SUMMARY
UCLA researchers in the Department of Electrical and Computer Engineering have developed a recurrent neural network-based volumetric imaging inference framework (recurrent-MZ) to speed up fluorescent microscopic image reconstruction of 3D samples. Recurrent-MZ permits the digital reconstruction of a sample volume over an extended depth-of-field using a few different 2D images to create a volumetric image.
BACKGROUND
3D fluorescent imaging of samples can be achieved with the use of volumetric, optical sectioning of samples using various microscopy techniques. Current modalities require the serial scanning of sample volumes, limiting the imaging speed, temporal resolution, and throughput. Solutions to improve throughput exist but they require complicated microscopy system designs, compromise image quality, or have long post-processing times. Iterative algorithms that aim to solve the inverse 3D imaging problem from lower dimensional projection are time-consuming and require user defined regularization of optimal processes. There is a need for an alternative method that allows the generation of 3D imaging in a rapid, simple, and robust manner.
INNOVATION
UCLA researchers in the Department of Electrical and Computer Engineering developed Recurrent-MZ, a recurrent neural networks (RNNs) method to perform volumetric microscopy 3D measurements using microscopic image reconstruction. The deep learning-based method uses 2D images that are sparsely captured by a standard wide-field fluorescence microscope at arbitrary axial positions within the sample volume. 2D fluorescence information from a few axial planes within the sample is explicitly incorporated to digitally reconstruct the sample volume over an extended depth-of-field. Recurrent-MZ has been successfully demonstrated to increase the depth-of-field of a 63×/1.4NA objective lens by approximately 50-fold with a 30-fold reduction in the number of axial scans required to image the same sample volume. The efficacy of the method permitted fast digital reconstruction of sample volume of multiple fluorescent specimens and demonstrated coverage of various axial permutations and unknown axial positioning errors. The recurrent-MZ method can allow the flexible and rapid volumetric imaging framework, overcoming the limitations of current 3D microscopy tools.
POTENTIAL APPLICATIONS
- 3D fluorescent imaging of samples
- 3D scanning microscopy
- High throughput volumetric sample imaging
- Computer vision
ADVANTAGES
- Fast reconstruction
- Flexible imaging
- Multiple axis coverage
- Generalization of varying imaging conditions
RELATED MATERIALS
DEVELOPMENT TO DATE
First successful demonstration by 3D image of fluorescence C. Elegans and nanobeads samples.