Cell quantification in digital contrast microscopy images with convolutional neural networks algorithm.

Area of Science:

• Biotechnology
• Computational Biology
• Microscopy

Background:

• High Content Screening (HCS) integrates high-throughput methods with cellular imaging for biological analysis.
• Accurate cell quantification is crucial for interpreting HCS data in various biological contexts.

Purpose of the Study:

• To evaluate Convolutional Neural Network (CNN) model performance in identifying cell counts from HCS digital contrast microscopy images.
• To assess the impact of image dataset size and quality on CNN-based cell quantification accuracy.

Main Methods:

• Utilized CNNs for cell counting in digital contrast microscopy images generated by HCS.
• Evaluated model performance using Mean Squared Error (MSE) and correlation analyses across A549, Huh7, and 3T3 cell lines.
• Systematically reduced image data to assess its effect on predictive accuracy.

Main Results:

• CNN models demonstrated varying performance across cell lines, with Mean Squared Error (MSE) ranging from 4,335.99 (A549) to 36,897.03 (3T3).
• Reducing image data significantly increased MSE, indicating a strong dependence on dataset size.
• Positive correlations (R=0.953 for A549, R=0.821 for Huh7) were observed, but no significant correlation was found for 3T3 cells.

Conclusions:

• CNNs show promise for cell quantification in HCS, but performance is sensitive to the quantity and quality of input images.
• Further optimization of CNN models and image acquisition protocols is necessary for robust cell counting across diverse cell types.