Detection of Circulating Tumor Cells in Blood Using Random Forest

Cancer has been the leading cause of death among Japanese since 1981. Recently, Circulating Tumor Cells (CTCs) in the blood have attracted attention as biomarkers of cancer metastasis. Traditionally, CTCs have been detected visually by physicians or by expensive machines. In addition, image processing has been used to detect CTCs, but it has the problem of frequent false positives because the region of interest is limited to only a small portion of the cell. In this paper, we propose a machine-learning-based classification method that focuses on the geometric shapes of cells and changes in brightness values across the entire surface. In the proposed method, multiple features are obtained for four types of cells in blood images: CTCs, Clusters, Normal Cells, and Vertical Cells. Based on the obtained features, cells are classified by Random Forest and their accuracy is evaluated. The effectiveness of the proposed method is demonstrated by comparing it with conventional methods.