingphenotypic maps of the tumors (Fig. 4A).

As expected, fusion of [18F]FDG and ADC in the same tumorsenabled the identification of solid and cystic regions within thetumors (Fig. 4A). Furthermore, beyond this classification, thesolid regions could be differentiated into two distinct pheno-typic patterns, solid acinar and very rare solid nodular regions.Solid acinar regions were characterized by [18F]FDG uptakevalues higher than 9.1 TMR; as shown previously, cystic andsolid nodular regions showed low [18F]FDG uptake. Althoughthe cystic areas showed high ADC values (ADC > 1.0 � 10�3

mm2/s), solid regions were characterized by low ADC values.

Thus, the solid nodular regions were characterized by a double-negative signature (low ADC and low [18F]FDG uptake),enabling the further differentiation of the three phenotypes,cystic hyperplasia, solid acinar, and solid nodular malignancy,by the ADC-[18F]FDG combination (Fig. 4B and Supplemen-tary Table S2).

The identification of the solid nodular phenotype was animportant finding and an example of the benefits of combinedPET/MRI, as this nodular malignancy was not identifiable withsingle-modality imaging. In the MRI it was distinguished fromthe background and identified as a solid region, but could notbe differentiated from the acinar region, whereas [18F]FDGcould not detect the lesion at all. Thus, if a comparable maskedphenotype is present in human disease, the differentiation ofthis phenotype from metabolic active solid phenotypes couldonly be provided by the combined imaging procedure ofPET/MRI.

Clinical translation: PET/MRI decodes intratumoralheterogeneity in human breast cancer[18F]FDG-PET is a feasible tumor marker To confirm the transla-tional potential of the described approach, we investigatedfive patients with biopsy-proven IBC-NST, using simultaneous

Figure 3.

Voxelwise analysis approach forADC-maps. A, the voxel data of 18tumors was examined. B, a fiveGaussian distribution was determinedas the best fit according to AIC andBIC. C, 3D color maps of the [18F]FDGand ADC of a representative tumorcorrelated with the correspondinghistology slide.

Schmitz et al.





PET/MRI 7 � 3 days before surgery. After surgical excision,pathologic analysis of the whole lesions was performed.

The first approach was the metabolic identification of tumortissue within the breasts of the patients, to verify the potentialof the method to identify malignant breast tissue, as describedfor the preclinical model. For this purpose, VOIs covering boththe entire healthy and the entire diseased breasts were drawnon the basis of the CE-MRI. The clustering approach for[18F]FDG was applied separately on the sum of all voxelswithin the healthy (Fig. 5A) and within the diseased breasts(Fig. 5B). The best fit for the healthy breasts was identified for asum of four Gaussian distributions, whereas the diseasedbreasts were best fitted by a number of five Gaussian distribu-tions. The highest uptake population within the diseasedbreasts, including SUV > 1.1, was completely absent in thehealthy breasts (Fig. 5A and B).

Applying this threshold of SUV > 1.1 to the whole breasts, thetumors within the breasts of all five patients were identified andconfirmed as malignant tumors by histologic analysis (Fig. 5E,first row). The dimensions of the tumors given by this approach(maximal diameter, exemplary tumor: 16 � 15 � 15 mm3; Fig.5E) showed a good correlation of the identified tumors as com-pared with the dimensions described by postsurgical pathologicanalyses (maximal diameter, exemplary tumor: 16 � 13 � 10mm3; Fig. 5E and Supplementary Table S3). This first approachcould provide a user independent method to noninvasivelyidentify malignant tumors.

Combined [18F]FDG-ADC-PET/MRI is able to cluster heteroge-neous structures As the main purpose of the study was the intra-tumoral classification of the lesions, we isolated all voxels within

the morphologically identified tumor regions, to use the intratu-moral clustering on the sum of all tumor voxels within the fiveprimary tumors. For [18F]FDG, the clustering approach identifiedfive Gaussian distributions, separated by the following thresh-olds: SUV ¼ 0.5, 0.9, 1.6, and 2.6 (Fig. 5C). The ADC data wereclustered by only two Gaussian distributions as the best fit,separated by the threshold of 1.1 � 10�3 mm2/s, distinguishinga low diffusion population from a high diffusion population(Fig. 5D). However, in this case, the populations were not clearlyseparated, yielding a high amount of a mixed population withinthe blue range with low ADC values.

