Inference in artificial intelligence with deep optics and photonics

Nature, Journal Year: 2020, Volume and Issue: 588(7836), P. 39 - 47

Published: Dec. 2, 2020

Language: Английский

---

Deep neural network models for computational histopathology: A survey

Medical Image Analysis, Journal Year: 2020, Volume and Issue: 67, P. 101813 - 101813

Published: Sept. 25, 2020

Language: Английский

---

DeepSTORM3D: dense 3D localization microscopy and PSF design by deep learning

Nature Methods, Journal Year: 2020, Volume and Issue: 17(7), P. 734 - 740

Published: June 15, 2020

Language: Английский

---

Deep learning in holography and coherent imaging

Light Science & Applications, Journal Year: 2019, Volume and Issue: 8(1)

Published: Sept. 10, 2019

Abstract Recent advances in deep learning have given rise to a new paradigm of holographic image reconstruction and phase recovery techniques with real-time performance. Through data-driven approaches, these emerging overcome some the challenges associated existing methods while also minimizing hardware requirements holography. These recent open up myriad opportunities for use coherent imaging systems biomedical engineering research related applications.

Language: Английский

---

Design of task-specific optical systems using broadband diffractive neural networks

Light Science & Applications, Journal Year: 2019, Volume and Issue: 8(1)

Published: Dec. 2, 2019

Deep learning has been transformative in many fields, motivating the emergence of various optical computing architectures. Diffractive network is a recently introduced framework that merges wave optics with deep-learning methods to design neural networks. Diffraction-based all-optical object recognition systems, designed through this and fabricated by 3D printing, have reported recognize hand-written digits fashion products, demonstrating inference generalization sub-classes data. These previous diffractive approaches employed monochromatic coherent light as illumination source. Here, we report broadband simultaneously processes continuum wavelengths generated temporally incoherent source all-optically perform specific task learned using deep learning. We experimentally validated success architecture designing, fabricating testing seven different multi-layer, systems transform wavefront THz pulse realize (1) series tuneable, single-passband dual-passband spectral filters (2) spatially controlled wavelength de-multiplexing. Merging native or engineered dispersion material deep-learning-based strategy, networks help us engineer light-matter interaction 3D, diverging from intuitive analytical create task-specific components can deterministic tasks statistical for machine

Language: Английский

---

Deep learning-based transformation of H&E stained tissues into special stains

Nature Communications, Journal Year: 2021, Volume and Issue: 12(1)

Published: Aug. 12, 2021

Pathology is practiced by visual inspection of histochemically stained slides. Most commonly, the hematoxylin and eosin (H&E) stain used in diagnostic workflow it gold standard for cancer diagnosis. However, many cases, especially non-neoplastic diseases, additional "special stains" are to provide different levels contrast color tissue components allow pathologists get a clearer picture. In this study, we demonstrate utility supervised learning-based computational transformation from H&E special stains (Masson's Trichrome, periodic acid-Schiff Jones silver stain) using sections kidney needle core biopsies. Based on evaluation three renal pathologists, followed adjudication fourth pathologist, show that generation virtual existing images improves diagnosis several diseases sampled 58 unique subjects. A second study performed found quality generated network was statistically equivalent those through histochemical staining. As into can be achieved within 1 min or less per patient specimen slide, stain-to-stain framework improve preliminary when needed, along with significant savings time cost, reducing burden healthcare system patients.

Language: Английский

---

Quantitative Phase Imaging: Recent Advances and Expanding Potential in Biomedicine

ACS Nano, Journal Year: 2022, Volume and Issue: 16(8), P. 11516 - 11544

Published: Aug. 2, 2022

Quantitative phase imaging (QPI) is a label-free, wide-field microscopy approach with significant opportunities for biomedical applications. QPI uses the natural shift of light as it passes through transparent object, such mammalian cell, to quantify biomass distribution and spatial temporal changes in biomass. Reported cell studies more than 60 years ago, ongoing advances hardware software are leading numerous applications biology, dramatic expansion utility over past two decades. Today, investigations size, morphology, behavior, cellular viscoelasticity, drug efficacy, accumulation turnover, transport mechanics supporting development, physiology, neural activity, cancer, additional physiological processes diseases. Here, we review field biology starting underlying principles, followed by discussion technical approaches currently available or being developed, end an examination breadth use under development. We comment on strengths shortcomings deployment key contexts conclude emerging challenges based combining other methodologies that expand scope even further.

Language: Английский

---

Phase imaging with computational specificity (PICS) for measuring dry mass changes in sub-cellular compartments

Nature Communications, Journal Year: 2020, Volume and Issue: 11(1)

Published: Dec. 7, 2020

Due to its specificity, fluorescence microscopy (FM) has become a quintessential imaging tool in cell biology. However, photobleaching, phototoxicity, and related artifacts continue limit FM's utility. Recently, it been shown that artificial intelligence (AI) can transform one form of contrast into another. We present PICS, combination quantitative phase AI, which provides information about unlabeled live cells with high specificity. Our system allows for automatic training, while inference is built the acquisition software runs real-time. Applying computed maps back QPI data, we measured growth both nuclei cytoplasm independently, over many days, without loss viability. Using method suppresses multiple scattering, dry mass content individual within spheroids. In current implementation, PICS offers versatile technique continuous simultaneous monitoring cellular components biological applications where long-term label-free desirable.

Language: Английский

---

Terahertz pulse shaping using diffractive surfaces

Nature Communications, Journal Year: 2021, Volume and Issue: 12(1)

Published: Jan. 4, 2021

Abstract Recent advances in deep learning have been providing non-intuitive solutions to various inverse problems optics. At the intersection of machine and optics, diffractive networks merge wave-optics with design task-specific elements all-optically perform tasks such as object classification vision. Here, we present a network, which is used shape an arbitrary broadband pulse into desired optical waveform, forming compact passive engineering system. We demonstrate synthesis different pulses by designing layers that collectively engineer temporal waveform input terahertz pulse. Our results direct shaping spectrum, where amplitude phase wavelengths are independently controlled through device, without need for external pump. Furthermore, physical transfer approach presented illustrate pulse-width tunability replacing part existing network newly trained layers, demonstrating its modularity. This learning-based framework can find broad applications e.g., communications, ultra-fast imaging spectroscopy.

Language: Английский

---

Deep learning-enabled virtual histological staining of biological samples

Light Science & Applications, Journal Year: 2023, Volume and Issue: 12(1)

Published: March 3, 2023

Histological staining is the gold standard for tissue examination in clinical pathology and life-science research, which visualizes cellular structures using chromatic dyes or fluorescence labels to aid microscopic assessment of tissue. However, current histological workflow requires tedious sample preparation steps, specialized laboratory infrastructure, trained histotechnologists, making it expensive, time-consuming, not accessible resource-limited settings. Deep learning techniques created new opportunities revolutionize methods by digitally generating stains neural networks, providing rapid, cost-effective, accurate alternatives chemical methods. These techniques, broadly referred as virtual staining, were extensively explored multiple research groups demonstrated be successful various types from label-free images unstained samples; similar approaches also used transforming an already stained into another type stain, performing stain-to-stain transformations. In this Review, we provide a comprehensive overview recent advances deep learning-enabled techniques. The basic concepts typical are introduced, followed discussion representative works their technical innovations. We share our perspectives on future emerging field, aiming inspire readers diverse scientific fields further expand scope applications.

Language: Английский

---