AI-powered single-cell imaging and morphology-guided label-free enrichment reveal tumor heterogeneity, genomic relatedness, and divergence in malignant effusions

AI-based single-cell brightfield analysis identified putative tumor cells in malignant effusions from 10 metastatic breast cancer patients, with median tumor cell fraction 5.4% (range 0.51–15.8%).

• Why it matters: Malignant effusions are accessible for liquid biopsy, but isolating tumor cells is hindered by abundant immune-cell background that complicates molecular characterization.
• What they did: A pre-trained deep learning model extracted morphological features from single cells and guided label-free microfluidic enrichment, then genomic profiling used low-pass whole-genome sequencing (n = 26) and whole-exome sequencing (n = 24).
• The result: Genomic comparisons between enriched tumor fractions, matched effusion cell-free DNA, and solid tumors showed shared alterations supporting tumor origin and relatedness, while analyte differences indicated genomic divergence and heterogeneity.

AI summary of the abstract · may contain errors · not medical advice · read the paper.