The thresholds were plotted back into theMRI data to illustratethe clustered populations spatially within the VOIs (Fig. 5E, thirdrow). In accordance to the preclinical data, the clinical datashowed the complementary pattern of [18F]FDGuptake andADC.Although the [18F]FDG uptake was higher in solid tumor regions,the ADC values were low in the solid tumor regions. However,higher ADC values were observed in the healthy gland structure ofthe breast and in single tumor regions, which were thereforesuggested to be less solid.

As the previous approach already showed that [18F]FDGuptakevalues up to SUV of 1.1 represent the background uptake ofhealthy breast tissue, the first two populations (SUV < 0.5 andSUV 0.5–0.9) could not be distinguished from the background ofthe healthy tissue. As these populations were only present in theperiphery of the tumors, surrounding the actual tumor areas,detected by the CE-MRI, they could be neglected as backgroundpopulations (Fig. 5E, third row).

The remaining three uptake populations, especially the twohighest populations, were considered as real tumor populationsand were spatially correlated to the histologic findings (Fig. 6).

Figure 4.

Multiparametric imaging approach.A, [18F]FDG andADCparametricmapswere calculated and combined withphenotypic maps. The combination ofboth criteria identified three distinctphenotypes (A and B): solid acinartumor regions ([18F]FDG TMR > 9.6),cystic hyperplastic regions (blue;ADC values > 1.0 � 10�3 mm2/s),and solid nodular regions with adouble-negative pattern (green;ADC <0.6� 10�3mm2/s and [18F]FDGTMR < 5.1; B). The solid nodularcluster (green) was mixed with thebackground population (A and B).






Comparing different regions of aggressiveness within thetumors scored by their proliferative activity and invasive growthpattern by histology, highly aggressive IC regions with highmitotic activity showed the highest [18F]FDG accumulation repre-sented by the highest uptake population, which is marked in redin the figures (SUV > 2.6).

Here, we present the tumors of three representativepatients. Figure 6A shows a tumor consisting homogeneously ofan invasive carcinoma, according to histology, which is mainlyrepresented by the highest red [18F]FDG population in the PETdata. However, less aggressive regions with lower mitotic activityshowed lower [18F]FDG accumulation, represented by the green(SUV 1.6–2.6) or blue (SUV 0.9–1.6) uptake populations. Thetumor in Fig. 6B contained large areas of sclerosis with less tumorcells and less mitotic activity that correlated with the green uptakepopulation. In contrast, sporadic small invasive carcinomaregions of high proliferation and aggressiveness within this tumorwere represented by small spots of the red [18F]FDGpopulation in

the imaging data (Fig. 6B). Focusing only on the CE-MRI scans,both tumors in Fig. 6A and B had very similar morphology. Theinter- and intratumoral biological and histological differences ofthese two tumors could only be distinguished by the metabolicPET clustering.

Moreover, combining [18F]FDG-PET, CE-, and DW-MRI, sev-eral regions in a tumor with a very heterogeneous morphologicand histologic pattern, composed of multiple foci of invasivecarcinoma and a fibroadenoma component, were distinguished(Fig. 7). Although the extent of the tumor tissue was delineatedby the CE-MRI, the aggressive spots of the invasive carcinomawere identified as described above by the high [18F]FDG uptakepopulation in red (SUV > 2.6), that also correlated with lowADC values (Fig. 7, Slice 1). In contrast, the fibroadenoma couldbe distinguished from the rest of the tumor by a lower [18F]FDGaccumulation (blue population, SUV 0.9–1.6) combined withthe high ADC population (ADC > 1.1 � 10�3 mm2/s; Fig. 7,Slice 2).

Figure 5.

