Colorectal cancer detection by immunofluorescence images of circulating tumor cells

Manual detection of ColoRectal Cancers (CRC) is subject to bias and time-consuming. To be objective and efficient, this study proposes a semiautomated method to predict CRC based on the number of Circulating Tumor Cells (CTCs). To count CTCs, this method uses SVM to find possible CTCs using features such as cell size, the ratio of the image area of the nucleus to that of the cytoplasm, and the color intensity distribution of the nucleus and cytoplasm. We propose the wrapping degree to which cytoplasm wraps the nucleus after SVM to significantly improve the results. The area under the receiver operating characteristic curve for CRC prediction performance was 0.78 for manual identification and 0.74 for our semiautomated method. This indicates that semiautomated identification was nearly as accurate in predicting CRC as manual identification. Our semiautomated method only required 20–40 min for detecting 400 CTCs, which was much faster than manual identification.