18. Mass characterization with diffusion
Invasive Ductal Carcinoma of the left breast
18
Se Sp
DCE 98% 76%
ADC < 0.9 92% 87%
DCE+ADC 96% 89%
Kul S.AJR 2011
21. Characterization with IVIM+ADC0+K
21
} N=22 patients
} Tumor > 8 mm
} 15 cancers
} 8 benign lesions
} Cancer
} ADC0 î
} K ì
} fIVIM ì
Lima M, Invest Radiol 2014
22. Characterisation with perfusion
22
} N=124 patients (59 cancers)
DCE-MRI
Parameters
Normal
n=59
Benign
n=65
Malignant
n=59
Benign vs
normal
Malignant vs
normal
Malignant vs
benign
Ktrans (min–1) 0.049 0.280 0.783 <0.001 <0.001 <0.001
Kep (min–1) 0.121 0.483 1.304 <0.001 <0.001 <0.001
Ve 0.523 0.633 0.620 0.020 0.008 0.760
Li L, Med Sci Monit 2015
DCE-MRI
Parameters
MDD
n=9
DCIS
n=14
IDC
n=41
DCIS vs MDD IDC vs MDD
Ktrans (min–1) 0.313 0.713 0.803 <0.001 <0.001
Kep (min–1) 0.449 1.282 1.338 <0.001 <0.001
Ve 0.729 0.601 0.617 0.292 0.329
ADC 1.221 1.008 0.947 <0.001 <0.001
MDD : mammary ductal dysplasia
23. Grade SBR prediction with perfusion
23
} Grade SBR ì (N=50)
} Ktrans ì (p=0.002)
} kep ì (p=0.005)
} Ve î (p=0.038)
Koo HR, Eur J Radiol 2012
24. Status RE prediction with perfusion
24
} ER- vs ER+ (N=50)
} Ktrans ì (p=0.056)
} kep ì (p=0.043)
} Ve î (p=0.015)
Koo HR, Eur J Radiol 2012
Whitney U-test was used for pairwise comparisons
with Bonferroni correction. Statistical analyses were
performed using commercially available software
(SPSS, v. 19.0; Chicago, IL). Statistical significance
was assigned if the P-value was less than 0.05.
grade (0.585 6 0.243, P ¼ 0.038), and was lower in
tumors with ER negativity (0.455 6 0.201) than with
ER positivity (0.912 6 0.651, P ¼ 0.015). Other
prognostic factors did not show any differences in
quantitative parameters (Ktrans
, kep, and ve) (Table 1,
Figure 1. A 58-year-old
woman with a favorable his-
tology and lower mean Ktrans
value. The tumor was con-
firmed as a 2.5-cm invasive
ductal carcinoma of histologic
grade 2, nuclear grade 2, ER-
positive, PR-positive, and
HER2-negative. a: An axial 3D
fast SPGR subtraction MR
image demonstrates an irregu-
lar enhancing mass in the left
breast. b: The pixels contrib-
uting to the AIF are selected
within the ipsilateral internal
mammary artery (arrow). c,d:
Permeability map in a breast
tumor and the fitting result of
dynamic MR data by the
pharmacokinetic model based
on the Tofts model. Three
lines denote AIF, dynamic MR,
and its fitted data.
148 Koo et al.
Whitney U-test was used for pairwise comparisons
with Bonferroni correction. Statistical analyses were
performed using commercially available software
(SPSS, v. 19.0; Chicago, IL). Statistical significance
was assigned if the P-value was less than 0.05.
RESULTS
Correlation Between Perfusion Parameters and
Prognostic Factors
Mean Ktrans
was higher in tumors with a high histo-
logic grade (0.567 6 0.334) than tumors with a low
histologic grade (0.371 6 0.234, P ¼ 0.007), higher in
tumors with a high nuclear grade (0.581 6 0.323)
than with a low nuclear grade (0.353 6 0.226, P ¼
0.002), and higher in tumors with ER negativity
(0.576 6 0.346) than with ER positivity (0.420 6
0.263) with borderline significance (P ¼ 0.056). Mean
kep was higher in tumors with a high histologic grade
(1.294 6 0.736) than tumors with a low histologic
grade (0.822 6 0.652, P ¼ 0.005), higher in tumors
grade (0.585 6 0.
tumors with ER n
ER positivity (0.9
prognostic factors
quantitative param
Figs. 1, 2).
