Hair-thin confocal fluorescence endo-microscopy for deep-brain in-vivo imaging

TL;DR

This work introduces a hair-thin holographic endoscope that combines a graded-index endcap with a step-index multimode fibre to create a confocal-like detection pathway for fluorescence imaging through a multimode fibre. By using two synchronized digital micro-mirror devices, the system forms diffraction-limited excitation foci and a dynamic annular detection aperture at the distal end, enabling selective rejection of out-of-focus light and substantially improved image contrast and resolution in deep brain imaging. Performance is quantified via 3D PSFs and collection-efficiency measurements, showing up to ~5× suppression of off-focus fluorescence as the confocal factor $w$ decreases, up to practical limits around $3w_{floor}$ where signal drops. In vivo demonstrations in sedated and awake mice illustrate enhanced structural connectivity imaging and neuronal activity monitoring deep in the brain, including motion-robust atlas views and high-sensitivity calcium signals, highlighting the method’s potential for minimally invasive, high-resolution neuroscience imaging at unprecedented depths.

Abstract

Confocal and multi-photon microscopy are widely used for in-vivo fluorescence imaging of biological tissues such as the brain, offering non-invasive access up to ~1 mm depth without major loss in performance. A recently-developed alternative is holographic endoscopy, which exploits controlled light transport through hair-thin optical fibres. With minimal invasiveness, it provides observations at comparable spatial resolution, while extending its applicability to unprecedented depths. It has been used to resolve details of sub-cellular structural connectivity, record neuronal signalling, and monitor blood flow from the deepest locations of the living brain. Yet, its use, particularly in densely labelled brain regions, has so far been constrained by significant contrast loss, primarily due to the absence of a practical mechanism for rejecting out-of-focus fluorescence light -- a capability inherently provided by confocal and multi-photon microscopy. Exploring opportunities in the structure of light modes of different MMF types we identify the possibility of achieving an analogue to confocal fluorescence microscopy through MMF-based endoscopes. Using a novel composite fibre probe that combines graded-index and step-index MMFs, we enable spatially resolved signal collection and selective rejection of out-of-focus light. This confocal filtering significantly enhances image contrast and resolution by suppressing background and off-plane signals. We demonstrate improved imaging performance on fine structural connectivity and intracellular calcium signalling in living mouse brain.

Figures (20)

