Implementation and Optimization of a Random Illumination Microscope: towards Robustness for Microscopy Core Facility

ABSTRACT

Super‐resolution microscopy has become an indispensable tool for investigating molecular architectures in their native cellular environment. However, most super‐resolution techniques face limitations that prevent rapid, deep imaging of live samples. Random Illumination Microscopy (RIM), based on natural laser speckle illumination, is a method of choice to overcome these challenges. RIM combines laser speckle illumination at the optical resolution with an algorithm that exploits the statistical invariance of speckle patterns.

In this approach, a stack of hundreds of random speckle images is acquired using a random diffusive element and then processed to reconstruct the super‐resolved optical section. The invariant statistical properties of speckle patterns, which persist even as they diffuse through biological samples, enable deep‐tissue imaging. Additionally, the wide‐field configuration of both illumination and detection ensures high acquisition speeds and minimal sample photodamage.

Here, we present the implementation of our RIM prototype within a microscopy core facility. We describe the system setup, characterization, and optimization, identifying the key elements required for its reliable operation. As a proof of concept, we also provide biological examples demonstrating the prototype's performance in resolving subcellular structures.

We present the implementation of our RIM prototype within a microscopy core facility. We describe the system setup, characterization, and optimization, and, as a proof of concept, we provide biological examples demonstrating the prototype's performance, here in resolving microvilli brush border intestine in adult C. elegans.

1. Introduction

Among the different super‐resolution methods, single molecule localization (Betzig et al. 2006; Hess et al. 2006; Rust et al. 2006) or stimulated emission depletion (Hell and Wichmann 1994; Klar and Hell 1999) have overcome the diffraction limit enabling researchers to investigate biological structures at length scales as small as 20 nm (Sahl et al. 2017). In contrast, SR‐SIM (Super‐Resolution Structured Illumination Microscopy) (Gustafsson 2000; Gustafsson et al. 2008; Heintzmann and Cremer 1999) presents more modest spatial resolution typically around 100 nm (Heintzmann and Huser 2017). However, this disadvantage is offset by the faster, higher signal‐to‐noise, and gentler imaging of SR‐SIM which makes it the method of choice for imaging live samples in super‐resolution (Wu and Shroff 2018).

In conventional wide‐field SR‐SIM, the main limitation is the imaging depth. As the structured illumination patterns propagate deeper into the sample, scattering, refraction, and optical aberrations degrade the pattern contrast, which compromises the quality of the reconstructed image. Alternative approaches such as Image Scanning Microscopy (ISM) have been developed to address these limitations. Unlike SIM, ISM relies on point‐by‐point (or multi‐point) scanning illumination, and the emitted light is detected with a spatially resolved detector. Each Point Spread Function (PSF) is captured with sub‐diffraction precision, enabling super‐resolution reconstruction (Wu and Shroff 2018). A well‐known commercial implementation of ISM is the Airyscan system from Zeiss, which allows the reconstruction of super‐resolved confocal images. However, because the image is reconstructed pixel by pixel, acquisition speed is significantly reduced compared to wide‐field techniques. Additional method using multifocal array was also developed to overcome this limitation (York et al. 2016).

To overcome the speed limitation of ISM and still be able to work in depth, the original idea was to use the natural speckle of a laser in order to structure the wide‐field illumination, “speckle microscopy”. This approach originates from the pioneering work of Jerome Mertz who used the natural structured illumination of the laser speckle to develop a method for wide‐field fluorescence sectioning (Lim et al. 2008). But in this first work, the simplification of the reconstruction method was not able to give a super‐resolved image, the result only achieved a z‐sectioning. More recently, the group of Anne Sentenac developed a new reconstruction method named Blind‐SIM, enabling to reconstruct super‐resolved images from distorted structured illumination (Mudry et al. 2012; Ayuk et al. 2013). This approach opens the possibility of imaging at depth, although it remains limited when the illumination contrast decreases. Another method was upgraded in order to unveil the central role of the sparsity of the illumination patterns in the mechanism driving super‐resolution (Labouesse et al. 2017). However, these inverse methods were not compatible with life science applications, as they required long computation times. Indeed, they were based on the joint estimation of the illumination function and the object. In addition, they were sensitive to the out‐of‐focus artifacts encountered in thick samples. Since the statistical properties of speckle are insensitive to scattering and aberrations, this approach is expected to enable deeper imaging than classical SIM.

