Microimager: a flexible thin-film miniaturized endoscope for optical biomedical imaging

Abstract

Optical imaging is the gold standard for visualizing the structure and function of biological tissue. Non-invasive imaging methods can only reach a limited depth while providing a high spatial resolution. On the other hand, implantable imagers that can access deep tissue are prohibitively large and invasive. Here, we present the Microimager, a flexible, miniaturized thin-film endoscope (7 × 400 μm) featuring multiple independent channels for high-resolution light delivery and collection from deep tissue. The Microimager consists of an array of parylene photonic waveguides implemented using a scalable microfabrication process. We experimentally demonstrate spatial discrimination and imaging of 30 µm features on a resolution mask, as well as distinct regions in mouse brain tissue. The Microimager is a useful addition to the optical biomedical imaging toolset and can provide access to deep tissue in a minimally invasive way for a wide range of applications.

1. Introduction

Optical biomedical imaging is a versatile and powerful technique that employs ultraviolet, visible, or near-infrared light to capture images of biological samples based on light-matter interaction through reflection, refraction, absorption, scattering, or fluorescent emission [1–3]. It provides detailed structural, functional, and molecular information [4]. Over the past few decades, it has been widely used in basic science research as well as clinical diagnostics. Several optical imaging modalities have been implemented. For example, fluorescent imaging has been used for molecular imaging, providing helpful insights into biological processes [5–8]. This technique involves illuminating the tissue with light at a specific wavelength to excite fluorophores—molecules that fluoresce upon excitation and are either genetically expressed or chemically attached to specific target cells. The excited fluorophores emit light at a longer wavelength, which is captured and filtered by the optical imaging system to provide detailed information about the location and function of the labeled cells [9]. Fluorescent imaging is used in many different applications, such as imaging malignant cells, which are often difficult to distinguish from healthy cells using conventional white-light optical imaging [10,11]. It has also played a key role in basic science discovery, especially in neuroscience, where it helps with monitoring neural tissue activity and identifying the role of specific cell types across populations of neurons to unravel the complex brain function that mediates behavior [12–14]. In clinical applications, fluorescence imaging in guided surgery helps with the identification of tumor locations and boundaries during operations. Visualization of tumors during surgery enhances resection precision by minimizing harm to healthy tissue [15,16].

Optical imaging has also been employed to enhance the accuracy of biopsy by allowing precision targeting [17]. Moreover, real-time visualization of the vascular system to prevent accidental damage during procedures is another key advantage of using optical imaging [18,19]. It has also been used for monitoring blood flow and tissue perfusion to ensure tissue viability in different conditions [20,21]. Optical imaging has also served as a powerful method for evaluating the efficacy of targeted therapies and monitoring the distribution of drugs within specific tissue regions [22].

Single-photon, multiphoton, confocal microscopy, and optical coherence tomography (OCT) are commonly used biomedical optical imaging methods [23,24]. The basic structure of an optical microscope consists of a lens system that projects the backscattered light from within the tissue onto an array of photodetectors to form the image of the underlying structure. However, traditional microscopes are typically limited to imaging superficial layers of tissue due to light scattering and attenuation, restricting the imaging depth and resolution [25,26]. They are often bulky and have a limited field of view (FOV) [16]. To address the bulkiness issue, miniaturized microscopes, called miniscopes, have been developed, which can be implanted into the tissue and offer enhanced maneuverability. The miniscopes are also composed of a compact lens system and a photodetector array (e.g., an image sensor), similar to conventional microscopes but in a much smaller form factor. They are primarily used in research, particularly in neuroscience, for imaging small, localized areas to monitor neural activity in live animal subjects using calcium imaging [27]. The invasiveness of microscopy techniques depends on the accessibility of the target region of interest. When imaging a superficial layer of tissue such as skin, the technique is considered noninvasive. However, accessing deeper regions may require surgical removal of superficial tissue layers, rendering this method invasive.

Optical imaging of deeper organs due to light absorption and scattering in biological tissue, requires endoscopic tools, including a long, rigid, or semi-flexible insertion tube that contains the light source and a miniaturized imaging system [28]. It is used for diagnostic and therapeutic purposes and is introduced through natural body orifices or small surgical openings in superficial tissues. Endoscopes are widely used in applications such as gastroenterology, cardiology, urology, pulmonology, and orthopedics. Compared to microscopes, endoscopes offer greater compactness and maneuverability, resulting in a larger FOV [16]. Endoscopes generally provide lower resolution than microscopes because they prioritize compactness, limiting the integration of high-resolution optics and sensors [29,30]. Insertion of endoscopes causes tissue damage. While such damage may be less critical in surgical procedures, minimizing it is essential during diagnostic procedures to preserve tissue integrity and function. The state-of-the-art endoscopes are still relatively bulky and can severely damage the tissue. There are different designs of endoscopes. An imaging system at the distal end equipped with a photodetector chip, capturing images directly at the tip, is referred to as a distal-chip endoscope. Recent innovations in implantable imaging have introduced lensless devices capable of delivering light and collecting fluorescence signals directly on-chip [31–33]. In contrast, in fiber-optic endoscopes, a fiber bundle relays the image from the distal end to a photodetector chip at the proximal backend. Each individual fiber relays a separate pixel of the scene to the image sensor array. The distal-chip endoscope has higher image quality, whereas the fiber-optic endoscope, due to the limited density of optic fibers, has lower resolution but is simpler and more compact at the distal end, making it less invasive and more flexible [34].

A recent advancement in fluorescent imaging for neuroscience research related to fiber optic endoscopy is fiber photometry, which typically employs a single fiber that penetrates deep into the brain to detect light emitted from fluorescently tagged cells. This method produces a single-pixel image, offering limited information about specific cell types within a small volume near the fiber tip without revealing more details. Increasing the number of pixels is desired, but using a bundle of fibers can become prohibitively invasive [35–37].

Despite ongoing efforts to miniaturize endoscopic tools, they often remain bulky, resulting in significant tissue damage. Using fiber bundles has limitations, including the relatively large diameter of fiber bundles, rigidity, and limited FOV, especially since all the fibers terminate at the same distal plane. Additionally, the tightly packed honeycomb arrangement of the fiber strands limits the customizability of the pixel layout.

Miniaturized flexible thin-film integrated photonic waveguides can be used to design the next generation of ultra-compact endoscopes addressing the limitations of fiber bundle endoscopes by reducing the mechanical mismatch with soft tissue and offering a customizable design of the imager. By leveraging planar microfabrication processes, a high density of on-chip optical waveguides can be implemented on flexible polymer substrates [38]. These thin-film endoscopes enable seamless integration with robotic surgical tools via lamination, and their customizable pixel arrangements support targeted illumination and imaging.

