| Περίληψη: | Breast cancer remains a major public health problem and is the second leading cause of death in both developing and developed countries. According to the American Cancer Society, only in 2016 they are about 249.260 estimated new cases in breast cancer, while the rate of breast cancer death is estimated about 40.890 cases in both sexes [Siegel R. et al., 2016].
Breast imaging includes X-ray mammography, ultrasound and Magnetic Resonance Imaging (MRI), used for breast cancer screening and diagnosis. Multi parametric MRI techniques such as Dynamic Contrast Enhanced (DCE), Magnetic Resonance Spectroscopy (MRS), Magnetic Resonance Elastography (MRE), Diffusion Weighted Imaging (DWI) and Diffusion Tensor Imaging (DTI) play a significant role in breast cancer diagnosis. The physical basis of the last two techniques is the Brownian motion of water molecules within tissue, as quantified by Apparent Diffusion Coefficient (ADC, [mm2/s]) values of various tissues.
In the last years, numerous studies have reported that DWI is useful for differentiation of malignant from benign lesions. It has been reported lower ADC values in malignant breast tumors than in benign or normal tissues [Kim SH. et al. 2009, Woodhams R. et al. 2010, Kinoshita T. et al. 2010, Jeh SK. et al. 2011, Partidge SC. et al. 2015], as diffusion of water molecules within breast tissue lesions is affected by tissue cellularity, fluid viscosity, membrane permeability and blood flow.
Besides diagnosis, DWI has been reported for prognosis, a complement to diagnosis, reporting tumor aggressiveness, patients’ outcome and response to therapy, by analyzing the correlation between ADC values and several breast cancer biomarkers, such as Estrogen Receptor expression status (ER), Progesterone Receptor expression status (PR), proliferation kinase index (Ki-67) and Human Epidermal growth factor Receptor 2 (HER2) status[Razek AA et al. 2010, Jeh SK et al. 2011, Choi SY et al. 2012, Molinari C et al. 2015, Park SK et al.2015].
The majority of the above mentioned studies, reported on mean ADC values extracted from Regions Of Interest (ROIs) sampling of the most active lesion areas avoiding necrotic, cysts or hemorrhagic regions of the lesions.
The aim of the current thesis is the quantitative evaluation of image texture of breast lesions, as they appear in DWI and investigation of image features correlations to widely accepted breast cancer biomarkers. Entire lesion-based image histogram analysis is
VI
considered in this study, in an effort to assess ADC histogram features contribution in diagnosis and prognosis tasks.
A dataset consisting of 62 histologically verified breast lesions, originating from 55 female patients with mammographically and/or ultrasonographically detected breast malignant lesions, is analyzed. Images were acquired with a 3.0 T MR scanner (Signa HDx GE Healthcare, Milwaukee, WI, USA) in the axial plane and a dedicated bilateral fourelement two-channel, phased array breast coil and DWI sequence with sensitizing diffusion gradients was applied in three orthogonal directions (x, y, z) with b values of 0 and 900 s/mm2.
Correlations of lesion histogram-based image features of a parametric ADC map, generated by weighting b0 and b900 DW acquisitions, are investigated with respect to histological subtype, histological grade, ER, PR and HER2 overexpression/ underexpression status (ER+/ER- , PR-/PR+, HER2+/HER2-), in the frame of identifying potential tumor diagnosis and prognosis imaging biomarkers. Specifically, as the dataset has not normal distribution, only the median value of the lesion histogram ADC features investigated: Mean, Standard Deviation, Skewness, Kurtosis, Entropy, Minimum, Percentile 25th, Percentile 50th, Percentile 75th, Maximum and Range.
Correlation analysis employs the Kruskall- Wallis H Test in case of experiments evolving multiple comparisons, while the Mann- Whitney U test is employed in case of two groups (paired) comparisons.
Five experiments were designed and carried out on the basis of the available data, aiming to identify the potential prognostic value of the ADC histogram features, in terms of differentiation of: (1) histological subtype, (2) histological grade, (3) Estrogen Receptor expression status, (4) Progesterone Receptor expression status and (5) Human Epidermal Growth Factor Receptor 2 expression status.
Histological subtypes that were investigated include three categories: benign breast lesions, Invasive Ductal Carcinomas (IDCs) and Invasive Lobular Carcinomas (ILCs). Histological grading of IDCs includes: Grade I, Grade II and Grade III, according to Nottingham Grading System. Finally expression of ER, PR and HER2 is binary either positive rated or negative rated.
Statistical analysis of histological types and ADC features resulted in no significant differentiation, in accordance to published study [Choi SY et al. 2012].
VII
In case of IDC grade differentiation, the sample median value of the feature ADC mean was 1.034 x10-3 mm2/s, 0.869 x10-3 mm2/s and 0.971 x10-3 mm2/s in case of grade I, grade II and grade III lesions, respectively. The decrease of mean ADC values between grades I and II, is in accordance to reported studies implying an inverse relationship between ADC values and histologic grades [Razek AA et al. 2010, Gouhar G. et al. 2011, Molinari C. et al. 2015], while the increase of mean ADC between grades II and III, seems to agree with studies reporting no corresponding associations. A significant trend was identified for the sample median values of the features ADC skewness (grade I: 0.044, grade II: 0.394, grade III: 0.936) and ADC kurtosis (grade I: 0.176, grade II: 1.165, grade III: 1.289) in terms of grade differentiation.
ER positive lesions demonstrated increased sample median value of ADC entropy (5.828), reflecting increased ADC heterogeneity, as compared to ER negative ones (5.480) highlighting the potential of this ADC feature in ER status differentiation.
PR positive lesions demonstrated increased sample median value of ADC entropy (5.842), also reflecting increased ADC heterogeneity, as compared to PR negative ones (5.486). Results of the PR status differentiation experiment are similar to the ER one, as expected, since ER is predominately expressed over PR status.
Finally HER2neu status differentiation experiment has demonstrated no statistically significant association of ADC features, in accordance to Choi SY et al. 2012.
In conclusion, results of the current study suggest the contribution of texture analysis methods in Diffusion-weighted breast imaging for the quantification of lesion tissue heterogeneity, providing complementary information for breast cancer prognosis. Histogram analysis of breast lesion ADC values demonstrate potential for differentiating highly aggressive breast carcinomas and identify prognostic imaging biomarkers. Worth to mention results of this study are that the skewness and kurtosis are parameters capable of distinguishing highly aggressive IDC carcinomas, whereas tumor heterogeneity, captured by entropy, may contribute in both ER and PR status differentiation.
Finally, future efforts will focus on investigating the correlation of extracted texture features, also considering second order texture, to additional prognostic factors, such as Lymph node status and proliferation index (Ki-67), as well as molecular subtypes.
|
|---|
