Automated characterization of Tumor-Infiltrating Lymphocytes (TIL) in histological breast images

Thumbnail Image
Publication date
Defense date
Journal Title
Journal ISSN
Volume Title
Google Scholar
Research Projects
Organizational Units
Journal Issue
Cancer illness has a big influence on society. Its extended proliferation and high aggressiveness make it a difficult problem to solve and therefore a big deal for science. Recently, a research trend has been focusing on how 3D tumor structure affects the development of the cancer and its outcome, especially metastasis. Stromal structure and tumor cell signaling are processes that highly influence tumor migration. Thus, histological analysis becomes a fundamental tool to study tumor structure, which provides valuable information about cell characteristics and organization. The relevance of histological study is supported by the increasing interest of anatomopathologists to have good automatic solutions to support the specialist’s diagnosis. For this purpose, the current thesis proposes an automated approach to analyze hematoxylin and eosin (H&E) stained histological images, particularly coming from breast cancer patients. The proposed method consists on the classification of the nuclei in H&E-stained histological images and the further analysis of tumor-infiltrating lymphocytes (TIL) present on the visualized section. The starting point of the approach is the automatic nuclei-segmented binary mask. Each of the segmented nuclei is classified into two types, cancerous or healthy. The classification is performed by a trained artificial neural network to give two binary masks, each of them containing one type of nuclei. Then, the algorithm can follow two different paths: classification of zones or TIL analysis. Classification of zones has the aim to provide a more comfortable support to perform cancer diagnosis, because it provides quantitative information of tumor lobule size. To achieve it, a nuclei correction step is executed, by which each nucleus class depends on the area surrounding it. In this way, a clearer vision of the existing zones is provided (tumor lobule or tumor microenvironment). The other approach is to perform TIL analysis. This technique is based on the nuclei classified binary masks and analyzes the immune system response against the tumor. This way, healthy cells of tumor microenvironment are detected and quantified. The ratio of TIL occupied area to free microenvironment area is computed as informational parameter. This ratio is calculated by the combination of a manually-segmented zone binary mask and the nuclei classified binary mask. In this way, only healthy nuclei of microenvironment zone are considered, dividing the sum of their area by the free sections of the microenvironment zone (i.e. area of microenvironment zone where nuclei are not present). Moreover, the TIL dispersion factor is computed to study their distribution throughout the area by dividing the microenvironment area in several zones and calculate the standard deviation of the area of lymphocytes within each of them. Afterward, the opposed of standard deviation is computed to obtain the dispersion factor. Automatic results are found to match the gold standard (the pathologist’s diagnosis), although some error is observed after evaluation. The approach taken in this work has a positive outlook, even though some aspects need to be polished, like the algorithm accuracy and the use of a larger set of images to claim a proper functionality for global cases.
Cancer, Histological imaging, Color deconvolution, Nuclei segmentation, Tumor-infiltrating lymphocytes (TIL)
Bibliographic citation