A new approach called Random illumination microscopy (RIM) combines speckle wide‐field illumination and a new statistical approach based on the covariance (Idier et al. 2018), and the variance (Mangeat et al. 2021; Affannoukoué et al. 2023; Labouesse et al. 2024; Mazzella et al. 2024). The method leverages the robustness of the autocorrelation of the speckle (speckle size grain). It is interesting to note that, despite the non‐linear nature of the data processing involved in taking the variance of the speckled images, mathematical demonstrations and experimental evidence have shown that RIM super‐resolved reconstruction is linearly linked to the sample fluorescence density. This results in a lateral resolution gain that is comparable to that of SIM. In addition, RIM exhibits an optical sectioning property comparable to that of an ideal confocal. Finally, this method is fairly universal and can be applied to different fluorescence modalities, in TIRF (Affannoukoué et al. 2023) and extended depth of field (Mazzella et al. 2024). Most importantly, it offers the first version of CARS‐type wide‐field nonlinear imaging (Fantuzzi et al. 2023).

In this manuscript, we present a detailed implementation and optimization of a RIM prototype in the context of a microscopy core facility. We chose to use a Spatial Light Modulator (SLM) to drive the different random patterns of speckle illumination and the Inscoper Imaging Solution to control the different synchronization for microscopy multi‐dimensional acquisition. The AlgoRIM solution (Mangeat et al. 2021) was then used for the processing of RIM stacks in order to determine super‐resolved optical sections of fluorescence images from biological samples.

2. Materials and Methods

2.1. Optical References

 

Resource	Source	Identifier
Ligth Sources
Laser 405 nm	Oxxius	LBX‐405‐100‐CIR‐PP
Laser 488 nm	Oxxius	LBX‐488‐100‐CIR‐PP
Laser 561 nm	Oxxius	LBX‐561S‐150‐COL‐PP
Laser 642 nm	Oxxius	LBX‐642‐130‐CIR‐PP
Optical Elements
Mounted Achromatic Half‐Wave Slide	Thorlabs	AHWP10M‐600
Mounted Achromatic Quarter‐Wave Slide	Thorlabs	AQWP10M‐580
5X Achromatic Galilean Beam Expander, AR Coated: 400 ‐ 650 nm	Thorlabs	GBE05‐A
Polarizing Beamsplitter Cubes	Thorlabs	CCM1‐PBS251/M
Spatial Ligth Modulater: SLM	4D Forth Dimension Displays	QXGA‐R11 Microdisplay System STR
Mounted Achromatic Doublets, AR Coated: 400 ‐ 700 nm	Thorlabs	ACT508‐400‐A‐ML
Broadband Dielectric Mirrors	Thorlabs	BB1‐E02
Opticomechanics
Nikon to Optical Cage Port Adaptater	Thorlabs	SM1A30
30 mm to 60 mm Cage Slide Adapter	Thorlabs	LCP33/M
30 mm to 60 mm Cage Slide Adapter	Thorlabs	LCP02/M
Cage Rotation Mount for Ø1″ Optics	Thorlabs	CRM1T/M
30 mm Cage System Iris Diaphragm	Thorlabs	CP20D
Right‐Angle Kinematic Mirror Mounts	Thorlabs	KCB1C/M
Cage Assembly Rod, 6″ Long	Thorlabs	ER6
Cage Assembly Rod, 3″ Long	Thorlabs	ER3
Cage Assembly Rod, 2″ Long	Thorlabs	ER2
Cage Assembly Rod, 1.5″ Long	Thorlabs	ER1.5
30 mm to 60 mm Cage Slide Adapter	Thorlabs	LCP4S
60 mm Cage Slide Optic Mount	Thorlabs	LCP08/M
60 mm Cage Slide Optic Mount	Thorlabs	LCP06/M
Mounting Adapter	Thorlabs	SM2A21
Microscope
Ti2 widefield microscope	Nikon	Tie2 eclipse
Quad Dio‐R405/488/561/635 dichroic mirror	Nikon	Custom (or Chroma or Semrock equivalent)
Objectives 10 x air 0.25 NA	Nikon	CFI ADL 10xF
Objectives 60 x oil 1.4 NA	Nikon	CFI Lambda S Plan Apo 60x Oil
Objectives 100 x oil 1.4 NA	Nikon	CFI HP Plan Apo VC 100x Oil
Objectives 100 x silicon 1.35 NA	Nikon	SR HP Plan APO 100x Sil
Z piezo plate	Nikon/Mad City Lab	MCL piezo NANO Z200‐N
Emission filter wheel	Nikon	TI2‐P‐FWB‐E Motorized BA Filter Wheel MEE54405
FF01‐434/17‐25 filter	Semrock	Brightline FF01‐434/17‐25
FF01‐520/35‐25 filter	Semrock	Brightline FF01‐520/35‐25
FF01‐600/37‐25 filter	Semrock	Brightline FF01‐600/37‐25
BLP01‐664R‐25 filter	Semrock	EdgeBasic BLP01‐664R‐25
Quad 446/523/600/677 filter	Semrock	BrigthLine FF01‐446/523/600/677‐25
cMOS camera	Hamamatsu	ORCA‐Fusion Digital CMOS camera C14440‐20UP
Microscope control
Inscoper box	Inscoper	Device controller Model M

2.2. System Settings

To characterize system settings, we used only 100x Silicon objective 1.35 NA with Inscoper software (version 8.5.18). All the images were done using 488 nm laser and emission filter FF01‐520/35‐25.

