Cancer Detection from Histopathology Images

Histopathological analysis of whole-slide images is one of the most widely used techniques for diagnosis of lung cancers. In this study, a fully automated pipeline was developed to detect cancer from histopathology slides of lung tissue. We obtained 1067 histopathology images of lung adenocarcinoma and 1060 images of squamous cell carcinoma from the legacy archive of The Cancer Genome Atlas (TCGA) dataset and used them to test the proposed methodology. At preprocessing step, we trained a classi cation model to detect clinically relevant patches of images using statistical measurements. In the next step, cells and nuclei of the cells were segmented and various texture and morphology features were extracted from images and segmented objects. At the nal step, different classi cation models were applied to distinguish between malignant tissues and adjacent normal cells. The results indicates that the usage of machine learning algorithms at pre-processing step for detecting relevant sections of whole slide images improves the performance of automated cancer detection systems substantially