In this work, we introduce the Microimager, a miniaturized, thin-film flexible endoscope designed for localized optical imaging. The Microimager utilizes a dense array of Parylene photonic waveguides, a fully flexible, compact, and biocompatible waveguide array fabricated using a scalable planar microfabrication process [39] that has been modified and improved in this study. Unlike traditional fiber bundles, the Microimager can be seamlessly integrated with light sources, detectors, and surgical tools, simplifying system design while its thin profile minimizes tissue damage for deep tissue penetration. Furthermore, its customizable pixel arrangement facilitates precise, localized illumination and imaging, and it can be incorporated into existing surgical tools to reduce procedural complexity.

To validate its performance, optical characterization experiments were conducted to identify the input-output relationship when the Microimager was illuminated with a localized light source. An optical mask was then used to show the ability of the Microimager to accurately relay input patterns to the output. Fluorescent imaging capabilities were assessed using fluorescent microspheres (FL beads). Proof-of-concept experiments included ex vivo imaging of live brain slices from transgenic mice expressing green fluorescent protein (GFP), showcasing the Microimager's spatial discrimination capabilities. Additionally, calcium imaging was performed to demonstrate the Microimager's ability to capture and relay temporal dynamics of localized neural activity.

2. Device design and architecture

Figure 1 illustrates the envisioned schematic design of the Microimager system. For fluorescent imaging, the Microimager should be capable of delivering excitation light and collecting the fluorescent emission from within tissue. To accomplish this, the Microimager is equipped with a dense array of flexible optical waveguides along an implantable thin-film shank. These waveguides relay a pixelized fluorescent image (for example, of the fluorescent protein-tagged neurons within the brain) to outside the tissue that can be detected by an image sensor, which is housed in the backend optoelectronics. The waveguides are also reciprocal, i.e., they can collect and deliver light. In the envisioned device design (Fig. 1), some of the waveguides in the array can be designated for light delivery, while the surrounding waveguides can be used to collect the fluorescent emission. This way, an ultracompact Microimager can be implemented with embedded light delivery and light collection waveguides and an integrated image sensor array. In this paper, we only focus on demonstrating the relay imaging capability of the Microimager. In other words, each waveguide is used exclusively for collecting fluorescent emission light. For light delivery, an external optical fiber was used, and the light from the output was projected onto an external microscope.

Fig. 1.

Envisioned schematic design of the Microimager. The Microimager, composed of a dense array of flexible optical waveguides along an implantable thin-film shank, relays pixelized fluorescent images of protein-tagged neurons to the image sensor. The central waveguide serves for excitation light while surrounding waveguides collect the fluorescent emission light.

To implement the Microimager, a biocompatible, flexible waveguide platform capable of efficiently transferring light was required. Among the available options [38], we selected Parylene photonics [39], a novel flexible integrated photonics platform entirely composed of biocompatible polymers. The waveguide core is made of Parylene C, a polymer with a high refractive index (n = 1.639) that remains transparent across the visible optical spectrum. Polydimethylsiloxane (PDMS) is used for the waveguide cladding due to its lower refractive index (n = 1.4) compared to Parylene C. This combination of Parylene C and PDMS provides a significant index contrast (Δn = 0.239) among biocompatible polymers, effectively confining the optical mode (Fig. 2(a)). The Parylene photonic platform is realized through a scalable micromachining process on the polymer layers sitting on a silicon substrate. Then, the flexible polymers are released from the substrate to realize a fully flexible implantable device.

Fig. 2.

Microimager design. a) The combination of Parylene C as a core and PDMS as a cladding effectively confines the optical mode. b) The cross-section of the Parylene photonics waveguide, including the Parylene C core, PDMS cladding, and input/output micromirror with out-of-plane illumination. c) Schematic of the Microimager with fully flexible shank integrated with backend optoelectronics. The inset figures show the input and output ports. d) Different design arrangements of input pixels are used to provide various coverages. The red lines indicated the thin-film shank outline, with a 60-degree tip to facilitate implantation.

A key feature of the Parylene photonics platform is the inclusion of embedded 45° micromirrors at the input and output ports, which facilitate broadband input/output light coupling and imaging [39]. These micromirror structures enable 90° out-of-plane light coupling and imaging (Fig. 2(b)). Unlike traditional optical waveguides and fibers that rely on end-firing light emission from the end facet, this out-of-plane approach is preferred for achieving high spatial resolution imaging and illumination along the thin-film shank. Additionally, it allows for the integration of backend optoelectronics on the surface, resulting in a compact design suitable for chronic experiments on freely moving animals. Image sensors, such as charge-coupled devices (CCD) or complementary metal-oxide-semiconductor (CMOS) sensors, can be directly integrated onto the surface for light detection. Furthermore, light sources can be coupled from the surface using an optical fiber or by integrating a laser diode with the optical waveguides (Fig. 2(c)). In this work, we utilized a microscope equipped with a CCD sensor positioned above the output to detect the emitted light. An optical fiber above the input was employed as the light source to excite the fluorescent medium.

The width of the optical waveguides can vary depending on the fabrication constraints, Microimager resolution, and the number of waveguides on the shank. In terms of fabrication, optical photolithography is used to pattern different device layers with a resolution of 1 μm. Therefore, theoretically, the smallest waveguide width and pixel size is 1 μm. However, the optical propagation loss of waveguides smaller than 5 μm is very high. To achieve smaller waveguides, the fabrication process must be optimized by using high-resolution lithography techniques, such as electron beam lithography, and optimizing the etching processes. In this study, the cross-section of the Parylene waveguide core is 10 µm × 5 µm (width × thickness) with a pitch size of 20 μm. The dense waveguide array consists of 20 waveguides, making the width of the whole Microimager shank 400 μm. The cladding thickness was 1 μm, as it had been shown that this thickness was sufficient for the waveguide to effectively confine the optical mode [39]. Therefore, the overall thickness of the implantable flexible shank is 7 μm, resulting in a thin device that minimizes damage to the tissue.

The implantable flexible Microimager is designed in various lengths, ranging from 4 to 8 mm, depending on the target depth of interest. The waveguide portion on the silicon (Si) backend is 2 mm, resulting in an overall optical waveguide length of 6 to 10 mm. During the microfabrication process to release the flexible polymer device from the Si substrate, a small portion of rigid Si is retained at the backend of the Microimager to facilitate integration with the optoelectronics system and to ease handling during assembly. Each waveguide relays optical intensity from the input port to the output port, forming a “pixel” of the Microimager. The current Microimager design has 20 pixels at the input/output ports. The output port is designed with a 2D periodic step-shaped arrangement consisting of 5 rows of 4 pixels. This design achieves a compact layout while ensuring that the pixels are separated to minimize interference between adjacent pixels, facilitating integration with an image sensor or a photodetector array. In general, the input pixel locations and the number of pixels can be customized in any arbitrary 2D arrangement on the Microimager implantable shank to provide the desired coverage or resolution within the imaging area of interest. For example, for imaging a larger area, a sparse 2D arrangement might be desired, and for imaging a local region with higher resolution, a dense arrangement is recommended. Figure 2(c) shows an example of a grid-like 2D arrangement of input pixels, and Fig. 2(d) shows additional 2D designs with different linear pixel arrangements.

