Neural Networks May Improve Accuracy of Fluorescence Microscopy

Scientists from Technion–Israel Institute of Technology have developed a new method for measuring the 3D distance between two fluorescent spots in cells, using neural networks and point spread function (PSF) engineering. This approach provides a more accurate and efficient method of measuring distance compared to traditional techniques of computing 3D distance between two fluorescent spots.

Fluorescence microscopy is commonly utilized to study cell structure and movement, molecule behavior, and drug effects by labeling specific parts of cells and tissues with glowing molecules. The resolution of optical microscopy, which is limited by the diffraction of light waves passing through a small aperture, can be improved by engineering the point spread function (PSF) to encode useful characteristics of a point source. 

In this case, the authors combined physical models and reconstruction networks to generate a neural network that will help uncover more about encoding and decoding the optimal phase mask for PSF modulation. The authors demonstrated the effectiveness of this method experimentally by using pairs of fluorescent beads and DNA loci in yeast cells.

The benefit of optimizing the imaging system for this form of direct measurement, rather than for separate emitter localizations, was validated through simulation. The results revealed that the distance network outperformed the localization network, primarily as noise levels increased to the range of experimental images.

This method has many potential applications, including single-color distance determination, labeling simplification, and improved signal-to-noise ratio for high-throughput imaging of cell populations.

This study highlights the usefulness of neural networks in task-specific microscopy design and optical system optimization in general. By using neural networks in combination with PSF engineering, the optical system's hardware and software can be designed in an "end-to-end" manner, optimizing both image acquisition and reconstruction. This development can potentially enhance the capabilities of traditional optical microscopes and open new possibilities for studying cellular processes such as migration and division.