High-speed light-sheet microscopy for the in-situ acquisition of volumetric histological images of living tissue

Abstract

Histological examinations typically require the excision of tissue, followed by its fixation, slicing, staining, mounting and imaging, with timeframes ranging from minutes to days. This process may remove functional tissue, may miss abnormalities through under-sampling, prevents rapid decision-making, and increases costs. Here, we report the feasibility of microscopes based on swept confocally aligned planar excitation technology for the volumetric histological imaging of intact living tissue in real time. The systems’ single-objective, light-sheet geometry and 3D imaging speeds enable roving image acquisition, which combined with 3D stitching permits the contiguous analysis of large tissue areas, as well as the dynamic assessment of tissue perfusion and function. Implemented in benchtop and miniaturized form factors, the microscopes also have high sensitivity, even for weak intrinsic fluorescence, allowing for the label-free imaging of diagnostically relevant histoarchitectural structures, as we show for pancreatic disease in living mice, for chronic kidney disease in fresh human kidney tissues, and for oral mucosa in a healthy volunteer. Miniaturized high-speed light-sheet microscopes for in-situ volumetric histological imaging may facilitate the point-of-care detection of diverse cellular-level biomarkers.

Over 6 million biopsy procedures are performed in the United States every year¹. However, the standard practice of excising, fixing and staining tissue for histopathological evaluation is costly and slow, delaying treatment while being confounded by sampling error. Although intraoperative frozen sections can provide results within 20 min, tissues suffer from freezing artefacts, poor quality sectioning, swollen cell morphologies and poor staining², with fatty tissues such as brain being particularly difficult to freeze and cut. The use of physical 2D histology slides also impedes evaluation workflow because pathologists must track features through multiple sections to gain a better understanding of 3D tissue morphology. Physical slides must be manually viewed through a microscope or else digitized prior to viewing, which requires additional time and resources.

Most importantly, however, both frozen and standard histology require physical removal of living tissue. For precious tissues such as the eye, heart or brain, conservative biopsy can lead to undersampling and either misdiagnosis or incomplete surgical resection³. The destructive nature of biopsies also means that they are almost never used for general surgical guidance, such as identification of tissue types or for screening large areas of the body. Ex vivo tissues also quickly lose features such as perfusion level and metabolic state, which could provide valuable biomarkers of tissue health or disease state.

An ideal solution to these challenges would be a microscope that could capture images of tissue cytoarchitecture in vivo at the point of care. Confocal microendoscopy utilizes confocal scanning through a fibre-optic conduit to generate 2D images of in situ tissues, and can be achieved through the channel of an endoscope. However, current commercial embodiments of confocal microendoscopy rely on systemic injection of bright fluorescent dyes such as fluorescein to provide contrast, while their ability to only capture 2D images over a small field of view has proven challenging to interpret reliably^4,5. Although specificity could be improved with fluorescent markers that can selectively highlight disease^6,7, regulatory approval of such agents can be prohibitively costly and complex in most cases. Two-photon fluorescence, second harmonic generation, fluorescence lifetime and stimulated Raman spectroscopy have also been demonstrated for both in vivo and bedside fresh-tissue imaging and have revealed impressive intrinsic or ‘label-free’ contrast^8–12. However, all of these approaches have limited acquisition speed, making them intolerant of in vivo motion and preventing real-time large-area or 3D imaging, while their reliance on costly and/or high-power pulsed laser sources has thus far restricted their use for in vivo clinical imaging, with a few exceptions^12–14.

