Human induced pluripotent stem cell formation and morphology prediction during reprogramming with time-lapse...

Area of Science:

• Biotechnology
• Stem Cell Biology
• Artificial Intelligence in Medicine

Background:

• Human induced pluripotent stem cells (hiPSCs) are crucial for regenerative medicine.
• Current methods for hiPSC generation have low efficiency.
• Early detection of reprogramming cells is challenging.

Purpose of the Study:

• To develop a deep learning-based method for early detection of hiPSC-forming cells.
• To predict the colony morphology of potential hiPSCs.
• To improve the efficiency of establishing new hiPSC lines.

Main Methods:

• Utilized time-lapse bright-field microscopy images of reprogramming CD34+ cells.
• Trained Convolutional Neural Network (CNN) models on cell features for classification.
• Employed U-net for cell segmentation and Recurrent Neural Network (RNN) for morphology prediction.

Main Results:

• Achieved 0.8 accuracy in predicting hiPSC formation within 7 days.
• 66% of predicted hiPSC-forming cells successfully formed colonies.
• Demonstrated high accuracy in predicting hiPSC colony areas and image similarity for future growth.

Conclusions:

• A novel deep learning approach efficiently identifies and predicts hiPSC colony development.
• This method enhances the efficiency of hiPSC line establishment.
• The findings offer a promising tool for stem cell research and regenerative medicine.