To load checkerboard and noise images, we used Metrocon software. To visualize the checkerboard and the speckles, we used a mirror slide with laser power set at 1.04 W/cm², with 10 ms camera exposure and without emission filter. Fast Fourier Transform analysis were done with FFT plugin in ImageJ. We collected the “r” value along a vertical and horizontal line. To have frequency information in cycle/µm, we done invert of “r” value.

For the visualization of the RimZone (illuminated zone on the camera), we used a “Green” chroma slide. We preset the settings at 0.22 W/cm² with 20 ms camera exposure. Intensity profile of the RimZone was obtained with MetroloJ plugin on ImageJ.

2.3. Acquisition Optimization and Resolution

Microtubule slide were done with U2OS WT cell line. To mark tubulin, we used the Abbelight Smart Staining Kit.

Fast Fourier Transform analysis were done with FFT plugin in ImageJ.

To find the best acquisition parameters, we used a slide of beads PS‐Speck^TM T7284 component a 0.1 µm fluorescent blue/green/orange/dark red form Molecular Probes.

To find system resolution, we used an Argo‐HM slide from Argolight taken from RT‐MFM metrology suitcase.

2.4. Biological Applications

2.4.1. C. elegans Brush Border

Strain was maintained under typical conditions as described (Brenner 1974). We used the strain FL378 with the following genotype erm‐1 (bab59[erm‐1:mNG^3xFlag]) I. For in vivo imaging, C. elegans larvae were mounted as described previously (Bidaud‐Meynard et al. 2021).

For the confocal imaging, we use Zeiss LSM880‐Airyscan equipped with a 63× 1.4 NA oil immersion objective (Zen Black software). The excitation/emission wavelengths for ERM‐1:mNG were 488 and 485/20 nm.

For RIM imaging, we use 100x, 1.35 NA silicon immersion objective. The excitation wavelength for ERM‐1:mNG was 488 nm and the emission filter was FF01‐520/35‐25. For each super‐resolution reconstruction, 400 frames were acquired with a SLM exposure time of 10 ms. Reconstruction parameters were w = 0.025 and r = 0.01.

2.4.2. Mitosis Microtubules

Routine culture of HK2 cells used in this study was performed in DMEM/F12 (Life technologies) supplemented with 10% FBS, 100 µg/mL penicillin, and 100 U/mL streptomycin, and maintained at 37°C in a 5% CO₂ incubator.

We fixed HK2 cell line with Methanol during 6 min à −20°C. Blocking was then performed using PBS1X/BSA 1% for 1 h at room temperature. Cells were incubated overnight at 4°C with the antibody DM1A‐AF488 (MERCK) diluted 1:500 to stain microtubules. Mounting was carried out between slide and coverslip with Mowiol.

For the confocal imaging, we used Zeiss LSM980‐Airyscan equipped with a 63× 1.4 NA oil immersion objective (Zen Blue software). The excitation wavelength for AF488 was 488 nm and the emission filter was SBS SP 615 nm.

For RIM imaging, we used 100x 1.35 NA silicon immersion objective. The excitation wavelength for AF488 was 488 nm and the emission filter was FF01‐520/35‐25. For each super‐resolution reconstruction, 200 frames were acquired with a SLM exposure time of 5 ms. Reconstruction parameters were w = 0.05 and r = 0.1.

2.4.3. Mitochondrial Network

MCF7 (HTB‐22) cells were purchased from the American Type Culture Collection and were kept free from mycoplasma throughout experiments. They were grown in Dulbecco's Modified Eagle's Medium (DMEM) already containing 1% L‐glutamine (Thermo Fisher Scientific), and supplemented with 10% FBS (Thermo Fisher Scientific) and 1% penicillin‐streptomycin (Thermo Fisher Scientific).

Cells were fixed in 4% paraformaldehyde (Euromedex), stained using standard immunocytochemical procedures, and mounted in ProLong Gold Antifade reagent (ThermoFisher Scientific). Mitochondria were stained with a polyclonal rabbit anti‐PMPCB (16064–1‐AP; Proteintech) used at a 1:500 dilution and a secondary anti‐rabbit antibody coupled to Alexa 647 at a 1:5000 dilution (ThermoFisher Scientific). PMPCB‐Alexa 647 was acquired with a Leica SP8 inverted confocal microscope (Leica) driven by the Leica Acquisition Suite (LAS) software and a 63x oil immersion objective (NA 1.4). The excitation/emission wavelengths for Alexa 647 were 633 and 650/20 nm.