Voxelwise analysis approach for clinical [18F]FDG-PET and DW-MRI. A and B, voxel data of the healthy (A) and diseased (B) breast were analyzed by theGMM approach for [18F]FDG. The best fits were identified for a sum of 4 (A) and 5 (B) Gaussian distributions, respectively. For intratumoral clustering, voxels withinall tumorswere fitted for [18F]FDGwith a best fit of fiveGaussian distributions (C) andADCwith a best fit of twoGaussian distributions (D). E, a representative tumor.CE-MRI as well as the parametric color maps produced for the whole breast approach for [18F]FDG, applying the threshold of 1.1 in SUV (first row), ADC andPET data (second row), and the parametric maps for the intratumoral clustering for [18F]FDG and ADC (third row).

Schmitz et al.





This clinical data demonstrated that a comprehensive tumorprofiling was possible using combined PET/MRI in the clinicalsetting. Thus, the method showed great potential for the clinicaltranslation in larger scale studies.

DiscussionCurrent trends in patient care aim for individualized medicine,

making diagnostics and treatment strategies more complex. Animportant prerequisite for individualized therapies is a sophisti-cated diagnostic strategy that reveals even small alterations in thedisease profile.

The standard diagnostic procedure in breast cancer involvesmammography, sonography, andhistologic validation.However,there are limitations to these techniques, for example in youngpatients with dense breast tissue (26). The value of PET and MRIfor breast cancer diagnosis remains controversial. As recom-mended by the national comprehensive cancer network (NCCN)in high-risk patients, MRI has clear advantages (27–30), but animprovement of patient outcomewas not yet confirmed (28, 31).Although not considered in the NCCN guideline to date, there isgrowing evidence for the value of PET and positron emissionmammography (PEM; refs. 32–36). Avril and colleagues corre-

lated [18F]FDG uptake with several prognostic markers such ashistologic tumor type (ductal vs. lobular), microscopic tumorpattern (nodular vs. diffuse), and tumor proliferation (37). Yet,standard [18F]FDG-PET appears insufficient as single informationfor breast cancer diagnosis (38). Other promising tracers, e.g.,addressing hormone-receptor status or gastrin-releasing peptidereceptor are currently under clinical investigation (32, 39–41).Major restrictions in most clinical studies compared with thisstudy are the exclusive use of single modality imaging andconventional image analysis. For example, when mammogra-phy is difficult, as in high-risk patients with mammographicdense breast tissue (42), the excellent soft tissue contrast of theMRI and the high sensitivity of the PET and functional MRI canprovide essential information in the diagnostic process (26, 37,43). Here, we showed that in combination with advancedanalysis approaches, molecular and functional imaging canprovide a more detailed characterization of the tumors inter-and intratumorally.

The assessment of tumor heterogeneity represents an emergingchallenge in cancer diagnostics and treatment, for example, poly-genic drug resistance is a major issue in cancer patient care to date(44). Potts and colleagues summarized outstanding work iden-tifying breast cancer heterogeneity and predicting therapy

Figure 6.

[18F]FDG uptake populationsrevealed histologic differences inheterogeneous tumors in the clinicalsetting. A, in a representative tumor,the highest [18F]FDG accumulationrepresented (red, SUV > 2.6)correlated with highly aggressiveIC regions in H&E histology(magnification, �400). B, anothertumor showed large areas of lower[18F]FDG uptake populations (green,SUV 1.6–2.6 or blue, SUV 0.9–1.6)correlating to areas of sclerosis in thehistologic overview and in the greenhistology box (magnification, �400);smaller IC regions within this tumor inthe red histology box (magnification,�400) correlated to small spots of thered population in the imaging data.






response and resistance on the basis of immunohistochemistry(4). To overcome the snapshot character of these biopsy-baseddiagnoses, in vivo diagnostic tools like MRI, CT, and PET, com-bined with advanced analysis approaches, such as multispectralanalysis or texture analysis, which are described as a powerfulprognostic feature to improve diagnostics, tumor staging as wellas therapy response assessment in many types of cancer (45–47)must be correlated to histology. Especially MRI, including func-tional and spectroscopic techniques showed great potential inimaging intratumoral heterogeneity and identifying biologicalsubpopulations within single tumors, distinguishing for exampleviable tumor from necrosis and adipose healthy tissues (48), orclassifying the tumor microenvironment in terms of vascularheterogeneity, hypoxia, and acidity (49, 50).

