Probabilistic spatial analysis in quantitative microscopy with uncertainty-aware cell detection using deep Bayesian...

Area of Science:

• Computational Biology
• Biophysics
• Microscopy Image Analysis

Background:

• Accurate cell identification is crucial for analyzing 3D microscopy data.
• Current deep learning methods for cell detection often lack uncertainty quantification.
• Extracting cell coordinates from density maps limits probabilistic interpretation.

Purpose of the Study:

• To develop a framework for probabilistic cell identification in large-scale 3D microscopy images.
• To integrate deep Bayesian learning for uncertainty-aware density map regression.
• To enable probabilistic spatial analysis for discovering subtle biological patterns.

Main Methods:

• Implemented deep Bayesian learning for uncertainty-aware density map regression.
• Utilized peak detection algorithms for generating cell proposals.
• Developed a mapping from proposals to a probabilistic space for calibrated predictions.
• Applied Monte Carlo sampling for probabilistic spatial analysis.

Main Results:

• The proposed framework successfully outputs probabilistic predictions for cell identification.
• Demonstrated the ability to reveal otherwise undetectable spatial patterns in a bone marrow dataset.
• Achieved accurate cell identification with quantifiable uncertainty.

Conclusions:

• The novel framework enhances cell identification accuracy and provides crucial uncertainty information.
• Probabilistic spatial analysis reveals complex biological patterns missed by traditional methods.
• This approach advances the analysis of large-scale biological imaging data.