SRRF images were acquired on the same sample using an inverted confocal microscope equipped with a spinning disk head (Leica) and coupled to an EMCCD camera. A 63x oil immersion objective was used. Alexa 647 was excited using a 638 nm laser, and emission was collected through a 685/40 nm filter. For each super‐resolution reconstruction, 200 frames were acquired with an exposure time of 50 ms. Reconstruction parameters were optimized using the parameter sweep function of the eSRRF plugin* (radius = 2, sensitivity = 2, magnification = 5x, Average Reconstruction).

For RIM imaging on the same sample, we use 100x, 1.35 NA silicon immersion objective. The excitation wavelength for PMPCB‐Alexa 647 was 642 nm and the emission filter was Quad 446/523/600/677. For each super‐resolution reconstruction, 200 frames were acquired with a SLM exposure time of 5 ms. Reconstruction parameters were w = 0.075 and r = 0.25.

2.5. AlgoRIM

All reconstruction of images were done on AlgoRIM software (Mangeat et al. 2021) (github link: https://github.com/teamRIM/tutoRIM).

To find the right parameters w and r a protocol is written in Supporting Information (user manual).

3. Results

3.1. RIM Microscope System Description

The Random Illumination Microscopy (RIM) system is a wide‐field fluorescence microscope composed of three main components: the optical module for speckle illumination, the microscope setup, and the control system for synchronizing the acquisition of RIM image stacks (Figure 1).

FIGURE 1.

System description. The path first passes through an optical density filter wheel (1), then a half‐wave slide (2a), then a beam expander (3), a polarization splitter cube (4), a second half‐wave slide (2b), a microdisplay called a spatial light modulator (SLM) (5), a quarter‐wave slide (7), an achromatic lens (8) with focal length f = 40 cm, and ends up at the microscope's epifluorescence illumination input.

The optical module is responsible for beam generation, shaping, and speckle formation. It integrates a laser bench equipped with four wavelength sources: 405 nm, 488 nm, 561 nm, and 642 nm. Each laser beam follows the same optical path. To adjust the excitation power, an optical density filter wheel is positioned at the output of the laser bench (Figure 1.1). A first half‐wave plate (Figure 1.2a), combined with a polarizing beam splitter cube (Figure 1.4), directs the laser beam toward the SLM by adjusting its polarization (Figure 1.5), thereby regulating the maximum beam intensity. Before reaching the beam splitter cube, a 5× beam expander (Figure 1.3) widens the laser wavefront to fully cover the active area of the SLM. The SLM is achromatic, making it suitable for wavelengths ranging from 400 nm to 650 nm. On the vertical polarization side of the cube, a second half‐wave plate (Figure 1.2b) rotates the laser polarization 45° clockwise. The beam reflects off the SLM and passes through the half‐wave plate again, rotating the polarization by an additional 45° clockwise, resulting in horizontal polarization. This ensures that the microdisplay is illuminated at normal incidence, preventing back‐reflection and directing the beam toward the microscope. The SLM is positioned in the object focal plane of an achromatic lens (2″ diameter, f = 40 cm). The microscope is aligned so that the back focal plane of the objective is conjugated with the image focal plane of the collimating lens. This configuration produces a collimated beam in the focal plane of the objective, maximizing the illumination field and ensuring a clear projection of the SLM pattern onto the sample plane. A quarter‐wave plate (Figure 1.7) is placed before the lens (Figure 1.8) and the cube (Figure 1.9). It converts the beam to circular polarization, reducing interference effects in the lens' object plane.

The wide‐field microscope is based on a Nikon Ti2 model, equipped with 100× silicon immersion (NA 1.35), 100× oil immersion, 60× oil immersion, and 10× air objectives. The optical module is attached to the rear of the microscope and interfaces through the epifluorescence port. A quad‐band dichroic mirror (Dio‐R405/488/561/635) (Figure 1.9) reflects the excitation lasers onto the sample. Fluorescence emission passes through the dichroic mirror and an emission filter wheel (Figure 1.10), which contains five filters: FF01‐434/17‐25, FF01‐520/35‐25, FF01‐600/37‐25, BLP01‐664R‐25, and Quad 446/523/600/677. Finally, a lens focuses the sample image onto a Hamamatsu ORCA Fusion scMOS camera.The microscope is also equipped with a piezo stage for Z‐axis translation and a motorized stage for X‐Y sample positioning.

The electronic system handles signal processing and control, managed by an on‐board acquisition card (Figure 1.11, Inscoper Imaging Solution). This system is connected to a computer and controls all microscope functions:

• ‐X‐Y‐Z translation.
• ‐Objective, dichroic mirror, and emission filter selection.
• ‐Laser activation/deactivation and power modulation.
• ‐Communication with the SLM controller to modify the illumination wavefront.
• ‐Camera configuration (acquisition parameters, image capture, and storage).