In our preclinical study, only combinedPET/MRI and advancedimage analysis enabled the detection and discrimination betweentwo malignant tumor phenotypes (solid nodular and acinar)from premalignant cystic hyperplasia. Especially the identifiedsolid nodular tumor regions with very low [18F]FDG uptakewould be misinterpreted as nonmalignant using a single PETapproach, if detected at all.

In the translational studywith five IBC-NST tumor patients, thesimultaneous PET/MRI approach including theGMManalysis ledto the differentiation of highly aggressive andmitotic active tumorregions to less aggressive and sclerotic regions as well as to benignlesions such as a fibroadenoma. These results pointed out theopportunities of in vivo classification of tumors in the diagnosticworkflow.

Using this analysis, phenotypic identification is based on data-derived thresholds, yielding a biology-driven distinction ratherthan user-dependent classification, showing prospective poten-

tial, where stable thresholds derived from the training data setwere successfully transferred to other tumors of the same origin inthe preclinical setting. Once defined, identified, and verified, thethresholds may be applied to other tumors within the samedisease without the need of the complete image analysis proce-dure for every individual dataset. Its applicability and diagnosticvalue inother preclinicalmodels aswell as a larger scale validationin the clinical setting, however, was beyond the scope of thiswork.This analysis describes that the coexistence of different pheno-typeswithin the same tumor can be visualized bymultiparametric[18F]FDG PET/MR imaging, in both in animals as well as inpatients suffering from breast cancer. This new approach has tobe tested in further prospective studies; we distinguished a clinicalsignificance to our approach.

Because treatment stratification is based on tumor biologyreceived byminimal invasive biopsy, an accurate tissue sampling,especially in multifocal and multicentric breast tumors, is highlydesirable. Therefore, our noninvasive tumor classification mightenable a precise target selection and biopsy guidance in breastcancer, by revealing the presence of tumor heterogeneity anddetecting the most aggressive part of the tumor, and thus wouldreduce the risk of undergrading and undertreatment.

Another application of this multiparametric imaging might bethe early response evaluation after one to two cycles of neo-adjuvant chemotherapy (NACT), in order to adapt cancer treat-ment regimen or to spare ineffective and toxicity-related therapy.Because a complete remission after NACT is related with longersurvival, an accurate prediction of a pathologic complete responseafterNACT and prior to surgery is highly desirable andmight havea significant impact on further treatment stratification, for exam-ple mastectomy versus breast conserving therapy especially in

Figure 7.

Combined [18F]FDG-PET and DW-MRI could distinguish several regions within heterogeneous disease: a representative tumor, composed of multiple fociof IC and a fibroadenoma component. In slice 1, the aggressive spots of the IC was identified by the high [18F]FDG population (red, SUV > 2.6). Thefibroadenoma in slice 2 could be distinguished from the rest of the tumor by a low [18F]FDG accumulation (blue, SUV 0.9–1.6) combined with high ADC values(ADC > 1.1 � 10�3 mm2/s); histologic slices are shown in �100 magnification, the inset in �400 magnification.

Schmitz et al.





multifocal/multicentric breast cancer. Finally, in cases of largerresection volumes, our approach might facilitate the final histo-pathological assessment and enhance the diagnostic accuracy, bydefining point of interest with suspicious vital/avital tumor areasthat would be processed separately after the surgical procedure.

Similar combined [18F]FDG/ADC approaches showed highdiagnostic value, for example, in lung and breast cancer patients(51, 52) but could not provide prognostic thresholds due to theexclusive use of single value analysis in the breast cancer study andhigh grades of heterogeneity in the investigated lung disease,although analyzed by a GMM approach.

Current trends in cancer research showed that radiomics pro-duce promising prognosticmarkers when applied on large patientcohorts (16, 17, 53). Our work further demonstrates that com-bined PET/MRI may complement these approaches by addinganother biology-derived feature into the workflow, potentiallymoving one step further toward personalized medicine andindividualized therapy planning.

Disclosure of Potential Conflicts of InterestNo potential conflicts of interest were disclosed.