Here, we report the development and applicability of an in vivo histological imaging methodology and microscope, which we named MediSCAPE, based on swept confocally aligned planar excitation (SCAPE) microscopy^15–19 that permits rapid, 3D, label-free, non-destructive in situ examination of living intact tissues on a microscopic level. MediSCAPE improves upon previous intravital microscopy approaches by leveraging light-sheet-based illumination to achieve high sensitivity with ultrafast 3D image acquisition (>10 volumes per second (VPS)). High sensitivity permits imaging of cellular features via weak intrinsic fluorescence contrast without the need for exogenous contrast agents. Real-time 3D imaging allows roving acquisition that tolerates in vivo motion and enables stitching, and thus multiscale surveying of 3D cellular structures across large tissue areas. We demonstrate the ability of MediSCAPE to image a wide range of healthy and diseased in vivo and freshly excised mouse and human tissues, comparing the normal and diseased structures visualized to gold standard histology. We also show the ability of MediSCAPE to capture in vivo functional features of living tissues, such as perfusion and responses to ischaemia and reperfusion. In addition to results obtained with a larger form-factor benchtop MediSCAPE prototype, we demonstrate near-equivalent performance using a miniaturized MediSCAPE design with a form factor amenable to intraoperative human use. The system is compact, relatively low cost and uses affordable, visible, continuous wave laser light sources (488 nm and 637 nm), with illumination levels equivalent to US Food and Drug Administration (FDA)-approved confocal microendoscopy.

This work shows the potential of MediSCAPE to become a rapid and low-cost alternative to conventional histopathology. Moreover, by providing real-time intraoperative feedback, MediSCAPE could enable closed-loop treatment decisions, including iterative assessment of surgical margins and more comprehensive surveillance of large tissue areas to guide biopsy site selection^20–22 without the need to permanently damage or remove tissues that could otherwise be conserved. The non-destructive nature of MediSCAPE could also make it valuable for applications that would generally avoid using histopathology, such as ‘tissue typing’ or perfusion assessment during robotic or orthopaedic surgeries. Rapid in situ histopathology could also be transformative for the evaluation of organs donated for human transplant, particularly kidneys, which are the most commonly transplanted and suffer from high inter-observer variability²³. Our diverse demonstrations highlight the advantages of MediSCAPE for intravital microscopy-based intrasurgical guidance.

Results

MediSCAPE system designs, imaging geometries and image acquisition modes.

Data for this study were collected using both benchtop and miniaturized MediSCAPE prototypes. The single-objective light-sheet imaging geometry and basic layout of these MediSCAPE systems are shown in Fig. 1a–f, with additional system details and characterization provided in Supplementary Note 1. In all cases, MediSCAPE uses an oblique light sheet to illuminate the sample and collects emitted fluorescence back through the same single, stationary high-numerical aperture (NA) objective lens. A galvanometer mirror within the system both sweeps the light-sheet from side to side (along x) and descans the returning fluorescence, mapping it onto a stationary conjugate oblique image plane, which is then focused onto a scientific complementary metal–oxide–semiconductor (sCMOS) camera (Andor Zyla 4.2+). Planes corresponding to oblique yz’ sections are acquired by the camera as the galvanometer mirror sweeps the sheet in x to generate a volumetric image. Since all components of the system remain stationary except the galvo mirror, which sweeps at one line per volume, imaging speed is limited only by camera read-out rate. The volume acquisition rate is thus determined by the number of x-steps spanning each volume, as well as the number of camera rows acquired on the camera (which corresponds to the depth in z’ imaged), with fewer rows permitting faster read-out (for example, ~2,000 frames per second for 100 rows on a standard sCMOS camera). In all systems, dual colour imaging was achieved using a home-built image splitter which spectrally separates the image into two images and positions them side by side across columns of the camera chip, permitting parallel imaging of both emission bands without compromising imaging speed.

