Towards label-free liquid biopsy: combining machine learning and tomographic phase imaging flow cytometry for the identification of tumor cells

Liquid biopsy is an emerging and promising biomedical tool that aims to the early cancer diagnosis and the definition of personalized therapies in non-invasive and cost-effective way, since it is based on the blood sample analysis. Several strategies have been tested to implement an effective liquid biopsy system. Among them, searching of circulating tumor cells (CTCs) released by the tumor into the bloodstream can be a valid solution. Within a blood sample, CTCs can be considered as rare cells due to their extremely low percentage with respect to white blood cells (WBCs). Therefore, a technology able to perform an advanced single-cell analysis is requested for implementing a CTCs-based liquid biopsy. Recently, tomographic phase imaging flow cytometry (TPIFC) has been developed as a technique for the reconstruction of the 3D volumetric distribution of the refractive indices (RIs) of single cells flowing along a microfluidic channel. Hence, TPIFC allows collecting large datasets of single cells thanks to the flow-cytometry high-throughput property in 3D and quantitative manner. Moreover, TPIFC works in label-free modality as no exogenous marker is employed, thus avoiding the limitations of marker-based techniques. For this reason, here we investigate the possibility of exploiting the 3D dataset of single cells recorded by TPIFC to feed a machine learning model, in order to recognize tumor cells with respect to a background of monocytes, which are the most similar cells among the WBCs in terms of morphology. Reported results aim to emulate a real scenario for the label-free liquid biopsy based on TPIFC.