Real-time Image Denoising of Mixed Poisson-Gaussian Noise in Fluorescence Microscopy Images using ImageJ

Mannam, Varun; Zhang, Yide; Zhu, Yinhao; Nichols, Evan L.; Wang, Qingfei; Sundaresan, Vignesh; Zhang, Siyuan; Smith, Cody J.; Bohn, Paul W.; Howard, Scott S.

doi:10.1101/2021.11.10.468102

Cited by 5 publications

(16 citation statements)

References 50 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Besides, our demonstrated model can work with other super-resolution target generation methods like STED/STORM/PALM/SIM. While we evaluated the technique on super-resolution fluorescence microscopy, this approach shows promise for extension to other deep learning based image enhancements (e.g., image denoising networks [10, 43], image super-resolution [26, 44, 45, 46, 47], image segmentation networks [28], and other imaging modalities like X-ray [48, 49, 50] and MRI imaging [51]).…”

Section: Discussionmentioning

confidence: 99%

“…FCNs [62] are used for pixel-wise prediction, e.g. semantic segmentation [39], image denoising [10], super-resolution [37] and low dose computer tomography X-ray reconstruction [63]. Fig.…”

Section: Fcns With Dense Encoder Decodermentioning

confidence: 99%

“…For example, the diffraction limit in fluorescence microscopy has been overcome by a wide variety of super-resolution techniques [4, 5, 6, 7]. To make these technical advances more widely available, machine learning (ML) approaches have been used to estimate super-resolution images obtained from those techniques while using conventional and commonly available imaging platforms [8, 9, 10]. These ML models are powerful and easily distributable; however, they require significantly large training datasets [9, 10] (≥ 10,000 images) that are often prohibitively expensive and time-consuming to generate.…”

Section: Introductionmentioning

confidence: 99%

“…To make these technical advances more widely available, machine learning (ML) approaches have been used to estimate super-resolution images obtained from those techniques while using conventional and commonly available imaging platforms [8, 9, 10]. These ML models are powerful and easily distributable; however, they require significantly large training datasets [9, 10] (≥ 10,000 images) that are often prohibitively expensive and time-consuming to generate. This limitation is especially true for biomedical imaging such as in vivo imaging, MRI imaging, and X-ray [11, 12, 13, 14].…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Small Training Dataset Convolutional Neural Networks for Application Specific Super-Resolution Microscopy

Mannam

Howard

2022

Preprint

Self Cite

View full text Add to dashboard Cite

Machine learning (ML) models based on convolutional neural networks (CNNs) have been used to significantly increase microscopy resolution, speed (signal-to-noise ratio), and data interpretation. The bottleneck in developing effective ML systems is often the need to acquire large datasets to train the neural network. This paper demonstrates how adding a dense encoder-decoder block can be useful to effectively train a CNN that provides super-resolution images from conventional diffraction-limited microscopy images when trained using a small dataset containing 15 field-of-views (FOVs). DenseED blocks use a dense layer that concatenates features from the previous convolutional layer to the next convolutional layer. Demonstrate using DenseED blocks in fully convolutional networks (FCNs) to estimate the super-resolution images when trained with a small training dataset (15 FOVs) of human cells from the Widefield2SIM dataset and the fluorescent-labeled fixed bovine pulmonary artery endothelial cells (BPAE samples). Conventional ML models without DenseED blocks trained on small datasets fail to accurately estimate super-resolution images while models, including the DenseED blocks can. The average resolution and peak signal-to-noise ratio (PSNR) improvements achieved using DenseED blocks in FCNs when trained with 15 FOVs are 2 times and ~3.2 dB, respectively. In addition, we evaluated various configurations of target image generation methods (experimentally captured target and computationally generated target) that are used to train the FCNs with and without DenseED blocks and showed with DenseED blocks outperform compared to simple FCNs without DenseED blocks. Hence, the proposed approach indicates that microscopy applications can use DenseED blocks to train on smaller datasets that are application specific/experimental modality specific imaging platforms such as MRI/X-ray and other in vivo imaging modalities.

show abstract

Section: Discussionmentioning

confidence: 99%

“…FCNs [62] are used for pixel-wise prediction, e.g. semantic segmentation [39], image denoising [10], super-resolution [37] and low dose computer tomography X-ray reconstruction [63]. Fig.…”

