DeepHCS: Bright-Field to Fluorescence Microscopy Image Conversion Using Deep Learning for Label-Free High-Content Screening

Lee, Gyuhyun; Oh, Jeong-Woo; Kang, Mi-Sun; Her, Nam-Gu; Kim, Myoung-Hee; Jeong, Won-Ki

doi:10.1007/978-3-030-00934-2_38

Cited by 8 publications

(7 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Using the simplest imaging methods for high-content imaging, we offer a general imaging approach for screening 3D tissues. Deep learning approaches to detect cell nuclei have been previously employed for 2D adherent cell cultures 8,9,11,12 or histological tissue slices 10 . Existing neural networks have not yet attempted to infer information from cell images acquired with lower numerical aperture objectives and xy-resolution 9–11 in 3D.…”

Section: Discussionmentioning

confidence: 99%

Label-free imaging of 3D pluripotent stem cell differentiation dynamics on chip

Atwell

Waibel

Boushehri

et al. 2022

Preprint

View full text Add to dashboard Cite

The dynamic chemical and architectural microenvironments of 3D stem cell cultures can be controlled by integration into a microfluidic chip. Massive parallelized 3D stem cell cultures for engineering in vitro human cell types require new imaging methods with high time and spatial resolution to fully exploit technological advances in cell culture. Here, we introduce a label-free deep learning method called Bright2Nuc to predict in silico nuclear staining in 3D from bright-field images obtained using traditional confocal microscopy. Bright2Nuc was trained and applied to several hundred 3D human induced pluripotent stem cell cultures differentiating towards definitive endoderm on a microfluidic platform. Combined with existing image analysis tools, Bright2Nuc segmented individual nuclei from bright-field images, quantified their morphological properties, predicted stem cell differentiation state, and tracked the cells over time. Our methods are available in an open-source pipeline that enables researchers to upscale 3D cell phenotyping in stem cell culture.

show abstract

Section: Discussionmentioning

confidence: 99%

Label-free imaging of 3D pluripotent stem cell differentiation dynamics on chip

Atwell

Waibel

Boushehri

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…The model was trained with the Adam optimizer with a decaying learning rate of 2e −4 for over 50,000 epochs to harness the benefits of heavy elastic deformation on the small annotated datasets. FusionNet has also been translated to PyTorch and pure TensorFlow for other applications, such as Image-to-Image translation (Lee et al, 2018) and MRI reconstruction (Quan et al, 2018). All training and deployment presented here was conducted on a system with an Intel i7 CPU, 32 GB RAM, and a NVIDIA GTX GeForce 1080 GPU.…”

Section: Methodsmentioning

confidence: 99%

FusionNet: A Deep Fully Residual Convolutional Neural Network for Image Segmentation in Connectomics

Quan

Hildebrand

Jeong

2021

Front. Comput. Sci.

Self Cite

105

View full text Add to dashboard Cite

Cellular-resolution connectomics is an ambitious research direction with the goal of generating comprehensive brain connectivity maps using high-throughput, nano-scale electron microscopy. One of the main challenges in connectomics research is developing scalable image analysis algorithms that require minimal user intervention. Deep learning has provided exceptional performance in image classification tasks in computer vision, leading to a recent explosion in popularity. Similarly, its application to connectomic analyses holds great promise. Here, we introduce a deep neural network architecture, FusionNet, with a focus on its application to accomplish automatic segmentation of neuronal structures in connectomics data. FusionNet combines recent advances in machine learning, such as semantic segmentation and residual neural networks, with summation-based skip connections. This results in a much deeper network architecture and improves segmentation accuracy. We demonstrate the performance of the proposed method by comparing it with several other popular electron microscopy segmentation methods. We further illustrate its flexibility through segmentation results for two different tasks: cell membrane segmentation and cell nucleus segmentation.

show abstract

“…In medical image synthesis, a majority of approaches require paired images in their training process, for example when learning to synthesise CT from MR images [18,19,20,21,22], generate MR images of a certain sequence from MR images of another sequence [23,24,25], denoise low-dose CT images [26,27] or perform super-resolution [28]. Cross-modality synthesis is also present in microscopy imaging, where the attempt to reduce time-consuming and laborious tissue preparation results in synthesizing fluorescence images from the bright-field pairs [29]. A consequence of the need for paired images during training is that both reference and synthetic images are also often available for evaluation.…”

Section: Pairwise Comparisonmentioning

confidence: 99%

“…Higher values of PSNR relate to better simulation. The application of PSNR to validate the plausibility of synthetic images is apparent in many papers [21,22,24,25,26,27,28,29,30].…”

Section: Peak Signal-to-noise Ratiomentioning

confidence: 99%

“…Exploiting known characteristics of the human visual system, Wang et al [31] proposed the structural similarity (SSIM), which compares local patterns of pixel intensities that have been normalised for luminance and contrast. SSIM has been widely adopted by the image synthesis community [24,25,27,28,29]. SSIM is computed as a function of three components, luminance, contrast and structure, as follows:…”

Section: Structural Similaritymentioning

confidence: 99%

See 1 more Smart Citation