Cycle-consistent deep learning approach to coherent noise reduction in optical diffraction tomography

Choi, Gunho; Ryu, Donghun; Jo, YoungJu; Kim, Young Seo; Park, WeiSun; Min, Hyun‐Seok; Park, YongKeun

doi:10.1364/oe.27.004927

Cited by 61 publications

(32 citation statements)

References 63 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our deep convolutional neural network is built on a cycle-generative adversarial network (cycle-GAN) architectural design [36], which has been used in medical image segmentation [37] and coherent noise reduction [22]. Our CNN is a cycle network, which contains two mirror symmetric generative adversarial networks (GAN).…”

Section: Architecture Of Cnnmentioning

confidence: 99%

“…In recent years, deep learning with CNN becomes a powerful tool to solve inverse problem in various optical imaging fields, including scattered image recovery [12], [13], wavefront sensing [14], [15], super-resolution [16], [17], fluorescence microscopy [18], [19], noise reduction [20]- [22] and phase recovery [23]. Since the reconstruction of DH can be also regarded as an inverse problem, the deep learning has been introduced into DH.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Digital Holographic Reconstruction Based on Deep Learning Framework With Unpaired Data

Yin¹,

Gu²,

Zhang³

et al. 2020

IEEE Photonics J.

View full text Add to dashboard Cite

Convolutional neural network (CNN) has great potentials in holographic reconstruction. Although excellent results can be achieved by using this technique, the number of training and label data must be the same and strict paired relationship is required. Here, we present a new end-to-end learning-based framework to reconstruct noise-free images in absence of any paired training data and prior knowledge of object real distribution. The algorithm uses the cycle consistency loss and generative adversarial network to implement unpaired training method. It is demonstrated by the experiments that high accuracy reconstruction images can be obtained by using unpaired training and label data. Moreover, the unpaired feature of the algorithm makes the system robust to displacement aberration and defocusing effect.

show abstract

Section: Architecture Of Cnnmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Digital Holographic Reconstruction Based on Deep Learning Framework With Unpaired Data

Yin¹,

Gu²,

Zhang³

et al. 2020

IEEE Photonics J.

View full text Add to dashboard Cite

show abstract

“…Noises can be induced by many resources, such as dust particles on the beam path, imperfections of various surfaces, multiple reflections from optical components, or misalignment of optical elements, etc. Thus, unexpected noises are formed in interference fringe patterns and speckle grains 35 . To quantitatively examine the background noise, the noise level was defined as the standard deviation (S.D., σ) of intensities of background, except for the particle signal.…”

Section: Generated Hologram Qualitymentioning

confidence: 99%

Deep learning-based hologram generation using a white light source

Lee

You

et al. 2020

Sci Rep

View full text Add to dashboard Cite

Deep learning-based hologram generation using a white light source taesik Go 1 , Sangseung Lee 2 , Donghyun You 2 ✉ & Sang Joon Lee 1 ✉ Digital holographic microscopy enables the recording of sample holograms which contain 3D volumetric information. However, additional optical elements, such as partially or fully coherent light source and a pinhole, are required to induce diffraction and interference. Here, we present a deep neural network based on generative adversarial network (GAN) to perform image transformation from a defocused bright-field (BF) image acquired from a general white light source to a holographic image. Training image pairs of 11,050 for image conversion were gathered by using a hybrid BF and hologram imaging technique. The performance of the trained network was evaluated by comparing generated and ground truth holograms of microspheres and erythrocytes distributed in 3D. Holograms generated from BF images through the trained GAN showed enhanced image contrast with 3-5 times increased signalto-noise ratio compared to ground truth holograms and provided 3D positional information and light scattering patterns of the samples. The developed GAN-based method is a promising mean for dynamic analysis of microscale objects with providing detailed 3D positional information and monitoring biological samples precisely even though conventional BF microscopic setting is utilized. Conventional bright-field (BF) microscopy using a white light source is a widespread imaging technique used in various areas, including industrial, biological, and medical fields. BF images have been commonly used to observe microscale objects, such as emulsions, microorganisms, and biological samples 1-5. Detection or diagnostic sensitivity is not high (usually less than 90% accuracy), and 3D monitoring of samples in a large volume is especially limited because BF images only provide two-dimensional (2D) amplitude information within the shallow depth of focus (DOF). Digital holographic microscopy (DHM) is a powerful imaging technique that encrypts the 3D information of a test sample into a single shot of 2D interference patterns (i.e., hologram). DHM has a deep observation depth that can overcome the technical limitations of BF microscopy by using a coherent light source. Therefore, DHM has been used in various fields including accurate biological sample monitoring 6-14 , environmental monitoring 15-17 , and particle or cell dynamic analysis 18-20 , because amplitude and phase information can be noninvasively obtained from the hologram without any labeling and mechanical scanning procedures. DHM systems with off-axis configuration can directly obtain phase information related to 3D morphological, mechanical, and biochemical properties 21. However, a complicated interferometric experimental setup, such as common-path and Mach-Zhender setups, is required in such systems. Digital in-line holographic microscopy (DIHM) has a relatively simple optical setup, that is, it only requires a single beam path. DIHM has been widely used in inve...

show abstract

“…Another recent work has successfully utilized a generative adversarial network (GAN) for reducing dynamic speckle noise in diffraction tomography images (Fig. 4d), using unregistered pairs of input and label images during the training process 27 .…”

Section: Introductionmentioning

confidence: 99%

Deep learning in holography and coherent imaging

2019

View full text Add to dashboard Cite

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 techniques have overcome some of the challenges associated with existing holographic image reconstruction methods while also minimizing the hardware requirements of holography. These recent advances open up a myriad of new opportunities for the use of coherent imaging systems in biomedical and engineering research and related applications.

show abstract

Cycle-consistent deep learning approach to coherent noise reduction in optical diffraction tomography

Cited by 61 publications

References 63 publications

Digital Holographic Reconstruction Based on Deep Learning Framework With Unpaired Data

Digital Holographic Reconstruction Based on Deep Learning Framework With Unpaired Data

Deep learning-based hologram generation using a white light source

Deep learning in holography and coherent imaging

Contact Info

Product

Resources

About