Correlation Betw
Immunohistochem
The Kruskal–Walli
ferent across the t
nificance (P ¼ 0.0
cancers showed a
than luminal bre
0.015) when comp
with Bonferroni co
tumors were iden
group than in the
vs. 31% [13/41],
perfusion parame
tochemical subtyp
Figure 2. A 66-year-old
woman with a poorer histol-
ogy, triple negativity, and higher
mean Ktrans
value. The tumor
was confirmed as a 1.8-cm
invasive ductal carcinoma that
was histologic grade 3, nuclear
grade 3, ER-negative, PR-nega-
tive, and HER2-negative. a: An
axial 3D fast SPGR subtraction
MR image demonstrates an
irregular enhancing mass in the
Perfusion MRI of Breast Cancers 149
with bilateral full breast coverage using parallel imag-
ing and through-plane zero-fill interpolation (ZIP).
MRI include bilateral image acquisition with a
positioning bilateral breast coil, temporal resolu
Figure 2. A 66-year-old
woman with a poorer histol-
ogy, triple negativity, and higher
mean Ktrans
value. The tumor
was confirmed as a 1.8-cm
invasive ductal carcinoma that
was histologic grade 3, nuclear
grade 3, ER-negative, PR-nega-
tive, and HER2-negative. a: An
axial 3D fast SPGR subtraction
MR image demonstrates an
irregular enhancing mass in the
right breast. b: The pixels con-
tributing to the AIF are selected
within the ipsilateral internal
mammary artery (arrow). c,d:
Permeability map in a breast tu-
mor and the fitting result of
dynamic MR data by the phar-
macokinetic model based on the
Tofts model. Three lines denote
AIF, dynamic MR, and its fitted
data. [Color figure can be
viewed in the online issue,
which is available at
wileyonlinelibrary.com.]
Perfusion MRI of Breast Cancers
25. Perfusion for prediction breast cancer subtypes
25
} Triple Negatif vs Luminal A (N=37)
} kep ì
} Ve î
Li S, Eur Radiol 2011
26. Texture for prediction breast cancer subtypes
26
} N=144 breast cancers
} 92 IDC, 45 ILC et 7 DCIS
} Luminaux A et B, HER2,TN
} Entropy in T1 Gd
} IDC > ILC p<.001
} Lum < HER2 p=.005
} Lum < TN p=.014
Waugh. Eur Radiol 2016
T1 Gd T2
Entropy = irregularity degree
27. Radiomics for prediction of molecular subtypes
27
} N=60 patients
Ming F, Plos One 2017
n AUC Se (%) Sp (%)
Luminal A 34 0.867 88 77
Luminal B 8 0.786 87 63
HER2 7 0.888 81 100
Basal Like 11 0.923 81 100
Total 0.869
29. Response prediction to NACT at baseline using ADC
29
ADC
(N=118)
Response
1.04 Progression
1.07 Partial Response
1.06 Complete Response
Richard R. Eur Radiol 2013
Baseline
30. Response prediction to NACT at baseline using ADC
30
RO. ADC
RO + 0.99
RO - 1.16
Grade ADC
I 0.95
II 0.96
III 1.11
Ki67 ADC
≧14 1.08
< 14 1.03
P53 ADC
Mutation 1.10
No Mutation 1.02
Richard R. Eur Radiol 2013
31. Response prediction to NACT at baseline using ADC
31
Luminal A Luminal B HER2 Basal like
0,6
0,8
1
1,2
1,4
1,6
1,8
ADC
Anova, P=0.0001
RC Non-RC
0,6
0,8
1
1,2
1,4
1,6
1,8
ADC
P=0.047
Richard R. Eur Radiol 2013
33. Radiomics for prediction of response to NACT at baseline
33
} N=57 patients (Main Cohort)
} 1.5T scan system
} 47 Responders (RECIST 1.1)
} 10 Non-Responders (RECIST 1.1)
Ming F, Eur J Radiol 2017
AUC = 0.910
Se = 87%
Sp = 90%
34. Radiomics for prediction of response to NACT at baseline
34
} N=57 patients (Main Cohort)
} 1.5T scan system
} 47 Responders (RECIST 1.1)
} 10 Non-Responders (RECIST 1.1)
} N=46 patients (Reproducibility)
} 3T scan system
} 37 Responders (RECIST 1.1)
} 9 Non-Responders (RECIST 1.1)
Ming F, Eur J Radiol 2017
AUC Main cohort Reproducibility
Main cohort
features
0.910 0.713
