Reply to: Breast cancer subtypes may obscure robust classification of breast tumors with brain metastases from other cancers
by Rafael Ortiz-Ramón, Andrés Larroza, Silvia Ruiz-España, Estanislao Arana, David Moratal1 (dmoratal@eln.upv.es)
Ortiz-Ramón, R., Larroza, A., Ruiz-España, S. et al. Eur Radiol (2018). https://doi.org/10.1007/s00330-018-5463-6Dear Editor,
First of all, we would like to thank Dr. Altundag for his interest in our work and thoughtful suggestions to improve our conclusions [1].
We agree with the reader that a possible reason for the unsatisfactory results when classifying brain metastases (BM) from breast cancer and melanoma could be related to the different molecular subtypes of the primary breast cancer and their proliferative histopathologic features. However, it is important to consider this statement carefully since some studies indicate that a percentage of molecular subtypes of breast cancer changes when the cancer metastasize to the brain, meaning that it is not possible to assume with certainty that the molecular subtype of the BM is the same that the molecular subtype of the primary breast cancer [2].
Unfortunately, we do not have all the molecular subtypes of breast cancer of the BM to make reliable conclusions about this hypothesis, since BM are not usually biopsied. This interesting hypothesis provided by the reader should be checked and confirmed in further analyses with a larger number of patients with accurate records of the molecular subtypes of cancer. In the present feasibility study, we only focused on analyzing the capability of texture features to differentiate BM by their primary site of origin on the whole, without considering the specific subtypes of cancer.
Furthermore, at this point it is important to reflect on the correlation between texture features and histopathologic features. Based on our results we only can affirm that the texture features tested in this work could not capture image heterogeneity differences between BM from breast cancer and melanoma. We believe that introducing and combining other texture features or exploring other imaging techniques could solve this problem. However, there could be other deeper reasons related to the pathological profile of the BM behind this failure, as stated by the reader.
Texture analysis quantifies the intrinsic heterogeneous properties that are usually imperceptible to the human eye, but it is still not clear if these heterogeneous properties reflect the biologic profile of the tissue. Although the information obtained from medical images is at the tissue level and cannot substitute histologic findings, the extracted textural parameters are believed to reflect pathophysiologic aspects of the tissue under examination [3].
Some studies have tried to identify the correlation between texture features extracted from medical images and the biologic profile behind the lesion or disease under analysis. The question to answer in these studies is clear: are textural and histopathological parameters linked because they measure the same biology in different ways? [4]. Most of these studies suggest that, whereas a single feature cannot still be directly linked to a specific biologic process, it is possible to assume that a combination of textural parameters may be closely related to underlying pathophysiologic processes [5]. Nevertheless, although several texture features in structural and functional imaging have been shown to characterize diseases or lesions, predict treatment response or be associated with survival, the biologic correlates of texture features are still largely unknown. There is a need to carefully investigate the correlation between texture features from different imaging modalities and histopathologic features that may influence image texture, either in a preclinical model or in humans for each specific application when tissue is available for exhaustive histologic analysis [6–8].
Based on that, we preferred not to make conclusions about the correlation between our results and the histopathologic analysis of the BM since we did not have all the information that an analysis of these dimensions would require. To our knowledge, a genetic link between breast cancer and melanoma is unclear at this point, and the study of this association goes beyond the objective of this work.