Artificial confocal microscopy for deep label-free imaging.

Area of Science:

• Biomedical Optics
• Microscopy
• Computational Imaging

Background:

• Widefield microscopy suffers from reduced contrast in thick specimens due to spatial crosstalk.
• Confocal microscopy offers improved depth resolution but introduces photobleaching and toxicity.
• Existing methods lack non-destructive, high-sensitivity imaging for unlabeled thick samples.

Purpose of the Study:

• To develop artificial confocal microscopy (ACM) for non-destructive, high-resolution imaging of unlabeled specimens.
• To achieve confocal-level depth sectioning, sensitivity, and chemical specificity using computational methods.
• To overcome the limitations of traditional confocal microscopy, including phototoxicity and sample damage.

Main Methods:

• Integrated a quantitative phase imaging module into a laser scanning confocal microscope.
• Trained a convolutional neural network (CNN) to translate phase images into fluorescence images.
• Utilized intrinsically registered phase-fluorescence image pairs for automated data acquisition and training.

Main Results:

• ACM demonstrated significantly enhanced depth sectioning compared to phase imaging alone.
• Recovered confocal-like tomographic volumes of microspheres, neurons, and cancer spheroids.
• Enabled non-destructive, quantitative, and dynamic data acquisition from thick samples.
• Successfully segmented individual nuclei in dense spheroids for cell counting and volume measurements.

Conclusions:

• ACM provides a non-destructive approach to achieve confocal-level imaging performance.
• The computational method enables recovery of chemical specificity and detailed structural information.
• ACM offers a versatile tool for analyzing thick, unlabeled biological specimens with high resolution and sensitivity.