Clinically Informed Intelligent Classification of Ovarian Cancer Cells by Label-Free Holographic Imaging Flow Cytometry

Abstract

Liquid biopsy, intended as the detection of circulating tumor cells (CTCs) in hematic specimens, is an emerging tool for both early cancer detection and estimation of prognosis. Herein, the strength of quantitative phase imaging (QPI) is investigated to achieve effective distinction of ovarian cancer (OC) from other blood cell populations based on label-free morphological biomarkers rather than conventional fluorescent imaging or other molecular parameters. At this purpose, QPI is implemented in high-throughput flow cytometry mode and combined with machine learning (ML), reliable and accurate OC cell phenotyping is achieved by developing ad-hoc multi-level ML classification architectures driven by a priori clinical information. It is shown that the latter allows increasing the overall classification accuracy when compared to noninformed ML classification systems. Thanks to its simplicity, the proposed intelligent system is compatible with various clinical applications, particularly in the context of CTC-based liquid biopsy during patient follow-up, when cancer subtype and other clinical information are already known.