Fig. 1 |. MediSCAPE system designs and image formation.

a, Geometry of the benchtop MediSCAPE system’s single-objective light-sheet excitation and detection. b,c, Basic optical layout of the benchtop system (b) and 3D schematic of benchtop system hardware (c). O1,2,3 are primary, secondary and tertiary objective lenses respectively; SL, scan lens; TL, tube lens; CL, cylindrical lens; PL, Powell lens. The slit permits adjustment of the light sheet’s NA. d, Basic optical layout of miniaturized MediSCAPE system. SMF and Asph denote single-mode fibre and aspheric lens fibre launch. e, 3D schematic of miniaturized system hardware (same scale as c). f, Illustration of hand-held miniaturized design with image relayed to camera via fibre bundle (black cylinder). g, Sterilizable imaging cap for clinical system holds water immersion interface with glass window at suitable working distance for tissue immobilization. h, yz’ slices are collected along the scan direction (x) to create an oblique volume. Autofluorescence (AF) excited at 488 nm is split into dual emission channels, displayed in yellow hot and blue colourmaps (see Extended Data Fig. 1). i, 3D rendering (ImageJ 3Dviewer) of same volume (tilt has been corrected (z’→ z)). j,k, Single xy slices from volume in i at two different depths. Depths are indicated on the bottom left corner as the distance from the top of the tissue surface. l, H&E histology of a similar region in the mouse kidney cortex. m, Miniaturized MediSCAPE results on fresh ex vivo wild-type mouse kidney xy (top) and yz (bottom) cross-sections. n(i,ii), A variable focal length tube lens can be used to easily switch between low (i) (×4.6) or high (ii) (×11.4) magnification as shown here for the ex vivo fresh mouse kidney tissue. See Supplementary Movie 1 for depth flythroughs. White dashed lines in m and n indicate the position of the yz and xz cross-sections shown in the corresponding xy cross-section images. All scale bars, 100 μm except in n (20 μm).

For initial proof-of-concept studies and comparisons to gold standard histology, we used benchtop inverted-configuration MediSCAPE prototypes similar to our previously reported SCAPE 2.0 designs characterized in ref. ¹⁷, with dual 488 and 637 nm excitation lasers and a 3-axis motorized stage when needed (Fig. 1a–c and Supplementary Note Tables 3 and 4). For in vivo clinical use, the first requirement was a more compact, light-weight layout with a narrower and longer imaging head that would permit manoeuvring in the surgical field without obscuring the surgeon’s access to the field. As shown in Fig. 1d–f, this feature was achieved in our miniaturized MediSCAPE prototype by unfolding the system’s usually orthogonal telescopes and positioning the galvanometer mirror within the primary elongated beam path. A smaller-diameter 60 × 1.0-NA water immersion primary objective lens (O1) was chosen, while the 2” diameter lenses used in the benchtop system were replaced with 12 mm diameter optics. To provide more mechanical stability to alignment, the laser illumination is introduced via a single-mode fibre and then directed into objective O2, simplifying the imaging head while enabling the image rotation objectives and laser launch to all be rigidly mounted on a plate distal to the imaging head. Another important component of this system is the imaging cap shown in Fig. 1g, which provides required water immersion, but also precision spacing of the primary objective (O1) from the tissue being imaged. This cap is essential for imaging unconstrained in vivo human tissues, as it could be pressed against the tissue to provide stability and the ability to scan while maintaining the tissue at the required working distance. This cap is also sterilizable and could ultimately be disposable for patient protection.

The prototype of this miniature design was assembled using off-the-shelf components and is illustrated in Fig. 1e (approximately to scale compared with Fig. 1c, see also Supplementary Note Table 2). Figure 1f shows a depiction of a further refined layout that would use a fibre-optic imaging bundle to position the system’s image splitter and camera remotely. Supplementary Note 1 provides further details of this miniaturized system design, including resolution characterization, component lists and extensions and customizations that could permit further miniaturization. Experimental and theoretical characterizations reveal equivalent or even superior resolution and light efficiency of the miniaturized system compared with benchtop systems, with resolution close to the beam waist of 0.811 ± 0.123 μm (y), 1.07 ± 0.115 μm (x) and 2.10 ± 0.479 μm (z), with the only trade-off being a modestly reduced field of view (~1 mm × 1 mm xy for systems A (Supplemental Note Table 3) and B (Supplemental Note Table 4) vs ~0.4 mm × 0.6 mm xy for the miniature system (Supplemental Note Table 2).

