Resolution enhancement with a task-assisted GAN to guide optical nanoscopy image analysis and acquisition.

Area of Science:

• Biophysics
• Microscopy
• Artificial Intelligence

Background:

• Super-resolution microscopy offers detailed nanostructure imaging in biological tissues.
• Conflicting objectives in microscopy include maximizing resolution while minimizing light exposure.
• Algorithmic post-acquisition approaches aim to improve image resolution and quality.

Purpose of the Study:

• To introduce a novel AI approach, the task-assisted generative adversarial network (TA-GAN), for enhanced biological nanostructure characterization.
• To evaluate TA-GAN's performance in improving generative accuracy across various microscopy modalities.
• To integrate TA-GAN into the microscope acquisition pipeline for automated optimization.

Main Methods:

• Developed and implemented a task-assisted generative adversarial network (TA-GAN) incorporating auxiliary tasks like segmentation or localization.
• Evaluated TA-GAN's generative accuracy against unassisted methods using diverse microscopy data (confocal, bright-field, STED, SIM).
• Integrated TA-GAN into the microscope's acquisition pipeline to predict nanometric content and guide imaging parameter selection.

Main Results:

• TA-GAN demonstrated improved generative accuracy compared to unassisted methods across different imaging modalities.
• The integrated TA-GAN pipeline automatically selected optimal imaging modalities and regions of interest.
• The system predicted nanometric content without requiring a super-resolved image acquisition, reducing light exposure.

Conclusions:

• TA-GAN offers a data-driven microscopy solution to overcome trade-offs in super-resolution imaging.
• This method enables automated optimization of microscopy acquisition sequences, reducing light exposure.
• TA-GAN facilitates the observation of dynamic molecular processes with enhanced spatial and temporal resolution.