Virtual Library

Start Your Search

Michelle Chi Kiu Lo



Author of

  • +

    FP12 - Tumor Biology and Systems Biology - Basic and Translational Science (ID 188)

    • Event: WCLC 2020
    • Type: Posters (Featured)
    • Track: Tumor Biology and Systems Biology - Basic and Translational Science
    • Presentations: 1
    • Moderators:
    • Coordinates: 1/28/2021, 00:00 - 00:00, ePoster Hall
    • +

      FP12.04 - Intelligent Label-Free Image-Based Profiling for Lung Cancer Cell Detection and Classification (ID 1930)

      00:00 - 00:00  |  Presenting Author(s): Michelle Chi Kiu Lo

      • Abstract
      • Presentation
      • Slides

      Introduction

      Ample evidence shows that the biophysical properties of cells (e.g. size, mass density, structural complexity etc.) are as effective descriptors of cellular heterogeneity, dysfunctional/malignant changes, compared to the conventional fluorescence markers. Because of its label-free nature, biophysical phenotyping of cells could also augment the adaptability and cost-efficiency of diagnosis. However, its widespread utility, particularly in lung cancer diagnosis/treatment monitoring, requires both high- throughput and high statistical power that are unattainable currently. To close this gap, we present an integrative strategy that extracts high-dimensional biophysical single- cell information and processes with millions of quantitative images at sub-cellular resolution assisted by deep learning. This advance provides sufficient discrimination power to delineate the biophysical phenotypes of lung cancer subtypes, and sensitively detect rare lung cancer cells spiked in human blood in a completely label-free manner.

      Methods

      We developed a microfluidic imaging flow cytometer, called multi-ATOM, which captures single-cell quantitative phase images (QPI) at an ultrahigh-throughput beyond 10,000cells/sec. We parameterized >80 label-free biophysical phenotypes from QPI and molecular-specific fluorescent markers simultaneously on the same cells. Further using deep neural-network with transfer learning, we investigated the feasibility of using these single-cells phenotypic profiles for (I) classification of three major lung cancer subtypes (adenocarcinoma, squamous cell carcinoma and small cell carcinoma) from 7 cell lines (H358, H1975, HCC827, H520, H2170, H526 and H69, all from ATCC) with a total population of >2,000,000cells; (II) detection of rare lung cancer cells (H2170) spiked in the human peripheral blood mononuclear cells (PBMCs) at ratios of 1:1,000, 1:10,000 and 1:100,000.

      Results

      wclc2020_resultsimage.png

      We demonstrated that this profiling strategy achieves the overall accuracy of ~90% to predict the three lung cancer subtypes. We further confirmed that lung cancer subtypes exhibit their distinct label-free phenotypic “fingerprint-like” profiles distilled from 7 cell lines. In the spiked-in tests, this method also achieves high sensitivity of >96% and specificity of 98%. In contrast to the detection method using cell size, our approach with the entire biophysical profile greatly improved the specificity of rare H2170 detection in PBMCs (increased from 89% to 98% at spike ratio of 1:10,000;and from 91% to 98% at spike ratio of 1:100,000.)

      Conclusion

      Massive label-free image-based profiling of single-cell showed the feasibility for rare lung cancer cells detection in blood, and lung cancer subtypes classification with the sensitivity and specificity which are otherwise challenging in the current label-free technologies. It could open a new avenue for exploiting cost-effective methods for cancer diagnosis and treatment monitoring.

      Only Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login, select "Add to Cart" and proceed to checkout. If you would like to become a member of IASLC, please click here.

      Only Active Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login or select "Add to Cart" and proceed to checkout.