Reproducibility
features
0.683 0.874
Common
parameters
0.717 0.781
35. Intra- and peritumoral radiomics
for pretreatment prediction of pCR
in NACT
N=117 Patients
AUC
Training set = 0.78
Independent testing = 0.74
Top radiomics features
Co-occurrence of Local Anisotropic Gradient
Orientations (homogénéhity-entropy)
Peak Laws level-ripple
35
CoLlAGe : Co-occurrence of Local Anisotropic Gradient Orientations
PLLR : Peak Laws level-ripple
TIL
TIL
TIL
➚ CoLlAGe
➚ CoLlAGe
➚PLLR
Braman NM, Breast Cancer Res 2017
36. ADC for prediction of response after 2 cycles to NACT
36 Pickles. Magn Reson Imaging 2006. Cohen. Can Assoc Radiol J 1996
Park SH. Eur Radiol 2012
Baseline 2nd Cycle
Prediction pCR (Se 100%, Sp 70%)
Cycle de CTNA ì ADC
C1 +15%
C2 +25%
C6 +55%IDC, RH-,HER-
37. ADC for prediction of response after 2 cycles to NACT
37
Baseline 2nd cycle
ADC = 1.2 x10- 3mm2/s ADC = 1.2 x10- 3mm2/s
38. Perfusion for prediction of response after 2 cycles to NACT
38
de Bazelaire C, Diagn Interv Imaging 2013
39. Perfusion for prediction of response after 2 cycles to NACT
39
de Bazelaire C, Diagn Interv Imaging 2013
40. Perfusion for prediction of response after 2 cycles to NACT
40
de Bazelaire C, Diagn Interv Imaging 2013
41. Perfusion for prediction of response after 2 cycles to NACT
41
de Bazelaire C, Diagn Interv Imaging 2013
42. Perfusion for prediction of response after 2 cycles to NACT
42
} î Ktrans 50% for responder
detection1
} Sensitivity 94%
} Specificity 82%
} î Ktrans de 40% pour détecter les
répondeurs2
} Sensitivity 75%
} Specificity 63%
1.Ah-See ML.Clin Cancer Res. 2008; 2.Wu LM. BCRT 2012
43. Texture for prediction of response to NACT at baseline
43
} N= 36 breast cancers treated by NACT
} 8pCR et 28 non responders
} IRM (T1,T2 and T1 Gd sub)
} Good response predicted by
} Uniformity ➚ T2 et T1
} Entropy ➘ T2 et T1
Parikh J. Radiology 2014
after 3cycles Se Sp
Entropy 88% 82%
Uniformity 88% 79%
Tumor size changes 50% 82%
45. Texture for prognosis prediction
45
} N=203 patients with breast cancer
} 4 death
} 22 relapse
} Relapse risk factors
} Stade N3 : Hazard Ratio (HR) = 11
} Triple Negative : HR=17
} Low entropy T1 Gd sub : HR=5
} High entropy T2 : HR=10
Kim J. Radiology 2017
T1 Gd T2
Entropy = irregularity degree
46. Radiomics for prognosis prediction
46
} 84 patients, 84 ans
} 88% CCI, 10% CLI, 2% mixtes
} Lum A, Lum B, HER2, Basal like, Normal like
} IRM
} Computer-Extracted Image Phenotype
} Correlation
} Radiomics and MammaPrint®, Oncotype
DX®, PAM50® (r=.5, p<.0001)
} Prognosis prediction
} AUC = 0,80
Hui L, Radiology 2016
CEIP algorithme
47. Radiomics for prognosis prediction
47
} N=261 patients
} Machine learning : Semiquantitative parameters
} Washout of tumor volume
} Washin of maximum intensity
} Proportion of tumor voxels that reach maximum intensity in the 1st post contast
} Proportion of tumor voxels that reach a treathold when FGT reached a value of the mean tumor
enhancement
} To discriminate
} High vs intermediate and low Oncotype DX scores
} AUC = 0.77
} High and intermediate vs low Oncotype DX scores
} AUC = 0.51
Saha A, J Cancer Res Clin Oncol. 2018
49. Take Home Messages
49
} Multiparametric MRI
} Diffusion, perfusion, radiomics
} All technics are usefull for
} Characterization: benign/malignant,Ttumor subtype
} Treatment monitoring
} Prognosis
} Perspectives
} To organize large studies+++
} To optimize MRI sequences
} To diffuse technics (software)
} To combine radiomics with diffusion and perfusion