Classification of cell morphology with quantitative phase microscopy and machine learning.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Optical Imaging

Background:

• Accurate cell classification is crucial for biomedical research.
• Label-free quantitative phase imaging offers a non-invasive method for cell analysis.
• Machine learning can enhance the accuracy and efficiency of cell classification.

Purpose of the Study:

• To compare two distinct machine learning approaches for cell classification.
• To evaluate the performance of label-free quantitative phase imaging combined with machine learning.
• To assess the utility of these methods for distinguishing cell states, such as under different gravity conditions.

Main Methods:

• Development of a multilevel integrated machine learning classifier (including artificial neural network, extreme learning machine, and generalized logistic regression).
• Application of a pretrained convolutional neural network using transfer learning.
• Validation using quantitative phase imaging to classify macrophages cultured in normal and microgravity environments.

Main Results:

• The multilevel integrated classifier achieved an average accuracy of 93.1%.
• The convolutional neural network approach attained an average accuracy of 93.5%.
• Both methods demonstrated comparable high performance in cell classification.

Conclusions:

• Machine learning, particularly integrated classifiers and convolutional neural networks, effectively performs cell classification using label-free quantitative phase imaging.
• These approaches offer a valuable tool for biomedical scientists, enabling easy and accurate cell analysis.
• The system is robust and applicable to distinguishing cell states under various conditions, including microgravity.