Parylene C, used as the waveguide core material in the Microimager, exhibits low autofluorescence in the green and red wavelengths of the visible light spectrum while showing higher autofluorescence under UV and blue excitation due to photo-induced dehydrogenation and oxidation [40]. In this work, although the excitation light propagates through the Parylene waveguides, autofluorescence is minimized by using appropriate emission filters before the camera and selecting GFP fluorophores with emissions outside high-autofluorescence wavelengths. Regarding power handling, the optical power used is well below the damage threshold for Parylene C, and no photothermal degradation is observed during operation.

3. Methods

3.1. Device fabrication

To implement the Microimager, we used the Parylene photonics platform [39]. Parylene photonics was fabricated at the wafer scale using a planar micromachining process to form the waveguide structures, which was reported in detail in prior work [39]. In this work, we reported some improvements to this microfabrication process.

The crystalline planes of the wafer are important for the Parylene photonics platform to form 45° sidewalls in a Si mold to define the micromirror surface and shape of the waveguide. The fabrication process began with a 4-inch Si wafer (n {100}) that had a 1-μm layer of thermal oxide (SiO₂) grown on top (Fig. 3(a)). The mask design features were oriented at 45° with respect to the (100) plane, which was necessary to expose the (110) crystal plane and define the 45° Si sidewalls and micromirror surface. The oxide layer was patterned using the described mask through optical lithography and anisotropic reactive ion etching (RIE), which then served as the hard mask for etching Si (Fig. 3(b)). Using the oxide hard mask, the waveguide template was etched in silicon in a solution of 2 M potassium hydroxide (KOH) mixed with 60 ppm Triton X-100 surfactant to achieve a smooth mold with angular sidewalls by revealing the (110) plane and reaching a target depth of 6 μm (Fig. 3(c)) [41]. Then the oxide hardmask was removed in 49% Hydrofluoric acid (HF) (Fig. 3(d)). A 300-nm SiO₂ layer was deposited conformally on the patterned Si surface using plasma-enhanced chemical vapor deposition (PECVD) as a sacrificial layer to enable the device release from the Si mold (Fig. 3(e)).

Fig. 3.

Fabrication process of the Microimager. a) The fabrication process began with a 4-inch Si wafer (n {100}) that had a 1-μm layer of SiO₂ grown on top. b) The oxide layer was patterned using optical lithography and anisotropic RIE, which then served as the hard mask for etching Si. c) Si was etched in a solution of 2 M KOH mixed with 60 ppm Triton X-100 to achieve a smooth mold with angular sidewalls by revealing the (110) plane and reaching a target depth of 6 μm. d) The hardmask was removed in 49% HF. e) A 300 nm SiO₂ was deposited on the Si surface using PECVD as a sacrificial layer to enable device release. f) a 1 μm diluted PDMS was spin-coated onto the Si mold as the lower cladding. g) A 300-nm layer of Parylene C film was deposited on the PDMS using CVD to act as an adhesion layer for the photoresist. h) Metal layers were deposited by evaporating 5 nm Pt and 100 nm Al. i) Micromirrors were patterned using a lift-off process. j, A 5 μm layer of Parylene C was deposited using CVD. k) The outline of the individual waveguide core was defined through anisotropic O₂ plasma etching using RIE. l) A 1-μm layer of diluted PDMS was spin-coated as the upper cladding. m) The outline of the Microimager was defined using the thick photoresist etch mask. Then, the entire 2-μm PDMS claddings were etched using O₂ and SF₆ plasma with RIE. n) The backside Si was etched using a DRIE process. o) The device was released by removing the sacrificial layer in buffered HF.

To form the substrate of the Microimager and lower cladding for the waveguide structure, PDMS (Sylgard 184, Dow Corning Corp, USA) was used. Due to its high viscosity, spin-coating thin conformal layers of PDMS was challenging. Therefore, at first, it was diluted in hexane with a ratio of 1:10, and then it was spin-coated (60 s, 2000rpm) onto the Si mold mold to a 1 μm thickness. Finally, the film was degassed (1 mTorr) and oven-baked (1 hr, 100°C) to cure (Fig. 3(f)).

To integrate micromirrors, they needed to be patterned on 45° sidewalls of the Si mold. Because of the low surface energy of PDMS [42], a very thin (300 nm) layer of Parylene C film was deposited on the PDMS using chemical vapor deposition (CVD) to act as an adhesion layer for the photoresist and enable optical lithography (Fig. 3(g)). Then metal layers were deposited by evaporation of 5 nm Pt and 100 nm Al (Fig. 3(h)), and micromirrors were patterned using a lift-off process through acetone soaking followed by sonication (Fig. 3(i)). Pt served as a strong adhesion layer to Parylene [43], while Al was an effective mirror surface due to its high reflectance across the visible spectrum [44].

The waveguide core was then fabricated in Parylene C. A 5-μm layer of Parylene C was deposited using CVD on top of the thin adhesion layer previously deposited during the micromirrors step (Fig. 3(j)). To define the outline of the individual waveguide core and remove unwanted Parylene C regions, a thick photoresist (12 μm, AZ4520) etch mask was used [45]. The waveguide patterns were aligned to the micromirrors using optical lithography. The patterns were subsequently transferred to Parylene C through anisotropic oxygen (O₂) plasma etching using RIE. PDMS acted as an etch-stop layer because this polymer was resistant to etching by O₂ plasma alone [46]. Finally, the photoresist etch mask was removed using acetone (Fig. 3(k)).

A 1-μm layer of diluted PDMS was spin-coated as the upper cladding on Parylene C core waveguides, similar to the process used for the lower cladding (Fig. 3(l)). Next, the outline of the Microimager was defined using the thick photoresist etch mask. Then, the entire 2-μm PDMS claddings at the device outline were etched using O₂ and sulfur hexafluoride (SF₆) plasma in an RIE process to singulate the individual devices (Fig. 3(m)).

