SHAPR predicts 3D cell shapes from 2D microscopic images

Area of Science:

• Computer Vision
• Biomedical Imaging
• Cell Biology

Background:

• Reconstructing 3D object shapes from 2D images is a key challenge in computer vision.
• Accurate 3D cell shape and size information is crucial for understanding cell function and disease.

Purpose of the Study:

• To develop a novel neural network-based autoencoder, SHApe PRediction (SHAPR), for reconstructing 3D cellular shapes from 2D microscopy images.
• To evaluate SHAPR's performance in reconstructing red blood cell shapes and predicting cell types.
• To assess SHAPR's utility for analyzing cell nuclei in human induced pluripotent stem cells.

Main Methods:

• A neural network-based autoencoder (SHAPR) was developed to predict 3D shapes from single-view 2D confocal microscopy images.
• SHAPR was validated using red blood cells, comparing its reconstruction accuracy against stereological models.
• The method was applied to analyze cell nuclei in human induced pluripotent stem cell aggregates.

Main Results:

• SHAPR demonstrated superior accuracy in reconstructing 3D red blood cell shapes compared to traditional methods.
• Feature-based prediction of red blood cell types improved significantly, with F1-score increasing from 79% to 87.4%.
• SHAPR successfully learned fundamental shape properties of cell nuclei, enabling prediction-based morphometry.

Conclusions:

• SHAPR offers a powerful tool for accurate 3D cellular shape reconstruction from 2D images.
• The method enhances cell type prediction and facilitates high-throughput image-based biomedical applications.
• SHAPR has the potential to reduce imaging time and data storage requirements, optimizing research workflows.