Self-Supervised Isotropic Resolution Enhancement of Expansion Microscopy via Quantized Compression.

Area of Science:

• Microscopy and Imaging Technologies
• Computational Biology
• Biomedical Engineering

Background:

• Expansion microscopy (ExM) offers nanoscale imaging for disease characterization but faces challenges in whole-organ analysis.
• Current super-resolution techniques struggle with depth-varying aberrations and storage demands in large datasets.
• Existing methods often require high-resolution ground-truth data or assume uniform point spread functions, limiting their applicability.

Purpose of the Study:

• To address resolution anisotropy and storage constraints in whole-organ ExM imaging.
• To develop a self-supervised framework for compression-aware isotropic super-resolution.
• To enable practical, large-scale ExM analysis for clinical biomarker discovery.

Main Methods:

• A single-stage, self-supervised framework combining a 2D lateral encoder and a lightweight volumetric decoder.
• Utilizing a vector-quantized variational autoencoder (VQ-VAE) for an information-sufficient bottleneck.
• Implementing compression-aware isotropic super-resolution directly on raw slices to manage memory limits.

Main Results:

• Achieved up to 128x slice compression and 8x axial resolution enhancement.
• Demonstrated approximately 1000x reduction in storage compared to fully isotropic volumes.
• Validated on human surgical tissues and diverse biological structures across multiple imaging modalities.

Conclusions:

• The proposed framework makes large-scale, whole-organ ExM analysis practical by enabling efficient on-demand isotropic reconstruction.
• It significantly reduces storage demands and improves throughput and scalability compared to prior methods.
• This approach addresses a key bottleneck in translating ExM to clinical biomarker discovery.