Morphological Signal Processing for Phenotype Recognition of Human Pluripotent Stem Cells Using Machine Learning...

Area of Science:

• Stem Cell Biology
• Regenerative Medicine
• Bioinformatics

Background:

• Human pluripotent stem cells (hPSCs) are crucial for regenerative medicine due to their self-renewal and differentiation potential.
• Automated selection of high-quality hPSC clones is essential but challenging.
• Label-free, non-invasive methods are needed to assess hPSC colony phenotype.

Purpose of the Study:

• To develop and evaluate machine learning models for classifying hPSC colony phenotypes ('good' or 'bad').
• To identify optimal morphological parameters and classification methods for predicting hPSC quality.
• To establish a foundation for automated hPSC quality control in medical settings.

Main Methods:

• Utilized phase-contrast imaging to extract morphological parameters from hPSC colonies.
• Applied machine learning algorithms including artificial neural networks, logistic regression, and random forest.
• Compared classification accuracy using cellular morphology, colony morphology, and combined parameters.

Main Results:

• Models using cellular morphology achieved 67% accuracy (artificial neural networks).
• Models using colony morphology achieved 75% accuracy (logistic regression).
• Combining cellular and colony morphology with random forest yielded 99% accuracy.

Conclusions:

• Both cellular and colony morphological features are necessary for accurate hPSC phenotype prediction.
• Machine learning models integrating diverse morphological data significantly enhance automated quality control for hPSCs.
• These findings support the development of automated systems for hPSC selection in medical applications.