Quality evaluation of induced pluripotent stem cell colonies by fusing multi-source features

Area of Science:

• Regenerative Medicine
• Stem Cell Biology
• Computational Biology

Background:

• Induced pluripotent stem cells (iPSCs) hold significant promise for regenerative medicine.
• Accurate quality assessment of iPSC colonies is crucial for their therapeutic applications.
• Current methods for iPSC evaluation can be time-consuming and labor-intensive.

Purpose of the Study:

• To develop an automatic quality evaluation model for iPSC colonies.
• To classify iPSC colonies into 'good', 'medium', and 'bad' quality categories.
• To facilitate efficient and rapid assessment for large-scale iPSC manufacturing.

Main Methods:

• iPSC samples were generated using the Sendai virus reprogramming method.
• Bright-field images of iPSC colonies were processed using adaptive gamma transform and data enhancement.
• Multi-source features were extracted using deep convolutional neural networks (DCNNs) and traditional descriptors, followed by classification with a support vector machine (SVM) after principal component analysis (PCA).
• Quality evaluation was validated using living cell fluorescent staining.

Main Results:

• The proposed automatic model achieved a classification accuracy of 95.55% on a dataset of 46,500 iPSC colony images.
• The ensemble learning approach effectively integrated features from multiple sources for robust classification.
• Principal Component Analysis (PCA) reduced computational cost and training time for the SVM classifier.

Conclusions:

• The developed model offers an efficient and rapid method for judging the biological quality of iPSC colonies.
• This automated approach can support and accelerate the large-scale manufacturing of iPSCs for clinical applications.
• The study demonstrates the potential of multi-source feature ensemble learning in stem cell quality control.