A direct link between the camera and the SLM ensures synchronization between wavefront changes and the end of image acquisition. The Inscoper graphical user interface allows users to:

• ‐Program the illumination mode.
• ‐Display acquired images.
• ‐Sequence acquisitions (see the “User Manual” in the Supporting Information).

The mechanical integration of the SLM is illustrated in Figure 2A. The SLM consists of a microdisplay and an electronic control board, both supplied by Forth Dimension Displays (model QXGA‐R11). The microdisplay is a liquid crystal matrix with a resolution of 2048 × 1536 pixels, covering an active area of 17.843 mm × 13.65 mm. The electronic board (version R11) drives the microdisplay and includes 4 Gb of non‐volatile flash memory (DRAM), capable of storing up to 1,024 1‐bit coded images at full resolution. Uploaded images can be displayed on the microdisplay to modulate the reflected wavefront.

FIGURE 2.

Integration of the SLM micro display in the optical module. (A) Electronic board and SLM support. (B) SLM operating sequence. Adapted from Hamamastu (C) Optical module design. (D) RIM optical bench.

To integrate the SLM into the optical module described in Figure 1, we designed custom mechanical components:

• ‐A housing assembly for the electronic board, allowing it to be suspended vertically using Thorlabs brackets.
• ‐An interface between the microdisplay and a Thorlabs cage mount (Figure 2A).

As shown in Figure 2B the image display mode used for the SLM. The goal is to display successive noise images on the SLM to modify the laser beam wavefront and generate dynamic speckle illumination patterns at the microscope's object plane. A library of 400 noise images is pre‐stored in the electronic board's memory. For each acquisition stack (typically 100 to 200 images), the library is continuously cycled, ensuring that the sequence of images displayed on the SLM can be considered random. The display sequence uses the “1‐bit balanced sequence” mode, which operates as follows:

• ‐A trigger signal is sent to the digital input channel of the electronic board, initializing the display and preparing the liquid crystals to receive an image.
• ‐After a brief exchange time, the positive frame of the image is displayed on the microdisplay. The exposure time is determined by the chosen sequence (2 ms, 4 ms, or 10 ms on our setup).
• ‐After a second exchange time, the SLM displays the negative frame of the image to protect the liquid crystals from damage. The exposure time for the negative frame is identical to that of the positive frame.

The final version of the RIM optical module is presented in Figure 2C. For stability, the module is mounted using Thorlabs 30 mm and 60 mm cage systems. The laser beam enters from the right, passing through:

• ‐A quarter‐wave plate.
• ‐A red beam expander.
• ‐A polarization splitter cube.

This optical assembly is bounded by two irises to align the laser beams and ensure they are coaxial with the beam expander's optical axis. An interface slide attaches to the epifluorescence inlet, securing the optics to the microscope. The cage assembly allows the lens to be positioned at a 40 cm distance from the microdisplay. Upstream, two mirrors mounted on kinematic mounts in a Z‐shape configuration direct the laser beams into the microscope.

As shown in Figure 2D the complete optical bench. The SLM optical module is visible in the foreground. The bench is fully enclosed to protect users from laser radiation. In the background, the laser bench includes:

• ‐A neutral density filter wheel for global control of laser intensity.
• ‐A shutter to control beam flow.
• ‐Two pinholes between the laser bench and the SLM optical module to adjust the injection of laser beams.

As illustrated in Figure 3A the signal exchange protocol between the various devices involved in the RIM system:

• ‐The camera sends a “Trigger Ready” output signal to the Inscoper control box (Figure 3A.1).
• ‐In response, the Inscoper box sends a signal to trigger camera acquisition (Figure 3A.2).
• ‐At the beginning of acquisition, the camera sends a signal to the SLM (Figure 3A.4).
• ‐Finally, the Inscoper box activates the lasers when the camera starts its acquisition (Figure 3A.3).

FIGURE 3.

Communication between the various electronic devices. (A) Exchange flow between the various electronic parts. (B) Camera operating mode and synchronization between each element. Adapted from Hamamastu.

Figure 3B shows the acquisition mode of the camera, a Hamamatsu ORCA Fusion C14440. Among the available operating modes, we selected the “Global Reset Level Trigger” mode, which functions as follows:

• ‐Acquisition is triggered by an external signal (Figure 3B.2).
• ‐The exposure of the CMOS sensor is maintained as long as the signal remains high.
• ‐On the falling edge of the signal, the camera reads out the active surface without further photon accumulation (highlighted in yellow).
• ‐The speckle pattern change is initiated by the camera on the rising edge of the “Global Exposure Time Out” signal, when the active surface begins to accumulate photons (Figure 3B.4).
• ‐Finally, the camera signals the end of the active surface readout on the rising edge of the “Trigger Ready Output” signal.

