This poster is published under an
open license. Please read the
disclaimer for further details.
Keywords:
Breast, Oncology, MR, MR-Diffusion/Perfusion, Computer Applications-Detection, diagnosis, Treatment effects, Cancer
Authors:
N. Michoux1, L. Bollondi2, A. Depeursinge3, A. Geissbuhler2, L. Fellah1, H. Müller3, I. Leconte1; 1Brussels/BE, 2Geneva/CH, 3Sierre/CH
DOI:
10.1594/ecr2015/B-1088
Conclusion
The main result of this study is that a multi-parametric model based on textons only,
i.e.
without the additional contribution of morphologic,
biologic or DCE-MRI parameters,
was able to predict non-response to NAC with a good performance level.
Texture analysis allows assessing the spatial distribution of the grey levels in the MR image of which distribution results from underlying structural properties of tissues affected by the disease processes; a concept which has been validated by histopathological analysis 16.
The usefulness of pre-NAC texture parameters in predicting response to NAC has been proven already but based on 2D analysis of breast MR images 17.
In a pilot work,
we combined kinetic and texture parameters extracted from a single subtracted MR image showing the largest area of the breast lesion with a high enhancement.
Using k-means clustering as statistical classifier,
a predictive model relying on 4 parameters (1 GLCM,
2 RLM,
1 kinetic) was found to perform with Se = 84% and Sp = 62% 18.
The predictive accuracy of the present 3D analysis is superior to that of 2D analysis (76% vs 68%).
However,
the gain in performance remains modest.
While a predictive model based on textons only improves the practicality of the analysis,
the 3D segmentation of breast lesions lengthened the processing time of MR images substantially.
These preliminary results warrant further investigations.
Especially,
testing alternative texture analysis techniques (multiple frequency scales 19,
S-transform 20),
exploring different and larger combinations of textons with BI-RADS,
kinetic and/or biologic parameters (Ki67>14%,
HR-/HER2+),
using other machine learning methods (since other types of classifiers than those tested in this study can be implemented,
with a possible impact on the performance of the model) may help improve the predictive performance,
and reach a definitive conclusion on the clinical practicality of texture analysis.
The rationale behind these investigations is the development of a computer-assisted solution based on the texture analysis of MR images that may contribute to an appropriate treatment outcome for patients with breast cancer initially eligible for NAC.