Authors' ContributionsConception anddesign: J. Schmitz, J. Schwenck,Q.Chen, B.Wietek, A. Staebler,R.D. Cardiff, B.J. Pichler, A.M. SchmidDevelopment of methodology: J. Schmitz, J. Schwab, Q. Chen, M. Hahn,B. Wietek, O.H. Aina, A.D. Borowsky, K. Nikolaou, B.J. Pichler, A.M. SchmidAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): J. Schmitz, J. Schwenck, M. Hahn, B. Wietek,N. Schwenzer, A. Staebler, N.E. Hubbard, A.D. Borowsky, S. Brucker, K. Niko-laou, R.D. Cardiff, B.J. PichlerAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): J. Schmitz, J. Schwab, J. Schwenck, M. Hahn,A. Staebler, N.E. Hubbard, R.D. Cardiff, B.J. Pichler, A.M. Schmid

Writing, review, and/or revision of the manuscript: J. Schmitz, J. Schwab,J. Schwenck, L. Quintanilla-Martinez, M. Hahn, B. Wietek, N. Schwenzer,A. Staebler, N.E. Hubbard, G. Reischl, A.D. Borowsky, S. Brucker, K. Nikolaou,C. la Foug�ere, B.J. Pichler, A.M. SchmidAdministrative, technical, or material support (i.e., reporting or organizingdata, constructing databases): J. Schwab, Q. Chen, G. Reischl, A.D. Borowsky,R.D. Cardiff, A.M. SchmidStudy supervision: B. Wietek, S. Brucker, R.D. Cardiff, A.M. SchmidOther (performed histological analysis and interpretation of mouse model):L. Quintanilla-MartinezOther (analysis, interpretation and taking pictures of the mouse histopa-thology): U. KohlhoferOther (provided the PyVmT MINO model, provided training in use oftransplants, and analyzed the histopathology. Visited Tuebingen to discussthe results and to helpwith their analysis. Dr. Ainawas a postdoctoral traineewho assisted and was paid from my grants.): R.D. Cardiff

AcknowledgmentsThe authors would like to thank Maren Harant, Funda Cay, Daniel Bukala,

Sandro Aidone, Carsten Groeper, Gerd Zeger, Holger Schmidt, Kerstin Fuchs,Laura K€ubler, and Judith E. Walls for excellent technical assistance and theradiopharmacy team for the extensive tracer syntheses. Further, the authorsthank Siemens Healthcare, Erlangen, Germany, for providing the dedicatedfour-channel breast MRI coil for the clinical measurements.

Grant SupportParts of this work were funded by fort€une grant 2223-0-0, the Swiss Werner

Siemens Foundation, the European Research Council grant 323196 (Image-Link), and by grants U01 CA141582 and U01 CA141541 from the NationalCancer Institute's Mouse Models of Human Cancers Consortium.

The costs of publication of this articlewere defrayed inpart by the payment ofpage charges. This article must therefore be hereby marked advertisement inaccordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Received March 9, 2015; revised June 13, 2016; accepted July 6, 2016;published OnlineFirst July 27, 2016.

References1. Tavassoli FA, Devilee P. Pathology and genetics of tumours of the breast

and female genital organs. World Health Organization; 2003.2. Weigelt B, Horlings H, Kreike B, Hayes M, Hauptmann M, Wessels L, et al.

Refinement of breast cancer classification by molecular characterization ofhistological special types. J Pathol 2008;216:141–50.

3. Gerlinger M, Rowan AJ, Horswell S, Larkin J, Endesfelder D, Gronroos E,et al. Intratumor heterogeneity and branched evolution revealed by multi-region sequencing. N Engl J Med 2012;366:883–92.

4. Potts SJ, Krueger JS, Landis ND, Eberhard DA, Young GD, Schmechel SC,et al. Evaluating tumor heterogeneity in immunohistochemistry-stainedbreast cancer tissue. Lab Invest 2012;92:1342–57.

5. Russnes HG, Navin N, Hicks J, Borresen-Dale AL. Insight into the hetero-geneity of breast cancer through next-generation sequencing. J Clin Invest2011;121:3810–8.

6. Swanton C. Intratumor heterogeneity: evolution through space and time.Cancer Res 2012;72:4875–82.

7. Andor N, Graham TA, Jansen M, Xia LC, Aktipis CA, Petritsch C, et al. Pan-cancer analysis of the extent and consequences of intratumor heterogene-ity. Nat Med 2016;22:105–13.