Figure 3B also shows the timing diagram for synchronizing signals between the four devices during RIM acquisition:

• ‐The camera sends an impulse to the Inscoper box to indicate it is ready to acquire an image.
• ‐The Inscoper box initiates camera acquisition by sending a high signal for the duration of the exposure time.
• ‐Simultaneously, the Inscoper box activates the illumination laser, which is turned off at the end of the exposure time.
• ‐When the scMOS sensor is exposed, the camera triggers the SLM to change the speckle pattern.

3.2. System Settings

To characterize the system, we project a checkerboard image onto the SLM rather than a random noise image. This approach allows us to verify the optical alignment of our system by ensuring the checkerboard lines are as sharp as possible and that the squares do not flash, as shown in Figure 4A.

FIGURE 4.

System settings. (A‐A’) We loaded an image of checkerboard on SLM to characterize the system's alignment by observing the fineness of the lines and the flashing of the checkerboard. (B‐B’) speckles pattern. (C‐C’) Speckles analysis with FFT analysis on image J. We can compare a good system's alignment (A‐B‐C) and a misalignment (A’‐B’‐C’). (D) The FFT analysis image gives us the amplitude per frequency that characterizes our speckles. We can extract two information from this graph: the contrast of our speckles, which corresponds to the ratio between the high plateau and the low plateau, and the size of our speckles, which corresponds to the frequency at which there is a break between the low plateau and the curve. (E) Illumination of a Green Chroma slide to observe the part of the camera illuminated by the SLM (RimZone). (F) Intensity profile of the RimZone.

We also investigated whether misalignment could affect the appearance of the speckles. To evaluate this, we deliberately misaligned the optical path by adjusting the position of the lens (Figure 1.8) and modifying the orientation of the half‐wave plate near the SLM (Figure 4A' and Figure 1.2b). After reintroducing random noise images onto the SLM, we observed the resulting speckle patterns (Figure 4B‐B').

To analyze these patterns, we performed Fast Fourier Transform (FFT) analysis using Fiji/ImageJ software (Figure 4C‐C'). By examining the plot profile along a line on the FFT images, we extracted two key pieces of information (Figure 4D). The first is the speckle contrast, determined by the ratio between the high plateau and the low plateau, where a higher ratio indicates greater contrast. The second is the speckle size, identified by the frequency at which there is a break between the low plateau and the curve, with higher frequencies corresponding to smaller speckles.

We compared four different conditions: a well‐aligned optical path, a misaligned lens position (10 cm for a focal lens of 40 cm), a misaligned half‐wave plate (1/8 turn rotation), and a condition where both the lens position and half‐wave plate were misaligned. Our observations revealed that the position of the lens has a minor effect on the appearance of the speckles, while modifying the half‐wave plate has a more significant impact (Figure 4D). When both parameters were misaligned, the effects accumulated (Figure 4D). Therefore, monitoring the system using the checkerboard pattern allows us to confirm that the system is optically aligned and produces correct speckles.

Due to the limited surface area of the SLM, the speckle field of view does not cover the entire camera sensor, resulting in a cropped acquisition area that we term the RimZone (Figure 4E). This RimZone was imaged by acquiring a homogeneous fluorescent sample, such as a Chroma slide, and corresponds to an area of 33.28 × 33.28 µm at the focal plane in our optical setup. However, we noticed that the RimZone is not homogeneously excited, displaying a heterogeneous illumination pattern that corresponds to the Gaussian beam profile of the laser (Figure 4F). It is possible to use the RimZone image from a homogeneous sample to correct for the non‐uniformity of the excitation field.

3.3. Acquisition Optimization

3.3.1. Sparkling Conditions

To perform RIM, it is necessary to acquire a batch of hundreds of speckle images. Figure 5A shows a region of interest with the first four speckles as an example. During acquisition, the structure appears to sparkle, revealing the speckle illumination pattern. However, the effectiveness of this sparkling must be verified, as it depends on two main adjustable parameters: laser power and acquisition exposure time. If the intensity is too low and/or the exposure time is too short, the sample will not be sufficiently illuminated to sparkle effectively.

FIGURE 5.

Acquisition optimization via sample sparkling. (A) example of a sparkling image sequence. Left: first image of the sequence. Right: corresponding ROI for 4 speckles images. (B) FFT analysis of the image in A. (C) Example of a plot profile comparison of FFT analysis. (D) Comparison of image reconstruction with 5 ms SLM exposure time, 200 speckles images and at different laser powers.

The effectiveness of the sparkling can be visually assessed and quantitatively analyzed using Fast Fourier Transform (FFT) on a single speckle image (Figure 5B). When the sample sparkles well, high frequencies appear in the FFT. Conversely, poor or absent sparkling results in a decrease in these high frequencies. The plot profile, after data normalization, shows a downward shift of the curve for poor sparkling (Figure 5C).