MediSCAPE’s method of 3D image formation is illustrated in Fig. 1h–k using images obtained on the in vivo kidney of a heavily anaesthetized wild-type mouse using the benchtop prototype (see Methods, and for full imaging parameters for all images shown, see Supplementary Table 1). Autofluorescence in the tissue was excited with 488 nm light and dual emission channels were collected simultaneously through 525/50 nm and 618/50 nm bandpass filters, shown here using blue and ‘yellow hot’ colourmaps that allow better visualization of the overlapping channels (see Extended Data Fig. 1 for more details). Figure 1h shows the individual yz’ planes captured by the camera as MediSCAPE sweeps the oblique light sheet along x. If these optically sectioned images are simply stacked, they form a 3D volume as shown in Fig. 1i (here, skew-correction has been applied to remove the effect of the oblique light sheet; see Methods). Figure 1j,k shows two individual xy planes from this 3D dataset whose overall volume size was 802 × 861 × 275 μm³ sampled at 1.0 × 1.4 × 1.1 μm³ voxel⁻¹ in 0.78 s.

Conventional histological sections stained with hematoxylin and eosin (H&E) from a similar region of a mouse kidney cortex shows normal tubular architecture (Fig. 1l). MediSCAPE images show similar structures, demonstrating that tubules in the in vivo mouse kidney show robust autofluorescence in both emission channels, with proximal tubules showing higher emission at ~525 nm (yellow hot) than distal tubules (blue/purple), with signal probably coming from flavins, elastin, porphyrins and lipofuscin (see Supplementary Table 2). Nuclei can be distinguished along the tubule walls as punctate dark regions. Figure 1m shows equivalent data acquired with the miniaturized MediSCAPE prototype on a freshly excised wild-type mouse kidney, both as a single xy plane and a yz depth plane (sampled from a full 3D volume). Similar features, resolution, contrast and signal to noise can be appreciated, while also showing the system’s more restricted field of view.

Although cellular features can be seen at this sampling density, we note that MediSCAPE has the ability to trade-off resolution with field of view to ‘zoom in’ on features of interest. Figure 1n(i,ii) and Supplementary Movie 1 show fresh mouse kidney data acquired using a variable 70–200 mm focal length tube lens as TL3¹⁷, switching between regular (×4.6) and high magnification (×11.4). This ‘zoom in’ feature reveals crisper and more detailed visualization of the tubular structure and could be readily automated and co-registered with coarser imaging over larger fields of view acquired at lower magnification. MediSCAPE’s resolution can also be quickly improved by increasing the light-sheet NA (for example, to 0.1–0.13), although this narrows the range of depths with high resolution to ~50–100 μm. Most of the images shown here were thus acquired with a relatively low light-sheet NA (~0.06–0.08 NA) to provide a longer ‘depth of field’ (~100–300 μm). It should be noted however, that the penetration depth achievable with MediSCAPE also depends heavily on tissue scattering as discussed further below.

Label-free detection of histological features in diverse tissue types.

Autofluorescence in living tissue can allow visualization of many morphological features routinely used for histological evaluation (see Supplementary Table 2 for a detailed list of sources of tissue autofluorescence). Furthermore, the distribution and concentration of intrinsic fluorophores, such as elastin and flavin adenine dinucleotide (FAD), provide a rich range of molecular information that can potentially indicate changes in tissue health even before structural changes become visible^10,24,25. Label-free imaging in humans is especially valuable because in vivo use of exogenous dyes is limited by safety restrictions, the time, complexity and cost of obtaining FDA-approval, limited dye penetration, heterogenous staining and time sensitivity of dye administration in a clinical setting.

