Iteratively training classifiers for circulating tumor cell detection

Abstract

The number of Circulating Tumor Cells (CTCs) in blood provides an indication of disease progression and tumor response to chemotherapeutic agents. Hence, routine detection and enumeration of CTCs in clinical blood samples have significant applications in early cancer diagnosis and treatment monitoring. In this paper, we investigate two classifiers for image-based CTC detection: (1) Support Vector Machine (SVM) with hard-coded Histograms of Oriented Gradients (HoG) features; and (2) Convolutional Neural Network (CNN) with automatically learned features. For both classifiers, we present an effective and efficient training algorithm, by which the most representative negative samples are iteratively collected to accurately define the classification boundary between positive and negative samples. The two iteratively trained classifiers are validated on a challenging dataset with high performance.