8. Sinn H-P, Kreipe H. A brief overview of the WHO classification of breasttumors. Breast Care 2013;8:149–54.

9. Ignatiadis M, Sotiriou C. Understanding the molecular basis of histologicgrade. Pathobiology 2008;75:104–11.

10. Heusch P, Buchbender C, K€ohler J, Nensa F, Gauler T, Gomez B, et al.Thoracic staging in lung cancer: prospective comparison of 18F-FDG PET/MR imaging and 18F-FDG PET/CT. J Nucl Med 2014;55:373–8.

11. Boellaard R, Delgado-Bolton R, OyenWJ, Giammarile F, Tatsch K, EschnerW, et al. FDG PET/CT: EANM procedure guidelines for tumour imaging:version 2.0. Eur J Nucl Med Mol Imaging 2015;42:328–54.

12. Collins CD. PET in lymphoma. Cancer Imaging 2006;6(Spec No A):S63–70.13. Krak NC, van der Hoeven JJ, Hoekstra OS, Twisk JW, van der Wall E,

Lammertsma AA. Measuring [18F] FDG uptake in breast cancer duringchemotherapy: comparison of analytical methods. Eur J Nucl Med MolImaging 2003;30:674–81.

14. Larson SM, Schwartz LH. 18F-FDG PET as a candidate for "qualifiedbiomarker": functional assessment of treatment response in oncology.J Nucl Med 2006;47:901–3.

15. Wahl RL, Jacene H, Kasamon Y, Lodge MA. From RECIST to PERCIST:evolving considerations for PET response criteria in solid tumors. J NuclMed 2009;50(Suppl 1):122S–50S.

16. Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, van Stiphout RG,Granton P, et al. Radiomics: extracting more information from medicalimages using advanced feature analysis. Eur J Cancer 2012;48:441–6.

17. Gillies RJ, KinahanPE,HricakH.Radiomics: images aremore thanpictures,they are data. Radiology 2016;278:563–77.

18. Paran Y, Bendel P, Margalit R, Degani H. Water diffusion in the differentmicroenvironments of breast cancer. NMR Biomed 2004;17:170–80.

19. Han S, Ackerstaff E, Stoyanova R, Carlin S, Huang W, Koutcher J, et al.Gaussianmixturemodelbased classification of dynamic contrast enhancedMRI data for identifying diverse tumor microenvironments: preliminaryresults. NMR Biomed 2013;26:519–32.

20. Guy C, Cardiff R, Muller W. Induction of mammary tumors by expressionof polyomavirus middle T oncogene: a transgenic mouse model formetastatic disease. Mol Cell Biol 1992;12:954–61.

21. Maglione JE, McGoldrick ET, Young LJT, Namba R, Gregg JP, Liu L, et al.Polyomavirus middle T–induced mammary intraepithelial neoplasia out-growths: single origin, divergent evolution, and multiple outcomes. MolCancer Ther 2004;3:941–53.






22. Cardiff RD, Borowsky AD. Precancer: sequentially acquired or pre-determined? Toxicol Pathol 2010;38:171–9.

23. Delso G, F€urst S, Jakoby B, Ladebeck R, Ganter C, Nekolla SG, et al.Performance measurements of the Siemens mMR integrated whole-bodyPET/MR scanner. J Nucl Med 2011;52:1914–22.

24. McLachlan G, Peel D. Finitemixturemodels. New York: JohnWiley & SonsLtd; 2004.

25. Kemp BJ, Hruska CB, McFarland AR, Lenox MW, Lowe VJ. NEMA NU 2-2007 performance measurements of the Siemens Inveon preclinical smallanimal PET system. Phys Med Biol 2009;54:2359–76.

26. Sardanelli F, Giuseppetti GM, Panizza P, Bazzocchi M, Fausto A, SimonettiG, et al. Sensitivity of MRI versus mammography for detecting foci ofmultifocal, multicentric breast cancer in fatty and dense breasts using thewhole-breast pathologic examination as a gold standard. Am J Roentgenol2004;183:1149–57.