To release fully flexible devices, the Si backend of the Microimager was retained while the excess silicon was removed. This was achieved by patterning the backside of the Si substrate using a thick photoresist etch mask, followed by etching through the Bosch process. The Bosch process was performed in two steps: a 12-second etch with O₂ and SF₆ plasma, followed by an 8-second passivation with C₄F₈. The sacrificial oxide layer served as the etch stop to protect the backside of flexible shank (Fig. 3(n)). After Si was removed, the etch mask and sacrificial layer were stripped using acetone and buffered HF acid, respectively, resulting in a released, flexible device with a Si backend (Fig. 3(o)). The devices were thoroughly rinsed in deionized water after release to avoid contamination of biological tissues by the process chemicals. The ability to keep the Si backend while achieving a fully flexible shank represented a significant advancement, as it made handling easy during assembly and facilitates integration with the backend optoelectronics, thereby enhancing their performance and versatility in neural interfaces.

3.2. Optical propagation loss measurement

For the waveguide to serve as an effective Microimager and to relay the images from the deeper regions in tissue to the device backend, the optical propagation loss of individual waveguides needs to be low enough to transmit the optical signal without excessive attenuation. We assessed the optical performance of the Microimager using a custom-designed bench-top optical setup [47]. For the light source, a single-mode fiber (S405-XP, Thorlabs Inc., USA) was used to transfer light from the fiber-coupled laser source (LP450-SF25, ThorLabs Inc., USA) at λ = 450 nm to the Microimager input, coupling the light into the waveguides. The sleeve and protective coating of the fiber were stripped to expose the bare fiber. To prevent light from escaping through the sides of the fiber, it was painted black, allowing light to emit only from the tip. To adjust the position of the fiber, it was mounted on a precise motorized XYZ-micromanipulator (PatchStar, Scientifica Inc.) for alignment with the Microimager input. The fiber was vertically located 20 µm on top of the input pixels to couple light more efficiently. A CCD image sensor (EO-5012 M, Edmund Optics) equipped with a zoom objective lens (600i, Edmund Optics) was used to image the Microimager output. The location of the sensor was also adjusted using an XYZ moving stage, along with a rotation stage (ThorLabs Inc., USA), to enable imaging of the output at an optimal angle.

The optical loss of waveguides was measured using the modified cut-back method [48]. The input coupling is optimized by adjusting the position of the laser fiber using a precision micromanipulator to maximize the intensity from the output pixels imaged by the image sensor. To measure the loss, we fabricated a series of waveguides ranging from 0.5 cm to 3 cm in length to use the modified cut-back method and measure the propagation loss. The blue laser with a wavelength of 450 nm at 5 mW power was used to characterize the propagation loss. This choice of wavelength provides the worst-case scenario for the propagation loss since it increases with decreasing the wavelength [39], and in practice, we are detecting fluorescent signals at higher wavelengths, i.e., λ = 520 nm.

The overall optical transmission through the waveguides is determined by a combination of input/output coupling efficiency and propagation loss within the waveguide structure. Coupling efficiency is affected by the location and distance of the source, reflections, and scattering losses introduced by surface roughness at the micromirrors. Propagation loss within the waveguide is primarily attributed to material absorption, which is minimal in the visible range due to the high optical transparency of Parylene C (transmittance >95%) [49,50] and scattering from sidewall roughness introduced during fabrication and higher-order mode coupling, resulting from imperfect mode matching between the optical fiber and the waveguide. Under ideal conditions, FDTD simulations using Lumerical-Ansys software indicate that >85% of the incident power can enter the input facet, and >75% can be coupled into the guided modes of the waveguide [51]. In practical implementations, however, fabrication imperfections reduce efficiency relative to this theoretical upper bound. However, by addressing fabrication imperfections, including micromirror and sidewall quality, optical transmission can be significantly improved.

3.3. Microimager experimental setup

To evaluate the performance of the Microimager, an experimental setup (Fig. 4) was designed. This setup is a modified version of the optical characterization setup with some additional components. A microscope with a CMOS Camera (Prime BSI, Teledyne Photometrics, Tucson, AZ, USA) and different objective lenses were used. To capture the image of the input and output port of the Microimager together, the microscope objective lens needs to move alternately between the ports. A 4x objective lens (Olympus Corp, Tokyo, Japan) was used at the input for imaging a wider field of view, and for the output, a 10x objective lens (AmScope, United Scope LLC, Irvine, CA) was used. Because of the small working distance of the lens and the short length of the Microimager, the optical fiber delivering the fluorophore excitation wavelength was positioned obliquely, 5 mm above the input, to allow the microscope to see the input while providing uniform light. By having a 5-mm distance from the surface, the Gaussian distribution of the laser beam was expanded and was nearly uniform over the input. The light source is a 450 nm laser (LP450-SF25, ThorLabs Inc., USA) used as the excitation light for fluorescent microspheres and fluorescent protein-tagged neurons, which emit green fluorescence with peak wavelengths of 515 nm and 509 nm, respectively. This green emission light needed to be captured exclusively from the output port by filtering out the blue excitation light. Thus, the microscope was equipped with a green filter (509/22 nm). To eliminate interference from stray light in the environment, light barriers, including a black silicone wall around the output and a black plastic conical wall around the lens, were used.

Fig. 4.

Microimager experimental setup.

For the ex-vivo experimental setup, a custom-designed 3D-printed chamber with an inlet and an outlet was used to allow the perfusion of artificial cerebrospinal fluid (aCSF) to maintain the viability of brain slices. The location of the outlet is designed to be higher than the inlet to ensure that the chamber is filled with aCSF and the level of aCSF remains constant for imaging. The on-chip Microimager device was glued to a predefined location measuring 20 mm × 15 mm at the center of the chamber.

3.4. Intensity measurement

After aligning the desired object with the Microimager input and the laser beam, the image of the input was captured by the CMOS camera with a 4x lens. The exposure time was set to less than 50 ms.

Then, to image the output pixels, the objective lens was repositioned to the output port. The signal at the output was weak due to the limited amount of filtered fluorescent light and optical losses within the waveguides. To enhance the signal and improve the SNR, the exposure time was increased to 500 ms. Additionally, to increase the numerical aperture (NA) for better light collection and for better resolution, a 10x objective lens was employed. As the output port may have low SNR in comparison to the brain slice at the input port, using a higher magnification lens would increase the apparent size of the signal, collect more light, and enhance the contrast relative to the background.

The location of each input and output pixel, designated as the region of interest (ROI), is labeled with a red and blue circle, respectively, and numbered in green. The diameter of each input ROI is 10 image pixels, while the diameter of each output ROI is 40 image pixels. The output ROI was chosen to be larger due to higher magnification and the diverging output beams. The average intensity within each input ROI was used as a representative measure of the intensity for each pixel.

