Artificial intelligence transforms resolution


Rebecca Pool

Thursday, December 20, 2018 - 20:30
Image: Original image (left) and deep-learning-enhanced image [Ozcan Lab/UCLA]
US-based researchers have developed a method based on artificial intelligence to digitally transform lower resolution fluorescence micrographs into higher resolution versions, as acquired using super-resolution microscopy. 
To achieve this transformation, Professor Aydogan Ozcan from UCLA California NanoSystems Institute, and colleagues, trained an artificial neural network using thousands of image pairs, which comprised low and high resolution images of the same samples.
In this way, the deep neural network was taught cross-modality image transformation from a much simpler and cheaper microscope into a high-end instrument.
According to Ozcan, once the training was complete, the deep neural network could resolve the nanoscale features in a lower resolution image, mimicking the performance of a much more advanced instrument.
The technique transforms low-resolution images from a fluorescence microscope (a) into super-resolution images (b) that compare favourably with those from high-resolution equipment (c). Images show sub-cellular proteins within a cell, and different panels correspond to different observation times. [Ozcan Lab at UCLA].
As he writes in Nature Methods: "Using this framework, we improve the resolution of wide-field images acquired with low-numerical-aperture objectives, matching the resolution that is acquired using high-numerical-aperture objectives."
"We demonstrate cross-modality super-resolution, transforming confocal microscopy images to match the resolution acquired with a stimulated emission depletion microscope," he adds. "We also demonstrate that total internal reflection fluorescence microscopy images of subcellular structures within cells and tissues can be transformed to match the results obtained with a TIRF-based structured illumination microscope."
The technique also allows imaging of dynamic events at nanoscale over a much larger sample volume, while reducing the toxic effects of illumination on living organisms and cells.
Original image (left), deep-learning-enhanced image (centre) and for comparison, a super-resolution image (right). [Ozcan Lab/UCLA]
Having matched the imaging resolution of advanced fluorescence nanoscopy tools using simpler microscopes, Ozcan and colleagues believe this nanoscopic image transformation framework builds bridges across different imaging modalities and instruments, 
"The deep network rapidly outputs super-resolved images, without any iterations or parameter search," says Ozcan. "This might help to democratize super-resolution imaging by enabling new biological observations at nanoscale beyond well-equipped laboratories and institutions."
Research is published in Nature Methods.
Website developed by S8080 Digital Media