27. NCCN. NCCN clinical practice guidelines in oncology (NCCN guidelines)breast cancer screening and diagnosis version I.2014 2014.

28. Pilewskie M, King TA. Magnetic resonance imaging in patients withnewly diagnosed breast cancer: a review of the literature. Cancer2014;120:2080–9.

29. Kim JY, Cho N, Koo HR, Yi A, Kim WH, Lee SH, et al. Unilateral breastcancer: screening of contralateral breast by using preoperative MR imagingreduces incidence of metachronous cancer. Radiology 2013;267:57–66.

30. Obdeijn I-M, Tilanus-Linthorst MM, Spronk S, van Deurzen CH, MonyeCD, Hunink MM, et al. Preoperative breast MRI can reduce the rate oftumor-positive resection margins and reoperations in patients undergoingbreast-conserving surgery. Am J Roentgenol 2013;200:304–10.

31. Sung JS, Li J, Da Costa G, Patil S, Van Zee KJ, Dershaw DD, et al.Preoperative breast MRI for early-stage breast cancer: effect on surgicaland long-term outcomes. Am J Roentgenol 2014;202:1376–82.

32. Buck AK, Schirrmeister H, Mattfeldt T, Reske SN. Biological characteriza-tion of breast cancer by means of PET. Eur J Nucl Med Mol Imaging2004;31:S80–87.

33. Jansson T, Westlin J, Ahlstr€om H, Lilja A, La�ngstr€om B, Bergh J. Positron

emission tomography studies in patients with locally advanced and/ormetastatic breast cancer: a method for early therapy evaluation? J ClinOncol 1995;13:1470–7.

34. Humbert O, Cochet A, Coudert B, Berriolo-Riedinger A, Kanoun S, Bru-notte F, et al. Role of positron emission tomography for the monitoring ofresponse to therapy in breast cancer. Oncologist 2015;20:94–104.

35. Yamamoto Y, Tasaki Y, Kuwada Y, Ozawa Y, Inoue T. A preliminary reportof breast cancer screening by positron emission mammography. Ann NuclMed 2016;30:130–7.

36. Smyczek-Gargya B, FersisN,DittmannH, VogelU, ReischlG,MachullaH-J,et al. PET with [18F] fluorothymidine for imaging of primary breast cancer:a pilot study. Eur J Nucl Med Mol Imaging 2004;31:720–4.

37. Avril N, Menzel M, Dose J, Schelling M, Weber W, J€anicke F, et al. Glucosemetabolism of breast cancer assessed by 18F-FDG PET: histologic andimmunohistochemical tissue analysis. J Nucl Med 2001;42:9–16.

38. Ambrosini V, Fani M, Fanti S, Forrer F, Maecke HR. Radiopeptide imagingand therapy in Europe. J Nucl Med 2011;52(Suppl 2):42S–55S.

39. Prignon A, Nataf V, Provost C, Cagnolini A, Montravers F, Gruaz-Guyon A,et al. 68 Ga-AMBA and 18 F-FDG for preclinical PET imaging of breastcancer: effect of tamoxifen treatment on tracer uptake by tumor. Nucl MedBiol 2015;42:92–8.

40. Mortimer JE, Bading JR, Colcher DM, Conti PS, Frankel PH, Carroll MI,et al. Functional imaging of human epidermal growth factor receptor2–positive metastatic breast cancer using 64Cu-DOTA-trastuzumab PET.J Nucl Med 2014;55:23–9.

41. van KruchtenM, Glaudemans AW, de Vries EF, Beets-Tan RG, Schr€oder CP,Dierckx RA, et al. PET imaging of estrogen receptors as a diagnostic tool forbreast cancer patients presenting with a clinical dilemma. J Nucl Med2012;53:182–90.

42. Boyd NF, Guo H, Martin LJ, Sun L, Stone J, Fishell E, et al. Mammographicdensity and the risk and detection of breast cancer. N Engl J Med2007;356:227–36.

43. SiegmannKC,Kr€amer B,ClaussenC.Current status andnewdevelopmentsin breast MRI. Breast Care 2011;6:87–92.

44. Rehemtulla A. Overcoming intratumor heterogeneity of polygenic cancerdrug resistance with improved biomarker integration. Neoplasia (NewYork, NY) 2012;14:1278–89.