3.5. Transfer function and normalization of intensities

To enhance the correlation between input and output, a transfer function was applied to calibrate the output intensities. First, each waveguide exhibits a unique end-to-end optical loss due to variations in propagation loss, primarily caused by scattering from fabrication imperfections combined with input/output coupling efficiency due to the surface quality of the waveguide facet and micromirrors, which in turn affects the output light intensity. To calibrate the waveguides and remove the effect of optical loss, all waveguides needed to receive equal light, and based on the corresponding output light, we can then adjust their relative intensities accordingly. As the laser beam exhibited a near-uniform Gaussian intensity distribution at the input due to the 5 mm distance of the fiber from the surface, it was used for calibration while maintaining all the conditions the same for each experiment. To achieve this, the image of output light, which only corresponds to the uniform laser beam, was captured (I_Calib). Another factor that affects the transfer function was the background noise of the actual output and calibration intensities. To find and remove the background noise, in the same condition, the laser was turned off, and the image of the output was captured (I_noise). As the camera location was stationary, the same coordinates of output ROIs were used for measuring the calibration and noise intensities. Now, with the above information, we can derive the transfer function for the relay Microimager system as

$$\frac{I_{output}-I_{noise}}{I_{Calib}-I_{noise}}$$ (1)

After applying the transfer function, the intensities were normalized to the maximum intensity for better comparison, resulting in values between 0 and 1. To compare the input and output pixels, the normalized intensities of the input ROIs and the output ROIs were plotted in a figure. Additionally, to facilitate interpretation of the step-shaped output pixels, they were unwrapped by presenting the normalized intensities of the ROIs horizontally under the intensity figure.

3.6. Imaging using Microimager

3.6.1. Relay imaging using a coherent focused light beam

To assess whether coherent light can couple to the waveguides, determine if the input and output of the Microimager are correlated, and verify that the Microimager can relay the pattern and gradient of the input light onto the output, the following experiment was conducted. The laser fiber was moved along the input pixels while the output pixels were imaged using the CMOS camera. In this experiment, the distance between the laser fiber and the surface was maintained at 1 mm to illuminate only a limited number of pixels. As a result, the calibration and transfer function were not applied; only the intensities were normalized. For these experiments, devices with the third input pixel arrangement, shown in Fig. 2, were used.

3.6.2. Relay patterns using an optical mask

To determine if the Microimager can relay patterns from the input to the output, an optical mask with bright and dark stripes of varying sizes, ranging from 20 to 100 μm, was designed (Fig. S1) and printed on a transparency film (Artnet Pro, San Jose, CA, USA). In this experiment, the third input pixel arrangement in Fig. 2 was used. The film was positioned over the input, and the oblique optical fiber was used to direct the laser beam at the input. Images of both the input and output were captured for all stripe widths. The local minimum and maximum of the output pixel intensities were considered dark and bright stripes, respectively.

3.6.3. Spatial discrimination by imaging fluorescent beads

To validate the fluorescent imaging capability of the Microimager, fluorescent green microspheres (Cospheric LLC, CA) were used. These fluorescent microspheres (FL beads) are made of polyethylene and have a size of 125–150 μm. Their peak emission is at 515 nm when excited at 414 nm. To make handling easier, they were embedded in 2% agar hydrogel and then sliced to a thickness of 350 μm using a vibratome (VT1200S). This thickness was chosen to match the brain slice experiment. In this experiment, the third input pixel arrangement in Fig. 2 was used, and the number of waveguides was 20.

The agar slice embedded with FL beads was located at the input of the Microimager chip. FL beads were positioned at the center of the input by adjusting the agar slice gently with a small brush. Then, using the oblique optical fiber, the laser beam was directed at the input/FL beads. In this experiment, the third input pixel arrangement from Fig. 2 was used.

3.6.4. Ex vivo experiments: Spatial and temporal discrimination

For validation of the fluorescent imaging capability of the Microimager in biological tissue, the Microimager setup was used, which includes live brain slices from transgenic mice. For this, we crossed the Ai93D mice (The Jackson Laboratory, USA) with SynCre mice to express the green fluorescent proteins (GFP) in excitatory neuronal cells. The peak excitation light for GFP is at a wavelength of 490 nm, and the peak emission light is at 509 nm. The brain was isolated from mice and transferred to the slicing station. Further, using a vibratome (VT1200S), thin slices of 350 µm were prepared in ice-cold aCSF perfused with oxygenated carbogen (95% O₂ and 5% CO₂) and kept in an incubation chamber containing oxygenated aCSF at room temperature for 45 min prior to use for imaging.

Slices were transferred to the custom-designed chamber for the Microimager experimental setup. The brain slice was placed over the Microimager input and anchored by a harp. As the brain slice covered the input completely, a captured input image was overlaid on the brain slice to assist in locating the inputs and its alignment with the desired brain region. After aligning the Microimager input, the desired brain slice region, and the laser beam, the input image was captured. In this experiment, the third input pixel arrangement in Fig. 2 was used.

To demonstrate the effectiveness of Microimager in detecting the temporal dynamics of neural activity through calcium imaging, we designed an ex vivo experiment using the same transgenic line as used in the fluorescent imaging experiments. In this experiment, the second input pixel arrangement in Fig. 2 was used.

Briefly, a mouse brain slice was placed on the Microimager input. The microscope light, provided by a broadband 130 W Mercury Vapor Short Arc lamp (U-HGLGPS, Evident Scientific, MA, USA) and filtered through a blue bandpass filter (480/17 nm), was used as the excitation light due to its larger coverage in the microscope field of view, making it easier to locate the response region. Initially, an exploratory experiment was conducted to identify the zone of maximal neural response in the somatosensory cortex of the brain slice. To achieve this, stimulation was performed in layer 5 of the cortex, and the neural response was recorded in the whole somatosensory cortex. For stimulation, a square biphasic pulse signal with an amplitude of 1 mA, a pulse width of 200 µs per phase, and a frequency of 0.2 Hz was applied using a platinum/iridium concentric electrode connected to the stimulation system (Fig. 4). Once the region with maximal neural response was observed, evident from significant fluorescence change from Ca imaging, the Microimager input was aligned to the same brain regions. The laser at the wavelength of 450 nm was targeted onto the response region instead of microscope light as the excitation light and the concentric electrode were repositioned at the exact previous stimulation point. Then 3 trials of stimulation parameters were applied to evoke Ca activity. The input was recorded with 20 frame/s for 15 s (20 frame/s × 15 s = 300 frames). To record the fluorescent change at the output, the objective lens was adjusted, and the same stimulation signal was applied. The emitted light at the output was recorded at 20 frames/s for 15 seconds.

