Label‐Free Typing of Colorectal Cancer by Optical Time‐Stretch Imaging Flow Cytometry With Multi‐Instance Learning

This is the authors' abstract. We don't add key points for this paper.

Colorectal cancer (CRC) is one of the most prevalent gastrointestinal malignancies, necessitating the study of cellular and molecular changes within the tumor microenvironment. While pathological image analysis remains the gold standard, its labor-intensive nature limits its broad application. This study proposes a label-free CRC typing approach using intelligent optical time-stretch (OTS) imaging flow cytometry combined with multi-instance learning. Specifically, we construct a high-throughput cell image acquisition system by integrating OTS imaging with microfluidic cell focusing, capturing 363 931 cell images from 10 clinical samples. To address cell diversity and heterogeneity, we employ a multi-instance learning framework, which incorporates a multi-level attention mechanism to explore feature interactions at both channel and instance levels. Finally, we apply a majority voting mechanism to enable efficient label-free CRC typing. Our method achieves an accuracy of 85.78% in distinguishing normal and cancerous cells, while encouraging CRC typing performance across all 10 clinical samples.