Deep Learning for Predicting Spheroid Viability: Novel Convolutional Neural Network Model for Automating Quality...

Area of Science:

• Biotechnology
• Tissue Engineering
• Artificial Intelligence

Background:

• Spheroids are crucial for 3D bioprinted tissue patches but can develop hypoxic cores and necrosis when larger than 500 μm.
• Accurate spheroid viability assessment is essential for fabricating high-viability tissue patches.
• Current viability assays are often time-consuming, labor-intensive, require expertise, and are prone to human bias.

Purpose of the Study:

• To develop a convolutional neural network (CNN) model for efficient and accurate prediction of spheroid viability using phase-contrast imaging.
• To establish a comprehensive dataset of mouse mesenchymal stem cell (mMSC) spheroids with corresponding viability percentages.
• To automate the classification of spheroid viability into predefined categories.

Main Methods:

• A dataset of mMSC spheroids with varying sizes and viabilities (determined by CCK-8 assays) was created.
• A CNN model was trained and validated using this dataset.
• The model was designed to classify spheroid viability into four ranges: 0-20%, 20-40%, 40-70%, and 70-100%.

Main Results:

• The CNN model achieved an average accuracy of 92% in predicting spheroid viability.
• The model demonstrated a consistent loss below 0.2 during training and validation.
• The model successfully classified spheroids into the defined viability categories.

Conclusions:

• The developed deep-learning model provides a non-invasive, efficient, and accurate method for assessing spheroid quality.
• This approach streamlines spheroid quality control, accelerating the development of bioengineered cardiac tissue patches.
• The model holds significant potential for advancing cardiovascular disease therapies through improved tissue engineering.