The average intensity of all ROIs at the input and output (F) was measured similarly to the fluorescent imaging experiment but over time. Specifically, the average intensity was calculated for each frame across all ROIs. The first 20 frames were considered as the baseline (F0). Finally, the input and output intensity changes were calculated as

$$\frac{F-F_0}{F}=\frac{\mathrm{\Delta }F}{F}$$ (2)

4. Results and discussion

4.1. Device design and fabrication

We successfully designed and fabricated the Microimager. Our fabrication method was based on planar micromachining of thin polymer films. Figure 5(a) shows the arrays of waveguides and their output pixels on the fabricated on-chip Microimager. The inset provides a closer view of the output pixels, highlighting different components of the Microimager, including the micromirrors, the Parylene waveguide cores, and Si molds. Figure 5(b) also shows the arrays of waveguides and their input pixels. Figure 5(c) demonstrates the released flexible thin-film device with the silicon backend. In this study, only the on-chip devices are used in all the experiments to showcase the imaging capability of the Microimager.

Fig. 5.

Optical microscope image of the fabricated device and optical loss measurement. a) Waveguide array and output pixels on the fabricated on-chip Microimager. The inset provides a closer view of the output pixels, highlighting different components of the Microimager. b) Waveguide array and input pixels on the fabricated on-chip Microimager. c) The flexible released device with the silicon backend d) The output intensity versus the waveguide length. The slope of the fitted line indicates the propagation loss of the waveguide.

In this study, the fabrication of Parylene photonics was modified. Previously, thin metal layers like chromium and aluminum were used as hard masks to etch the Parylene waveguides. However, patterned metals have rough edges, which are transferred to the Parylene substrate during the etching process, resulting in rough waveguide sidewalls and, as a result, increasing the optical loss [52]. To mitigate this roughness, another thin layer of Parylene was deposited on the etch waveguide to smoothen the sidewalls. However, in this study, to simplify the process, a thick photoresist etch mask is used instead of the metal hard mask. Photoresist masks have smooth edges, and as a result, the etched Parylene sidewalls are smooth (Fig. 5(a)).

To validate the modified fabrication process, the device loss was measured. Figure 5(d) illustrates the scatter plot of the logarithm of the output intensity in different waveguide lengths. The yellow line shows the best-fit curve derived from an exponential regression model. The slope of the fitted line indicates the propagation loss of the waveguide, measured at 8.48 dB/cm for a 10 µm wide waveguide at a wavelength of λ=450 nm. Compared with the previously reported results that indicated a propagation loss of 6.1 ± 1.4 dB/cm [39] for a wider waveguide (30 µm wide) with an overlay of a smoothening layer, the achieved propagation loss in this study for a much narrower waveguide using a simpler fabrication process is very encouraging. This shows that process simplification (using a photoresist mask instead of a hardmask) can yield a reasonable performance with fewer steps. We should also note that the propagation loss in the previous work was measured using an outscattering measurement method and not the modified cut-back approach used here. Overall, using a photoresist mask, as opposed to a metal hardmask, enabled us to create smooth sidewalls and achieve low enough propagation loss for very compact waveguides.

Although the selectivity of metals is better than that of the photoresists for etching Parylene (approximately 1:1), to address this issue, the photoresist should be thicker than the desired etch depth. To be on the safe side, at least 2 times the desired etch depth is recommended. For example, in this study, a 12-μm photoresist etch mask was used to etch 5 μm Parylene C. Metal hard masks usually have intrinsic stress, which can cause cracks in the substrate. In contrast, photoresist masks usually exhibit low stress compared to metal hardmasks [53]. Another advantage of photoresist masks is their ease of use. They can be easily deposited by spin coating and patterned using optical lithography. Metal hardmasks require complex deposition tools like sputtering or evaporation, followed by patterning with a photoresist. The pattern then needs to be etched onto the metal layer, which finally needs to be etched again to be removed. The 2 μm PDMS substrate of the device was also etched to define the device outline using a thick photoresist etch mask, which replaced the previously used metal hard mask. Previously, XeF₂ was used to remove the Si substrate and release the device. However, XeF₂ etching is isotropic and not directional, which also etches the Si backend that needs to be preserved. In this study, anisotropic deep reactive ion etching (DRIE) was employed to achieve vertical sidewalls and protect the Si backside (Fig. 5(c)).

In this study, we aimed to keep the width of the Microimager under 400 μm, resulting in a footprint of approximately 2800 μm² with a width of 7 μm. In contrast, the thinnest commercial fiber used in fiber photometry has a diameter of 200 μm [54–57], resulting in a footprint of 31,400 μm², which is 11.2 times larger than our design. Additionally, the Microimager is flexible and features 20 pixels, whereas commercial fibers are rigid and have only a single pixel.

In this design, the pitch size includes the width of the waveguides (10 μm) and two 45° sidewalls (2 × 5 μm) on either side of each waveguide, resulting in an overall pitch size of 20 μm. The width of the 45° sidewalls is determined by the etch depth; as the etch depth increases, the width of the sidewalls also increases. Instead of having a separate silicon mold for each waveguide, we can merge them and have only one large trench for all the waveguides. This approach will reduce the distance between waveguides. However, the potential for crosstalk between waveguides must be considered.

Optical mode simulations show that the platform can be scaled to support ultracompact waveguides down to 1 µm × 1 µm with minimal crosstalk [39], enabling up to 200 channels within the same footprint, assuming optimized fabrication. Further scalability can be achieved through multilayer stacking.

4.2. Imaging and relaying a coherent focused light beam

After designing and fabricating the Microimager, the performance of the device was evaluated in different conditions. In the first experiment, the behavior of the Microimager using a coherent light at the input was evaluated. By moving the laser beam along the input pixels (Fig. 6(a)), the Microimager relayed the location of the Gaussian light source at the input as the peak intensity at the output (Fig. 6(b)). In other words, the position of the moving fiber over the Microimager input can be estimated by observing the change in the peak intensity location at the output. The color of the ROIs and the peak intensities are matched.

Fig. 6.

a) Moving the laser beam along the input pixels b) The corresponding output pixels. c) The normalized intensity at the output and the unwrapped output pixels. The intensity peaks show the location of the Gaussian light source at the input.

4.3. Characterizing the Microimager resolution using an optical mask

In this experiment, the Microimager was examined to see if it could resolve the pattern of the optical mask from the input to the output. Figure 7(a) shows the input of the Microimager covered by the optical mask with a 30 µm width. Figure 7(b) shows output pixels corresponding to the optical mask. The intensity of each output pixel is displayed in Fig. 7(c). As can be seen, the Microimager successfully resolved all the dark and light stripes. The output intensities for the remaining stripe sizes are shown in Figure S2. The smallest stripe size that the Microimager could fully resolve was 30 µm. Additionally, the Microimager was able to resolve 87% of the stripes with a width of 25 µm and 75% of the stripes with a width of 20 µm.