45. Davnall F, Yip CS, Ljungqvist G, Selmi M, Ng F, Sanghera B, et al.Assessment of tumor heterogeneity: an emerging imaging tool for clinicalpractice? Insights Imaging 2012;3:573–89.

46. SoussanM,Orlhac F, BoubayaM,Zelek L, ZiolM, Eder V, et al. Relationshipbetween tumor heterogeneity measured on FDG-PET/CT and pathologicalprognostic factors in invasive breast cancer. PLoS One 2014;9:e94017.

47. O'Connor JP, Rose CJ, Waterton JC, Carano RA, Parker GJ, Jackson A.Imaging intratumor heterogeneity: role in therapy response, resistance, andclinical outcome. Clin Cancer Res 2015;21:249–57.

48. Carano RA, Ross AL, Ross J, Williams SP, Koeppen H, Schwall RH, et al.Quantification of tumor tissue populations by multispectral analysis.Magn Reson Med 2004;51:542–51.

49. Jackson A, O'Connor JP, Parker GJ, Jayson GC. Imaging tumor vascularheterogeneity and angiogenesis using dynamic contrast-enhanced mag-netic resonance imaging. Clin Cancer Res 2007;13:3449–59.

50. Gillies RJ, Raghunand N, Karczmar GS, Bhujwalla ZM. MRI of the tumormicroenvironment. J Magn Reson Imaging 2002;16:430–50.

51. Schmidt H, Brendle C, Schraml C,Martirosian P, Bezrukov I, Hetzel J, et al.Correlationof simultaneously acquired diffusion-weighted imaging and2-deoxy-[18F] fluoro-2-D-glucose positron emission tomography of pulmo-nary lesions in a dedicated whole-body magnetic resonance/positronemission tomography system. Invest Radiol 2013;48:247–55.

52. Karan B, Pourbagher A, Torun N. Diffusionweighted imaging and 18Ffluorodeoxyglucose positron emission tomography/computed tomogra-phy in breast cancer: Correlation of the apparent diffusion coefficient andmaximum standardized uptake values with prognostic factors. J MagnReson Imaging 2016;43:1434–44.

53. Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Cavalho S,et al. Decoding tumor phenotype by noninvasive imaging using a quan-titative radiomics approach. Nature Commun 2014;4006. doi: 10.1038/ncomms5006.


Schmitz et al.




2016;76:5512-5522. Published OnlineFirst July 27, 2016.Cancer Res Jennifer Schmitz, Julian Schwab, Johannes Schwenck, et al.

Imaging: A Translational StudyIn VivoMultiparametric Decoding Intratumoral Heterogeneity of Breast Cancer by

Updated version

10.1158/0008-5472.CAN-15-0642doi:

Access the most recent version of this article at:

Material

Supplementary

http://cancerres.aacrjournals.org/content/suppl/2016/07/27/0008-5472.CAN-15-0642.DC1

Access the most recent supplemental material at:

Cited articles

http://cancerres.aacrjournals.org/content/76/18/5512.full#ref-list-1

This article cites 49 articles, 15 of which you can access for free at:

Citing articles

http://cancerres.aacrjournals.org/content/76/18/5512.full#related-urls

This article has been cited by 2 HighWire-hosted articles. Access the articles at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

Subscriptions

Reprints and

[email protected]

To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at

Permissions

Rightslink site. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC)

.http://cancerres.aacrjournals.org/content/76/18/5512To request permission to re-use all or part of this article, use this link



http://cancerres.aacrjournals.org/lookup/doi/10.1158/0008-5472.CAN-15-0642

http://cancerres.aacrjournals.org/content/suppl/2016/07/27/0008-5472.CAN-15-0642.DC1

http://cancerres.aacrjournals.org/content/76/18/5512.full#ref-list-1

http://cancerres.aacrjournals.org/content/76/18/5512.full#related-urls

http://cancerres.aacrjournals.org/cgi/alerts

mailto:[email protected]

http://cancerres.aacrjournals.org/content/76/18/5512


decoding intratumoral heterogeneity of breast cancer by ... · published onlinefirst july 27, 2016;...

Documents