One of RIM's challenges is to find a balance between effective sparkling and photobleaching. To achieve efficient sparkling, it is sometimes necessary to increase the laser power or the acquisition time, both of which can lead to photobleaching during speckle data acquisition.

Reducing the laser power results in poorer contrast but minimizes photobleaching, still allowing for reasonable image reconstruction (Figure 5D). Achieving super‐resolution images with RIM requires careful attention to both contrast acquisition and image reconstruction quality.

3.3.2. AlgoRIM Reconstruction Parameters

After acquiring our batch of speckle image datasets, super‐resolution image reconstruction is performed using the AlgoRIM software (version 20_RAM_3_V1) (Mangeat et al. 2021). Two key parameters must be optimized: the pre‐filtering parameter (Wiener filtering, denoted as w) and the regularization parameter (r). A detailed protocol for determining the optimal parameters is provided in Supporting Information (user manual).

Briefly, the optimization process for these parameters is conducted independently. Initially, both w and r are set to 0.1. The value of w is then adjusted in logarithmic increments until an appropriate range is identified, at which point a dichotomous search is employed to refine the value. Once the optimal value for w is determined, the same procedure is repeated for r.

As illustrated in Figure 6 an example of this optimization process is shown using 0.1 µm fluorescent beads, selected for their photostability and suitability for evaluating the resolution of the method. The quality of the reconstruction is assessed based on the separation of beads in the plot profile along the red line (Figure 6A and B). In this example, the optimal parameters are found to be w = 0.01 and r = 0.1 (Figure 6C).

FIGURE 6.

Choice of AlgoRIM reconstruction parameters. (A) Reconstructed image of 0.1 µm beads. (B) Plot profile along the line in A by modifying the pre‐filtering parameter (Wiener filtering or w) and the regularization parameter (r). (C) Reconstruction result by modifying the various w‐ and r‐ values.

The selection of w and r values is dependent on the specific batch of acquired images. Each sample requires a new set of parameters to be determined. Furthermore, any modification in the acquisition parameters for the same sample necessitates re‐optimization of w and r. However, once the acquisition conditions and reconstruction parameters are established for a specific application, further re‐optimization is not required.

3.4. System Resolution

To estimate the resolution of our system, we used an Argolight Argo‐HM slide and compared it with widefield acquisition corresponding to the mean intensity of the speckle stack (Figure 7A). We can also note that the contrast is much better in RIM than in widefield. Using the best conditions seen above, we were able to separate lines 150 nm apart (Figure 7B–D). Thus, the resolution of our system lies between 100 nm and 150 nm in the planar axis.

FIGURE 7.

System resolution (A) RIM reconstruction of a calibrated sample (Argo‐HM slide, Argolight). Left panel: RIM reconstruction. Right panel: widefield image. (B–D) Comparison of plot profiles along the line (75px) in A of the RIM reconstruction image and the widefield image.

3.5. Biological Applications

To finalize the validation of our system, we wanted to compare the RIM technique with other super‐ and high‐resolution microscopy techniques.

The first application is the observation of microvilli structures at the brush border of the C. elegans intestine. Specifically, ERM‐1/ezrin, a structural protein of microvilli, was imaged. The aim is to be able to image the 150 nm periodic organization of the microvilli with our prototype, as already done using high‐resolution techniques like Zeiss Airyscan confocal microscope (Bidaud‐Meynard et al. 2021) and the original RIM prototype (Mangeat et al. 2021). Indeed, we reproduce the results on the same sample using Airyscan microscopy and our RIM system (Figure 8). It is also clearly shown the capacity of optical sectioning of the RIM reconstruction, as microvilli are seen individually with a better contrast than the confocal Airyscan image. However, the Airyscan acquisition time remains significantly slower than the RIM acquisition time, with around 1 min and 20 s, compared to 2 s, respectively.

FIGURE 8.

Comparison of microvilli observation between the RIM system and confocal Airyscan (Zeiss). Microvilli brush border intestine were identified by ERM‐1/ezrin tagged with mNeonGreen and visualized in adult C. elegans.

The acquisition time becomes even more critical when performing 3D imaging. To visualize larger structures, such as mitotic spindles, z‐stacks are required. With the Airyscan setup, acquiring a single plane takes 42 s, whereas it only takes 2 s with the RIM setup (Figure 9). During this process, Airyscan captured 10 z sections in 7 min, while RIM achieved 24 z sections in only 48 s. Thus, the RIM technique significantly reduces the acquisition time while preserving the good resolution. Again, here the optical sectioning capability of the RIM acquisition is presented with a higher contrast than Airyscan, suggesting a sharper z resolution.

FIGURE 9.