Fig. 7.

Resolving pattern: a) The Microimager input is covered with the optical mask. b) The corresponding output pixels. c) The normalized intensity of input and output pixels with the unwrapped output pixels show the Microimager can resolve the bright and light stripes with 30 µm distances.

As a potential application, with the power of fully resolving 30 µm-width features, this Microimager can successfully discriminate the different layers of mice somatosensory cortex in which the thickness of 6 layers ranges from 100 µm to 400 µm [58]. Also, it should be able to differentiate even layers of the hippocampus with thicknesses ranging from 50 µm to 400 µm [59]. If different cell types within these layers are tagged with fluorescent proteins, the Microimager can detect those neurons, allowing us to study their activity and explore potential connections within the other cortical layers.

4.4. Three-dimensional localization and imaging of FL beads

After optical characterization using coherent light and the optical mask, a simplified model using fluorescent green microspheres, also called fluorescent beads (FL beads) embedded in the agar hydrogel block, was employed to validate and characterize the fluorescent imaging capability of the Microimager. FL beads were distributed in the agar block randomly at different distances and depths.

Fluorescent imaging using filters for green light enables us to separate the excitation light from the emitted light, ensuring that the Microimager captures and relays light generated within the fluorescent medium. The green fluorescent light emitted from the fluorescent medium follows a light path that includes entering the device, striking and reflecting off the input micromirror, coupling into the waveguides, and subsequently striking and reflecting off the output micromirror before finally exiting into free space. In this light path, the blue excitation light also accompanies the green emission light. However, by using a green filter, the blue excitation light is filtered before reaching the camera sensor, ensuring that only the emitted fluorescent light, i.e., green light, is detected.

In the first experiment, only one FL bead was positioned over the input. The left image in Fig. 8(a) shows the input without the green filter, where the Microimager input, the FL bead, and the laser beam are all aligned in the same position. The right image in Fig. 8(a) displays the input with the green filter applied, where the FL bead is primarily visible due to its emitted green fluorescent light. To enhance clarity, this image is overlaid with the Microimager image. Figure 8(b) presents the green-filtered output, where the fluorescent light relayed by the waveguides illuminates the output pixels.

Fig. 8.

Fluorescent imaging using one FL bead. a) Microimager input, Left: without filter, Right: with green filter. b) Microimager output with green filter, which relayed the light from the FL bead, c) Comparison of the normalized input and output intensities with the unwrapped output pixels. To enhance visualization, the output intensity curve is smoothed using moving average d) Schematic of FL beads and the scenario of shifted light.

Figure 8(c) compares the input and output intensities after applying the transfer function and normalization. The blue dashed line represents the actual output intensities, and the solid blue line shows the smoothed output intensities, using a moving average filter with three spans to enhance visualization. As can be seen, the correlation between the input and output is evident. However, there is a slight shift in the output intensity towards one side compared to the input, which is attributed to the non-uniform light illumination on the fluorescent bead caused by the oblique angle of the laser fiber. The oblique laser fiber was used because we needed to illuminate the input while simultaneously imaging the output, and we needed to observe the input to adjust the laser beam and position of the FL bead. Additionally, the limited working distance of the objective lens and using laser fiber at the same time, necessitated this oblique arrangement. In the left image of Fig. 6(a), without the green filter, there is a shadow on the right side of the beam caused by the left oblique laser beam. This indicates that the left side of the FL bead received more excitation light while the right side received less. Consequently, the left side emitted more photons than the right side, causing the resultant emission light to shift towards the left side. Figure 8(d) illustrates the schematic of this scenario. Agar gel is a scattering medium due to its semi-transparent and heterogeneous nature. The emitted light scattered within the agar as it traveled to reach the input, intensifying the light shift. This shift, which is an artifact of using the external light source, does not affect the function of the Microimager in the in vivo experiment. We envision allocating one or more of the waveguides to deliver light, serving as excitation light within the tissue. In this case, similar to an endoscope, the light source and input would be positioned next to each other.

The results showed that the fluorescent light successfully traversed the light path within the Microimager. This fluorescent light illuminated only the corresponding output pixels whose inputs were exposed to the light from the FL beads, while the rest of the output pixels remained dark. The Microimager effectively detected the single fluorescent bead in the 2D, XY plane.

Further, two fluorescent beads were used to determine if the Microimager could distinguish between two fluorescent objects at the 3D plane at different XYZ levels. To achieve this, two fluorescent beads were identified in the agar slices at a distance smaller than the input width and then positioned on the input (Fig. 9(a)). In addition to different locations on the XY plane, these beads are at different heights, which is a good model for 3D spatial discrimination. As seen from Fig. 9(b), FL bead 1 was almost at the surface of the device, while FL bead 2 was positioned 70 µm above bead 1. Figure 9(c) shows the corresponding output pixels.

Fig. 9.

Fluorescent imaging using two FL beads. a) The image of two FL beads on the input with a green filter. b) The z location of the two beads c) The output relayed the light from FL beads. d) Comparison of the input and output intensities after applying the transfer function. To enhance visualization, the output intensity curve is smoothed using moving average e) FL bead 1 is moved to the location of FL bead 2. f) The z location of the bead 1 after moving to the location of bead 2. g) Relayed output image after the FL bead 1 was moved. h) Comparison of the input and output intensities of moved FL bead 1 after applying transfer function. To enhance visualization, the output intensity curve is smoothed using moving average.

Figure 9(d) compares the input and output intensities after applying the transfer function and normalization. As shown, there is a clear correlation between the input and output. Like the single FL bead, this demonstrates that the Microimager can detect the presence of the 2 beads in the XY plane. In addition, the amplitude of the FL bead 1 near the surface of the device is higher than that of the bead 2 positioned 70 µm farther away. This shows that the Microimager can also detect differences in the z-level distances. To confirm this effect, the agar slice was moved to bring FL bead 1 to the location of FL bead 2 (Fig. 9(e),(f)), ensuring the same conditions and ruling out other variables. Figure 9(g) displays the corresponding output pixels. To compare the two conditions, the input and output intensities were normalized based on Fig. 9(d) and then plotted in Fig. 9(h). As seen, the amplitude of fluorescent bead 1 near the surface, now positioned where fluorescent bead 2 was, remains high, confirming the sensitivity of the Microimager in distinguishing z-level differences.

In this experiment, not only was the 2D location of the FL beads detectable by the Microimager, but the height from the input surface was also discernible, as beads further from the surface exhibited lower intensity compared to those nearer. Therefore, the Microimager has 3D spatial discrimination capability.

