Deep Learning to Improve Optical Microscopy

The new technique uses deep learning to produce high-resolution pictures from lower-resolution microscopic images. (Source: UCLA, Ozcan Research Group)
Read more at: https://phys.org/news/2017-11-ucla-deep-reconstruct-holograms-optical.html#jCp

Deep learning is one of the key techno­logies behind recent advances in applications like real-time speech recog­nition and automated image and video labeling. The approach, which uses multi-layered artificial neural networks to automate data analysis, also has shown signi­ficant promise for health care: It could be used, for example, to auto­matically identify abnor­malities in patients’ X-rays, CT scans and other medical images and data. UCLA researchers report that they have developed new uses for deep learning: recon­structing a hologram to form a micro­scopic image of an object and improving optical micro­scopy.

Their new holo­graphic imaging technique produces better images than current methods that use multiple holo­grams, and it’s easier to imple­ment because it requires fewer measure­ments and performs compu­tations faster. The research was led by Aydogan Ozcan, an associate director of the UCLA Cali­fornia Nano­Systems Institute and by post­doctoral scholar Yair Rivenson and graduate student Yibo Zhang, both of UCLA’s electrical and computer engi­neering department.

For one study, the researchers produced holograms of Pap smears, which are used to screen for cervical cancer, and blood samples, as well as breast tissue samples. In each case, the neural network learned to extract and separate the features of the true image of the object from undesired light inter­ference and from other physical byproducts of the image recon­struction process. “These results are broadly applicable to any phase recovery and holo­graphic imaging problem, and this deep-learning–based framework opens up myriad oppor­tunities to design funda­mentally new coherent imaging systems, spanning different parts of the electro­magnetic spectrum, including visible wave­lengths and even X-rays,” said Ozcan.

Another advantage of the new approach was that it was achieved without any modeling of light–matter inter­action or a solution of the wave equation, which can be challen­ging and time-consuming to model and calculate for each indi­vidual sample and form of light. “This is an exciting achievement since tradi­tional physics-based hologram recon­struction methods have been replaced by a deep-learning–based compu­tational approach,” Rivenson said.

For the second study the researchers used the same deep-learning framework to improve the resolution and quality of optical micro­scopic images. That advance could help diagnos­ticians or patho­logists looking for very small-scale abnor­malities in a large blood or tissue sample, and Ozcan said it represents the powerful opportunities for deep learning to improve optical micro­scopy for medical diag­nostics and other fields in engi­neering and the sciences. (Source: UCLA)

Reference: Y. Rivenson et al.: Phase recovery and holographic image reconstruction using deep learning in neural networks, Light: Sci. & App., online13 Ocotober 2017; DOI: 10.1038/lsa.2017.141 • Y. Rivenson et al.: Deep learning microscopy, Optica 4, 1437 (2017); DOI: 10.1364/OPTICA.4.001437

Link: Ozcan research group, Electrical and Computer Engineering Dept., University of California, Los Angeles, USA

Speak Your Mind

*