Real-time image denoising of mixed Poisson–Gaussian noise in fluorescence microscopy images using ImageJ

Fluorescence microscopy imaging speed is fundamentally limited by the measurement signal-to-noise ratio (SNR). To improve image SNR for a given image acquisition rate, computational denoising techniques can be used to suppress noise. However, common techniques to estimate a denoised image from a single frame either are computationally expensive or rely on simple noise statistical models. These models assume Poisson or Gaussian noise statistics, which are not appropriate for many fluorescence microscopy applications that contain quantum shot noise and electronic Johnson–Nyquist noise, therefore a mixture of Poisson and Gaussian noise. In this paper, we show convolutional neural networks (CNNs) trained on mixed Poisson and Gaussian noise images to overcome the limitations of existing image denoising methods. The trained CNN is presented as an open-source ImageJ plugin that performs real-time image denoising (within tens of milliseconds) with superior performance (SNR improvement) compared to conventional fluorescence microscopy denoising methods. The method is validated on external datasets with out-of-distribution noise, contrast, structure, and imaging modalities from the training data and consistently achieves high-performance ($><#comment/>8dB$) denoising in less time than other fluorescence microscopy denoising methods.

Authors:
; ; ; ; ; ; ; ; ;
Publication Date:
NSF-PAR ID:
10369404
Journal Name:
Optica
Volume:
9
Issue:
4
Page Range or eLocation-ID:
Article No. 335
ISSN:
2334-2536
Publisher:
Optical Society of America
1. We experimentally demonstrate a camera whose primary optic is a cannula/needle ($diameter=0.22mm$and$length=12.5mm$) that acts as a light pipe transporting light intensity from an object plane (35 cm away) to its opposite end. Deep neural networks (DNNs) are used to reconstruct color and grayscale images with a field of view of 18° and angular resolution of$∼<#comment/>0.4∘<#comment/>$. We showed a large effective demagnification of$127×<#comment/>$. Most interestingly, we showed that such a camera could achieve close to diffraction-limited performance with an effective numerical aperture of 0.045, depth of focus$∼<#comment/>16µ<#comment/>m$, and resolution close to the sensor pixel size (3.2 µm). When trained on images with depth information, the DNN can create depth maps. Finally, we show DNN-based classification of the EMNIST dataset before and after image reconstructions. The former could be useful for imaging with enhanced privacy.
2. An optical parametric oscillator (OPO) is developed and characterized for the simultaneous generation of ultraviolet (UV) and near-UV nanosecond laser pulses for the single-shot Rayleigh scattering and planar laser-induced-fluorescence (PLIF) imaging of methylidyne (CH) and nitric oxide (NO) in turbulent flames. The OPO is pumped by a multichannel, 8-pulse Nd:YAG laser cluster that produces up to 225 mJ/pulse at 355 nm with pulse spacing of 100 µs. The pulsed OPO has a conversion efficiency of 9.6% to the signal wavelength of$∼<#comment/>430nm$when pumped by the multimode laser. Second harmonic conversion of the signal, with 3.8% efficiency, is used for the electronic excitation of the A-X (1,0) band of NO at$∼<#comment/>215nm$, while the residual signal at 430 nm is used for direct excitation of the A-X (0,0) band of the CH radical and elastic Rayleigh scattering. The section of the OPO signal wavelength for simultaneous CH and NO PLIF imaging is performed with consideration of the pulse energy, interference from the reactant and product species, and the fluorescence signal intensity. The excitation wavelengths of 430.7 nm and 215.35 nm are studied in a laminar, premixed–air flame. Single-shot CH and NO PLIF and Rayleigh scatter imaging is demonstrated in a turbulent$CH4−<#comment/>H2−<#comment/>NH3$diffusion flame using a high-speed intensified CMOS camera. Analysis of the complementary Rayleigh scattering and CH and NO PLIF enables identification and quantification of the high-temperature flame layers, the combustion product zones, and the fuel-jet core. Considerations for extension to simultaneous, 10-kHz-rate acquisition are discussed.
3. One of the major challenges in large scale optical imaging of neuronal activity is to simultaneously achieve sufficient temporal and spatial resolution across a large volume. Here, we introduce sparse decomposition light-field microscopy (SDLFM), a computational imaging technique based on light-field microscopy (LFM) that takes algorithmic advantage of the high temporal resolution of LFM and the inherent temporal sparsity of spikes to improve effective spatial resolution and signal-to-noise ratios (SNRs). With increased effective spatial resolution and SNRs, neuronal activity at the single-cell level can be recovered over a large volume. We demonstrate the single-cell imaging capability of SDLFM within vivoimaging of neuronal activity of whole brains of larval zebrafish with estimated lateral and axial resolutions of$∼<#comment/>3.5µ<#comment/>m$and$∼<#comment/>7.4µ<#comment/>m$, respectively, acquired at volumetric imaging rates up to 50 Hz. We also show that SDLFM increases the quality of neural imaging in adult fruit flies.
4. Three-dimensional (3D) refractive index (RI) tomography has recently become an exciting new tool for biological studies. However, its limitation to (1) thin samples resulting from a need of transmissive illumination and (2) small fields of view (typically$∼<#comment/>50µ<#comment/>m×<#comment/>50µ<#comment/>m$) has hindered its utility in broader biomedical applications. In this work, we demonstrate 3D RI tomography with a large field of view in opaque, arbitrarily thick scattering samples (unsuitable for imaging with conventional transmissive tomographic techniques) with a penetration depth of ca. one mean free scattering path length ($∼<#comment/>100µ<#comment/>m$in tissue) using a simple, low-cost microscope system with epi-illumination. This approach leverages a solution to the inverse scattering problem via the general non-paraxial 3D optical transfer function of our quantitative oblique back-illumination microscopy (qOBM) optical system. A theoretical analysis is presented along with simulations and experimental validations using polystyrene beads, and rat and human thick brain tissues. This work has significant implications for the investigation of optically thick, semi-infinite samples in a non-invasive and label-free manner. This unique 3D qOBM approach can extend the utility of 3D RI tomography for translational and clinical medicine.
5. Lightweight and head-mountable scanning nonlinear fiberscope technologies offer an exciting opportunity for enabling mechanistic exploration of ensemble neural activities with subcellular resolution on freely behaving rodents. The tether of the fiberscope, consisting of an optical fiber and scanner drive wires, however, restricts the mouse’s movement and consequently precludes free rotation and limits the freedom of walking. Here we present the first twist-free two-photon fiberscope technology for enabling neuroimaging on freely rotating/walking mice. The technology equips a scanning fiberscope with active rotational tracking and compensation capabilities through an optoelectrical commutator (OEC) to allow the animal to rotate and walk in arbitrary patterns during two-photon fluorescence (TPF) imaging of neural activities. The OEC provides excellent optical coupling stability ($<<#comment/>±<#comment/>1%<#comment/>$fluctuation during rotation) and an extremely high torque sensitivity ($<<#comment/>8mN⋅<#comment/>m$). In addition, the new technology is equipped with a custom grating and prism to effectively manage the temporal properties of the femtosecond excitation pulses through the fiber-optic system, which improved neuroimaging signal by more than$2X$. This TPF fiberscope imaging platform has been tested forin vivoimaging, and the results demonstrate that it enablesmore »