Differentiating Engineered Tissue Images and Experimental Factors to Classify Cardiomyocyte Content

Area of Science:

• Biotechnology and Biomedical Engineering
• Stem Cell Biology
• Machine Learning in Healthcare

Background:

• Accurate prediction of cardiomyocyte (CM) content is crucial for scaling up human-induced pluripotent stem cell (hiPSC) differentiation for engineered cardiac tissues.
• Current methods for assessing differentiation often rely on subjective visual inspection or destructive assays, necessitating non-destructive, objective alternatives.
• Machine learning (ML) has shown promise in modeling complex biological processes, including stem cell differentiation.

Purpose of the Study:

• To develop and validate a machine learning model for classifying cardiomyocyte content in hiPSC-derived microspheroids at an early differentiation stage.
• To evaluate the predictive power of non-destructive phase-contrast imaging and tunable experimental parameters for CM content prediction.
• To create an unbiased method for assessing differentiation batch quality, complementing experimenter's visual intuition.

Main Methods:

• Convolutional neural networks (CNNs) were employed to build a binary classifier for predicting sufficient vs. insufficient CM content on differentiation day 10.
• Two datasets were used as input features: (1) phase-contrast images from differentiation days 3 and 5, and (2) images supplemented with experimental parameters (e.g., cell concentration, microspheroid size).
• Model performance was evaluated based on classification accuracy.

Main Results:

• A CNN model using only phase-contrast images achieved 63% accuracy in classifying CM content.
• Incorporating experimental parameters alongside images significantly improved classification accuracy to 85%.
• The best-performing model utilized differentiation day 5 images and experimental features, demonstrating the importance of early-stage data and process parameters.

Conclusions:

• Machine learning, particularly CNNs, can effectively predict cardiomyocyte content in hiPSC microspheroids using non-destructive imaging and experimental data.
• The integration of tunable experimental parameters substantially enhances predictive accuracy, highlighting their role in controlling differentiation outcomes.
• This approach offers a valuable tool for real-time monitoring and quality control in the large-scale production of engineered cardiac tissues.