Human embryonic stem cell classification: random network with autoencoded feature extractor.

Area of Science:

• Stem cell biology
• Biomedical imaging
• Machine learning

Background:

• Automated analysis of human embryonic stem cell (hESC) videos is vital for precise quantification and classification of hESC states and health.
• Applications in regenerative medicine necessitate reliable methods for assessing hESC quality and behavior.

Purpose of the Study:

• To develop an ensemble deep learning model for classifying hESC states from phase contrast microscopy videos.
• To create an automated system for hESC classification, reducing manual annotation time.

Main Methods:

• A deep learning-based random network (RandNet) was developed, incorporating an autoencoded feature extractor.
• The autoencoder was pre-trained on unlabeled data and fine-tuned with annotated data for classifying hESCs into six distinct states.
• An ensemble method and bagging of deep learning classifiers were employed.

Main Results:

• The RandNet approach achieved a high classification accuracy of 97.23% ± 0.94%.
• The proposed method outperformed existing state-of-the-art techniques in hESC classification.
• The approach demonstrated a significantly lower training cost compared to other deep learning methods.

Conclusions:

• RandNet provides an efficient and effective solution for classifying hESC images using a combination of subnetworks trained on both labeled and unlabeled data.
• The developed method can serve as a valuable tool for annotating new videos, saving substantial manual labor.
• This automated approach facilitates the quantified analysis and classification of hESC states for regenerative medicine applications.