Label-free segmentation of mitochondria for simultaneous morphological and metabolic studies

Area of Science:

• Cell Biology
• Biophysics
• Artificial Intelligence in Imaging

Background:

• Mitochondrial structure and function are crucial for cellular health.
• Traditional fluorescence microscopy for mitochondria is limited by phototoxicity, photobleaching, and staining artifacts.
• Label-free imaging is needed to overcome these limitations.

Purpose of the Study:

• To develop an AI-enhanced imaging platform, RedoxSegNet, for label-free mitochondrial segmentation and analysis.
• To enable concurrent morphological and functional assessment of mitochondria without staining dyes.
• To investigate mitochondrial morpho-functional dynamics in native cellular environments.

Main Methods:

• Utilized high-resolution two-photon excitation fluorescence microscopy.
• Employed a custom-built conditional diffusion model to reconstruct mitochondrial features from NAD(P)H autofluorescence images.
• Applied post-processing algorithms for segmentation, achieving <6% error compared to stained images.

Main Results:

• RedoxSegNet successfully segmented mitochondria from label-free images.
• The platform facilitated mapping of mitochondria-specific optical redox ratios, elucidating metabolic heterogeneity.
• Validated under stress conditions (FCCP), demonstrating capture of mitochondrial fragmentation and metabolic responses.

Conclusions:

• RedoxSegNet provides a non-invasive and reliable method for analyzing mitochondrial dynamics.
• The AI-enhanced platform overcomes limitations of traditional fluorescence imaging.
• Enables detailed investigation of mitochondrial morpho-functional characteristics in cellular contexts.