Label-free classification of cells based on supervised machine learning of subcellular structures.

Area of Science:

• Biomedical Engineering
• Cell Biology
• Machine Learning Applications

Background:

• Accurate cell classification is crucial for disease diagnosis.
• Traditional cell analysis often requires staining, which can be cytotoxic and time-consuming.
• Developing label-free, non-cytotoxic methods for cell analysis is a significant goal.

Purpose of the Study:

• To demonstrate cell classification using machine learning on subcellular structures of unstained live cells.
• To evaluate the efficacy of quantitative phase microscopy (QPM) for label-free cell imaging and analysis.
• To develop an automated method for distinguishing between human white blood cells (WBCs) and cancer cell lines.

Main Methods:

• Quantitative Phase Microscopy (QPM) was used to image unstained live human white blood cells and five cancer cell lines.
• Morphological information, specifically optical thickness, was quantitatively extracted from QPM images.
• Subcellular features were extracted from the images to create training datasets for machine learning algorithms.
• A machine learning classifier was trained and tested on these datasets.

Main Results:

• The developed machine learning classifier achieved high accuracy in distinguishing WBCs from cancer cell lines.
• The area under the ROC curve was 0.996, indicating excellent classification performance.
• The method successfully utilized subcellular structures from QPM images for classification.

Conclusions:

• Label-free, non-cytotoxic cell classification based on subcellular structure is feasible using QPM and machine learning.
• This approach shows potential for automated diagnosis of single cells.
• The technique offers a promising alternative to traditional staining-based cell analysis methods.