Section: Fcns With Dense Encoder Decodermentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Small Training Dataset Convolutional Neural Networks for Application Specific Super-Resolution Microscopy

Mannam

Howard

2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…, exhibited the usage of a specialized software on tissue and tumor images thereby imparting a stable basis for the continuing improvement of tissue imaging software. Fluorescence microscopy imaging pace is essentially restricted through the dimension signal-to-noise ratio (SNR), Mannam et al, (2022). Furthermore, their study revealed that in order to enhance picture SNR for a given picture acquisition rate, computational denoising strategies may be used to suppress noise.…”

Section: Related Workmentioning

confidence: 99%

Deep Learning and Computer Vision: Machine Learning Analysis and Image Processing of Puromycin Treated Microscopy

Owoeye¹,

Ojo-Omoniyi²

2022

Int.J.Curr.Microbiol.App.Sci

View full text Add to dashboard Cite

Digital image processing involves the usage of a functional algorithm to process images with special regions of interest. In most case scenarios, this is termed as an active aspect of digital signal processing; image processing comes with several rewards over analog image processing. Its relevance and application spans Autonomous Vehicles, Biometric fingerprint technologies as well as Face recognition applications. Reliable statistics through feature engineering from the image can be extracted and in turn serve as focus points of deep learning insights. Besides, its application in monitoring Climatic changes, Agricultural crop yields, security measures, industrial manufacturing as well as medical fields exponentially advances each day. Meanwhile, deep learning being a feature of Artificial Intelligence has brought forward several useful models that is being used as transfer base for further model accuracies and baselines. In this study, we make use of a certain Microscopy datasets, sampling one of the images for digital processing, in order to gain useful insights through Cropped Quantizing, Laplace Edge Detection and Gaussian noise with sigma methods respectively. The statistical results of the extracted image features through Support Vector Method (SVM) give accuracy of up to 75%.

show abstract

Improving fluorescence lifetime imaging microscopy phasor accuracy using convolutional neural networks

Mannam,

P. Brandt,

Smith

et al. 2023

Front. Bioinform.

View full text Add to dashboard Cite

Introduction: Although a powerful biological imaging technique, fluorescence lifetime imaging microscopy (FLIM) faces challenges such as a slow acquisition rate, a low signal-to-noise ratio (SNR), and high cost and complexity. To address the fundamental problem of low SNR in FLIM images, we demonstrate how to use pre-trained convolutional neural networks (CNNs) to reduce noise in FLIM measurements.Methods: Our approach uses pre-learned models that have been previously validated on large datasets with different distributions than the training datasets, such as sample structures, noise distributions, and microscopy modalities in fluorescence microscopy, to eliminate the need to train a neural network from scratch or to acquire a large training dataset to denoise FLIM data. In addition, we are using the pre-trained networks in the inference stage, where the computation time is in milliseconds and accuracy is better than traditional denoising methods. To separate different fluorophores in lifetime images, the denoised images are then run through an unsupervised machine learning technique named “K-means clustering”.Results and Discussion: The results of the experiments carried out on in vivo mouse kidney tissue, Bovine pulmonary artery endothelial (BPAE) fixed cells that have been fluorescently labeled, and mouse kidney fixed samples that have been fluorescently labeled show that our demonstrated method can effectively remove noise from FLIM images and improve segmentation accuracy. Additionally, the performance of our method on out-of-distribution highly scattering in vivo plant samples shows that it can also improve SNR in challenging imaging conditions. Our proposed method provides a fast and accurate way to segment fluorescence lifetime images captured using any FLIM system. It is especially effective for separating fluorophores in noisy FLIM images, which is common in in vivo imaging where averaging is not applicable. Our approach significantly improves the identification of vital biologically relevant structures in biomedical imaging applications.

show abstract

Real-time Image Denoising of Mixed Poisson-Gaussian Noise in Fluorescence Microscopy Images using ImageJ

Cited by 5 publications

References 50 publications

Small Training Dataset Convolutional Neural Networks for Application Specific Super-Resolution Microscopy

Small Training Dataset Convolutional Neural Networks for Application Specific Super-Resolution Microscopy

Deep Learning and Computer Vision: Machine Learning Analysis and Image Processing of Puromycin Treated Microscopy

Improving fluorescence lifetime imaging microscopy phasor accuracy using convolutional neural networks

Contact Info

Product

Resources

About