The Microimager waveguides, equipped with micromirrors, exhibit optimal coupling efficiency when paired with coherent light. This efficiency is evident, as more light can enter the device, strike the micromirror, and effectively couple to the waveguides compared to the scattered light that propagates randomly. In experiments using FL beads embedded in agar, the emitted light slightly scattered before reaching the device yet still coupled effectively to the waveguides.

The waveguides exhibit a theoretical acceptance angle of approximately 48°, based on the refractive index contrast between Parylene C (n ≈ 1.639) and PDMS (n ≈ 1.4). This defines the solid angle within which fluorescent emission can be collected. Although isotropic emission inherently limits the overall collection efficiency, employing a waveguide array and optimizing their arrangement increases angular coverage.

4.5. Ex vivo fluorescent imaging of cortical layers in excised brain tissue to show spatial discrimination capability

After characterizing the fluorescent imaging with FL beads, the Microimager was used to image biological tissue. Figure 10(a) shows a live brain slice covering the Microimager input. As mentioned earlier for imaging the optical mask, the Microimager can potentially discriminate different regions of the hippocampus in mouse brain. Hence, we aimed to image fluorescent-expressing regions in the hippocampus. Figure 10(b) compares the input and output intensities after applying the transfer function and normalization. As shown, there is a clear correlation between the input and output. However, the output is slightly shifted to the left due to non-uniform light illumination, similar to what was observed in the FL beads experiment.

Fig. 10.

Ex vivo experiements to show the spatial discrimination using brain slice. a) The GFP brain slice is over the Microimager input, which is overlaid to show the location of output pixels, and the laser beam illuminates the slice. The third input pixel arrangement in Fig. 2 was used. b) The average intensity of ROIs at input and output after applying the transfer function. To enhance visualization, the output intensity curve is smoothed using the moving average.

Some regions of the hippocampus, such as the St. oriens and the St. pyramidale, which covered the center of the input, expressed GFP with higher intensities compared to other layers. In contrast, layer 6 of the cortex, which expressed lower GFP intensity and comparatively appeared darker, covered the right of the input, while the St. radiatum, with moderate GFP expression, covered the left. Therefore, it was expected that the output pixels located in the center would be brighter, those on the right would be darker, and those on the left would show moderate brightness. The Microimager not only met these expectations but also revealed additional details. Specifically, the St. oriens and St. pyramidale—two brighter hippocampus regions—showed distinct intensity peaks. This correlation effectively demonstrates that the Microimager possesses the spatial discrimination capability to differentiate between various parts of the brain, which advances the study of cell type specificity and neural circuit dynamics.

Similarly to FL bead experiments, in ex vivo fluorescent imaging using live mouse brain slices, despite the high scattering nature of the brain tissue, the excitation light was able to propagate through the tissue, generating fluorescent light that successfully traversed the tissue, coupled to the waveguides and relayed to the output pixels. These experiments substantiate that even scattered fluorescent light can effectively couple into the waveguides, demonstrating the versatility and efficacy of the Microimager in both coherent and scattered light environments.

4.6. Ex vivo Ca imaging to show the temporal dynamics in the brain

To demonstrate that the Microimager can detect the temporal dynamics of neural activity, we performed a calcium imaging experiment. Figure 11(a) shows the Microimager experimental setup that was used, and Fig. 11(b) demonstrates the colormap of the Microimager input with annotated ROI locations and the stimulation electrode. In contrast to other experiments, the second input pixel arrangement in Fig. 2(d) was used. Layer 5 of the cortex was stimulated, and the response was seen at layers 2 and 3. The input and output intensity changes (ΔF/F) were calculated based on Eq. (2) and then smoothed using moving and plotted in Fig. 11(c). As can be seen, the intensity changes of the input and output are correlated. The analysis revealed a moderate to strong positive correlation between the input and output intensity changes, with a Pearson correlation coefficient of 0.6. This indicates that the light emitted by neurons at the input, due to evoked calcium activity, is effectively relayed to the output pixels. Therefore, the Microimager can detect the temporal dynamics of brain function through Ca imaging.

Fig. 11.

Ca imaging to show the temporal dynamics in the brain a) Microimager experimental setup for Calcium imaging. b) The colormap of the response region that covers the input and the laser beam that illuminates the slice. The concentric stimulation electrode was used to evoke the calcium activity at the region of interest. The second input pixel arrangement in Fig. 2 was used. c) The intensity change due to Ca activity at input and output. To enhance visualization, the output intensity curve is smoothed using moving average.

The results indicate that the total intensity change at the output is approximately two orders of magnitude lower than at the input, primarily due to optical losses. Integrating across multiple waveguides is used to boost the SNR at the expense of spatial discrimination. Future improvements in the fabrication process to minimize these losses could enhance the SNR of individual output channels, enabling more precise measurement of spatially distributed signals.

In this study, the calcium activity from neurons was recorded in response to electrical neural stimulation. Other neural stimulation modalities, such as optogenetic neuromodulation, can also be used in the future to evoke electrophysiological activity that can be revealed through calcium indicators using our Microimager.

5. Conclusion

In this paper, we discussed the design, fabrication, characterization, and testing of an implantable, flexible, miniaturized endoscope called the Microimager for localized brain fluorescent imaging. The Microimager is implemented on a Parylene photonics platform, which is biocompatible and fabricated using scalable planar microfabrication techniques, enabling customizable design, size, and number of channels (pixels). Operating across the entire visible spectrum, the Microimager functions effectively both for coherent and scattered light. Its compact size, compared to other commercial optical tools for fluorescent imaging, minimizes tissue damage, making it a minimally invasive solution. Additionally, the Microimager can offer a greater number of channels compared with optical fiber bundle endoscopes, and thus enhancing spatial resolution of endoscopes. In the proof-of-concept experiments conducted in this work, the Microimager demonstrated both spatial discrimination and temporal dynamics capabilities. However, a more comprehensive study is needed to characterize and optimize its spatiotemporal resolution. Notably, because of its compact form factor and mechanical flexibility, the Microimager can be implanted into the tissue for chronic biomedical imaging experiments. For example, the Microimager can be implanted into the brain of freely moving animals, where one or more waveguides could be dedicated to light delivery and the rest of the waveguides for collecting and relaying the fluorescent image to image the tissue structure and function. The Microimager discussed in this paper is a groundbreaking innovation that can be used in a gamut of applications involving the imaging of biological tissue structure and function in a minimally invasive way.

Funding

National Science Foundation 10.13039/100000001 ( 1926804).

Disclosures

The authors declare no conflicts of interest.

Data availability

Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.

Supplemental document

See Supplement 1 ^{(1.4MB, pdf)} for supporting content.