• Figure 1: The concept of confocal endo-microscopy.a, Ray model of the light propagation through side-view MMF fibre probe. Light from a point at the focal plane ($z=z_0$) is converted by the graded-index (GRIN) fibre endcap to a collimated beam, which is, due to the conservation of the propagation constant, delivered by the step-index (SI) fibre to an annular zone of the proximal far-field. b, Simplified concept of the imaging geometry. DMD1 provides structured excitation light resulting in a diffraction-limited focus scanning along the distal end focal plane. Fluorescent signal is collected backwards through the probe, isolated from the excitation path and spatially filtered at DMD2. Confocal images are formed by signal detected at PMT1, non-confocal images are formed by combining signals from both PMT1 and PMT2. c-e and f-h, Series of images of fluorescent beads (2µm in diameter) obtained in the non-confocal and the confocal regime respectively. For all confocal images, the confocal parameter (width of the virtual aperture annulus) has been set to $w/w_{floor}=1$. Images d and g were taken with the empirically obtained TM. c and f were obtain with numerically recalculated TM for a focal plane shifted by 10µm towards the fibre probe. In contrast, e and h were obtain with a TM, numerically recalculated for a focal plane shifted 10µm away from the fibre probe. All images c-- h are shown with no alpha manipulation. The minimum of the false-colour interval corresponds to the true (calibrated) zero of the PMT detectors, the upper boundary is set such, that the maximum of 0.1% of pixels is saturated. The colour-bar for the confocal modality is calibrated relatively to the non-confocal counterpart to accurately reflect the signal loss due to the confocal rejection.
• Figure 2: Quantitative assessment of out-of-focus signal suppression.a, Lateral and axial point spread function (logarithmic scale). b, Collection PSF for a non-confocal regime and two confocal factors $w$ (linear scale). c, Effective PSF calculated as a product of the excitation and collection PSFs (logarithmic scale). d, Measured average collection efficiency as a function of the distance from the facet for non-confocal and confocal regimes with different confocal factors. e, Measured collection efficiency at the focal point, normalised to that of the non-confocal regime, as a function of the confocal factor. f and g, Measured collection efficiency, normalised to the non-confocal regime, for focal points displaced axially (f, $\Delta z$, where $z_R$ is the Rayleigh length of the excitation focus) and laterally (g, $\Delta r$, measured in units of the half-width at half-maximum of the excitation focus).
• Figure 3: Confocal endo-microscopy in deep-brain structures. a, Simplified experimental arrangement of structural connectivity imaging in anaesthetized animal model. b, d and c, e, Full-field of view non-confocal and confocal records of neurons in a Thy1-GFP-M line. b and c are obtained using the standard (raster) scanning sequence, while d and e are obtained with the 'block-wise' scanning sequence designed to suppress the heart-beat-induced motion artefacts. f and g, Atlas-view non-confocal and confocal records across the brain depth region spanning 4.1mm. To visualise whole extent of the brain in one image, the brightness is locally scaled. h and i, Full-resolution samples from f and g respectively. Panels f -- e, h and i are shown with no alpha manipulation. The minimum of the false-colour interval corresponds to the true (calibrated) zero of the PMT detectors, the upper boundary is set such, that the maximum of 0.1% of pixels in each image is saturated. j, Simplified experimental arrangement of signalling activity monitoring in awake animal model. k--p, Monitoring GCaMP activity of cell somata in a Ai162D$\times$Camk2a-CreERT2 mouse. k, Three strongest principal components of the record shown in RGB channels. l and m, Reference frames with the highest score of the three principal components obtained in the non-confocal and the confocal modality, respectively. n, Selected integration zones for activity monitoring. o and p, Recorded fluorescence signal collected within the integration zones, in the standard $\Delta F / F$ form, i.e. the relative change in fluorescence intensity over the baseline. During the record, the animal model has been exposed to a visual stimulus indicated as black dashes. q--v, The equivalent of k--p for the records of neuronal processes (dendrites). No visual stimulus has been applied. In all studies, the confocal parameter (width of the virtual aperture annulus) has been set to $w/w_{floor}=3$.
• Figure S1: An illustrative ray-model of the light transport for the straight-view terminated composite MMF fibre probe. The medium outside the distal end of the probe (left) is considered to be water, while the medium outside the proximal end (right) is considered to be the air. Individual light paths originate outside the probe near its distal extremity in a series of foci. Each focus lies in the focal plane for which the length of the endcap has been optimised. Each focus is further located at a specific radial distance from the optical axis. The model traces the rays all the way to the proximal far-field plane, passing through the water-filled space, the straight-view-terminated GRIN endcap, the SI segment and the proximal lens surrounded by two air-filled volumes.
• Figure S2: Numerical simulation of light propagation throughout the composite fibre probe, considering single wavelength of $\lambda_{vac}=511nm$. The GRIN fibre forming the endcap has core of 78µm in diameter, on-axis refractive index of 1.45 and NA of 0.29. Light propagation in both fibre types is modelled as linear superposition of propagation invariant modes Ploschner2015NPBoonzajerFlaes2018PRL. The endcap length is set to 248µm, which leads to the working distance of 50µm. The SI MMF has core of 50µm in diameter and the NA of 0.22. The length of the SI segment is set to 4mm. The focal length of the focusing lens (indicated in the geometry by a pair of dashed lines) is 500µm. Each of the RGB channels represents intensity distribution originating in a focus at different radial position across the focal plane. The light distribution is 'filtered' by the composite fibre, i.e., only light signal that is transported through the probe is visible. The imaged volume is compressed along the direction of the fibre axis by factor of 2. The image is computed in spherical projection, placing camera 1.5mm in front of the probe's distal facet and 3mm off the probe's axis.