Abstract
Identification and enumeration of circulating tumor cells (CTCs) in peripheral blood are proved to correlate with the progress of metastatic cancer and can provide valuable information for diagnosis and monitoring of cancer. Here, we introduce a bright-field image cytometry (BFIC) technique, assisted by a multi-frame image correlation (MFIC) algorithm, as a label-free approach for tumor cell detection in peripheral blood. For this method, images of flowing cells in a wide channel were continuously recorded and cell types were determined simultaneously using a deep neural network of YOLO-V4 with an average precision (AP) of 98.63%, 99.04%, and 98.95% for cancer cell lines HT29, A549, and KYSE30, respectively. The use of the wide microfluidic channel (400 μm width) allowed for a high throughput of 50 000 cells per min without clogging. Then erroneous or missed cell classifications caused by imaging angle differences or accidental misinterpretations in single frames were corrected by the multi-frame correlation analysis. This further improved the AP to 99.40%, 99.52%, and 99.47% for HT29, A549, and KYSE30, respectively. Meanwhile, cell counting was also accomplished in this dynamic process. Moreover, our imaging cytometry method can readily detect as few as 10 tumor cells from 100 000 white blood cells and was unaffected by the EMT process. Furthermore, CTCs from 8 advanced-stage cancer clinical samples were also successfully detected, while none for 6 healthy control subjects. Although this method is implemented for CTCs, it can also be used for the detection of other rare cells.