Ovarian cancer is one of the most lethal gynecological malignancies, frequently accompanied by ascites formation in advanced stages. Accurate identification of ovarian cancer cells within ascitic fluid is clinically important yet technically challenging due to pronounced cellular heterogeneity. Here, we establish a quantitative holographic imaging flow cytometry framework for ovarian cancer cell discrimination under ascites-mimicking conditions using single-cell phase images acquired by microfluidic digital holographic microscopy. A six-cell-type dataset was constructed to emulate the heterogeneous tumor-associated microenvironment, introducing substantial morphological and biophysical overlap. Within this unified experimental setting, we systematically compared multidimensional feature-based machine learning models with end-to-end deep learning approaches to assess their relative performance in cancer cell detection. Deep learning models demonstrated improved robustness and sensitivity in complex backgrounds while preserving high-throughput capability. This study provides a structured evaluation of quantitative phase–driven cell classification and supports the development of rapid, automated, label-free screening strategies for ascites analysis.
Li et al. (2026) studied this question.