Comparison of 3D spindles observation between the RIM system and confocal Airyscan (Zeiss). Microtubules were stained by DM1A‐AF488 in HK2 cell line.

The final application focuses on imaging the mitochondrial network. Here, we compare the RIM system with Super‐Resolution Radial Fluctuation (SRRF), which is a super‐resolution technique generally considered to offer a similar resolution to that of RIM. Conventional confocal microscopy provides a general overview of the mitochondrial network structure (Figure 10). This can be further refined using the SRRF technique, which offers enhanced resolution of the network architecture. However, RIM appears to deliver even more detailed insights, particularly regarding mitochondrial sub‐compartments (Figure 10). These finer structures, which are not visible with SRRF, could be further investigated in dedicated studies. Notably, RIM not only reveals more subcellular details but also does so with significantly faster acquisition times (see Table 1).

FIGURE 10.

Comparison of mitochondria observation between the RIM system, SRRF and confocal SP8 (Leica). Mitochondria were stained by PMPCB‐Alexa fluor 647 in MC7 cell line. Left panel: full image. Right panel: crop from red box.

TABLE 1.

Comparison of different microscopy techniques.

Microscope	XY resolution (nm)	Acquisition time (1 image)	Reconstruction time	Bleaching	Live compatibility	Dimension
SP8	180‐250	∼5 min	0	+++	yes	3D
Airyscan	140	42 s to 1 min 20 s	0	+++	yes	3D
SRRF	60‐120	45 s	2 min	++	yes	3D
RIM	100–150	2 s	1 s	+	yes	3D

The MRic core facility was involved in assembling and validating the prototype. A user manual was drafted to allow users to operate the RIM system independently, without assistance from engineers (see Supporting Information). Each user must adapt the system parameters to their specific biological question, such as the number of speckles, the SLM exposure time, or the use of multidimensional acquisition. In addition, users are responsible for performing data analysis to determine the optimal pair of reconstruction parameters.

4. Discussion

Random illumination microscopy (RIM) is a super‐resolution technique that enables imaging of live samples, including at greater depths than many other super‐resolution methods. Compared to conventional confocal microscopy, RIM offers faster acquisition times and reduced phototoxicity, making it particularly suitable for studying dynamic biological processes.

One of the key strengths of our RIM setup is its optical stability, ensured by a cage system that secures and stabilizes the optical path, thus guaranteeing robust and reproducible imaging conditions. Additionally, alignment can be easily monitored through an integrated metrology setup, allowing precise verification of key elements such as speckle quality and structure. We have also optimized acquisition parameters to enhance performance and reliability.

However, our results highlight that mastering the RIM process can be challenging, as it requires fine‐tuning multiple parameters for each individual sample. A crucial step involves becoming familiar with the speckle pattern in the raw data, which provides an initial indication of the necessary adjustments. Following this, a detailed exploration of each parameter is required to identify the optimal combination. It is essential to critically evaluate the reconstruction results, as achieving super‐resolution does not always guarantee a high‐quality speckle pattern, which is key to reliable imaging.

Several improvements can be made to further enhance the system and enable standardized image analysis. One of our key objectives is to improve the intensity profile of the RimZone, as the current illuminated area lacks homogeneity. We aim to optimize the reflection of illumination by the SLM to produce a more uniform illumination field. Achieving this would make the system suitable for applications involving quantitative fluorescence intensity measurements.

From a technical standpoint, multiple aspects can be refined. Regarding illumination, the current beam expander only partially covers the active area of the SLM. Replacing it with a Galilean telescope composed of achromatic and astigmatic lenses would allow us to cover the full surface of the SLM and potentially extend beyond it. As previously mentioned, the wavefront could also be improved. The current Gaussian intensity profile could be corrected by introducing a lens with a 20 cm focal length, which would expand the RimZone and produce a perfectly uniform illumination field.

In terms of perceptual performance, we also aim to enhance the resolution of our system to match that of the RIM setup at the Laboratoire d'Interactions et Dynamique Cellulaires (LITC) in Toulouse (Mangeat et al. 2021), which achieves a resolution of approximately 100 nm, compared to our current resolution between 100 and 150 nm.

5. Conclusion

In conclusion, we have successfully implemented the RIM system within a core facility setting. Biologists can now operate the system independently, without requiring assistance from a facility engineer. To ensure easy access and autonomous use by all users, we have developed a standardized training protocol and a user‐friendly manual. This implementation marks a significant step toward making advanced super‐resolution imaging techniques more accessible to the broader biological research community.

Conflicts of Interest

M. T. is co‐founder and scientific advisor of Inscoper. S. L. and T. M. are co‐founder of Rimeo. All other authors declare no competing interest.

Supporting information

Supporting File 1: boc70060‐sup‐0001‐SuppMat.pdf.