To characterize the tissue structures that are visible with autofluorescence, we imaged a wide variety of freshly excised mouse tissues with both our benchtop (Fig. 2) and miniaturized (Extended Data Fig. 2) MediSCAPE systems. Figure 2 shows xy lateral slices at two different depths compared with H&E histological sections showing the same or adjacent regions in the mouse tissue. Supplementary Movies 2–5 show complete depth flythrough movies of each 3D volume for both benchtop and miniaturized MediSCAPE systems. The furthest depth at which histology-level resolution is possible is tissue dependent and also varies with excitation wavelength. For many tissues, resolution starts to degrade after 50 μm with 488 nm excitation, yet in skeletal muscle, for example, cellular-level contrast is observed 121 μm into the tissue (Fig. 2f).

Fig. 2 |. Label-free imaging of a variety of fresh mouse tissues with MediSCAPE.

En face view of autofluorescence in a range of fresh intact mouse organs imaged using a benchtop MediSCAPE system. xy slices are shown from a single volume, with depth from the tissue surface indicated on the bottom left of each slice. H&E histological sections from the same or adjacent regions are shown for comparison. a, Cardiac muscle fibres in the heart ventricle. b, Alveoli and visceral pleura in the lung. c, Cerebellum in a sagittally cut surface of the brain. d, Classic hepatocyte cord formations and capsule in a liver lobule. e, Red pulp and the surrounding capsule in the spleen. f, Muscle fibres visible deep within thigh muscle. g, Superficial layers in the bladder mucosa, with pixel intensities shown on a log scale for better visualization. h, Crypts of Lieberkühn in the colon mucosa. i, Uniform organization of acinar cell structures in normal pancreas. j, Pancreatic ductal adenocarcinoma tumour (KrasLSL^.G12D/+; Trp53LSL.R172H^/+; Pdx1-Cretg^/+ (KPC) mouse²⁷), with chaotic distribution of cells and clear cystic structures below the tumour surface. See Supplementary Movie 2, for flythroughs of lateral cross-sections in the first 50 μm depth for tissues a–d and Supplementary Movie 3 for flythroughs of the first 100 μm depth for tissues e–h. See Supplementary Movie 4 for a flythrough of pancreas data in i and j. Supplementary Movie 5 and Extended Data Fig. 2 show a subset of the same tissues imaged with the miniaturized MediSCAPE system. All images were acquired at a volume size of 801 × 1,065 × 275–330 μm³ in xyz, with a sampling density of 1 ×1.4 ×1.1 μm³ per voxel, respectively, at 100 fps, with 5–7 mW of laser power at the sample. All scale bars, 100 μm.

With solely intrinsic contrast, micron-scale structures were visible in all fresh tissues studied and corresponded well to structures visible in H&E histological sections. The shape and diameter of crypts of Lieberkühn, for instance, can be clearly distinguished in the colon mucosa in Fig. 2h and Extended Data Fig. 2d,e. Alveoli in lung tissue are distinctly lined with intensely fluorescent elastin (Fig. 2b) and can be assessed intact, whereas histology often shows major distortions from sectioning of the delicate air-filled tissue. Layers within tissues, such as the bladder mucosa (Fig. 2g), are distinct and can be assessed in 3D, allowing more comprehensive evaluation than 2D histology sections and single-plane confocal microendoscopy. Figure 2i shows healthy mouse pancreas revealing a normal, uniform pattern of acinar structures interspersed with bright stellate cells (caused by vitamin A droplets²⁶). Figure 2j shows data acquired on a large pancreatic ductal adenocarcinoma (PDAC) freshly resected from a Kras^LSL.G12D/+; Trp53^LSL.R172H/+; Pdx1-Cre^tg/+ (KPC) mouse²⁷. Here we see a chaotic distribution of cells, paucity of mature blood vessels and the presence of a clear cystic structure below the tumour surface, consistent with the H&E histological section of the same tumour also shown (Supplementary Movie 4).

To compare the autofluorescence contrast captured by MediSCAPE to standard point-scanning confocal and two-photon microscopy, we imaged fresh mouse colon mucosa and kidney samples with all three techniques, as shown in Supplementary Note 2 Fig. 1. Cellular and tissue-level features were similar across all three techniques, but the point-scanning methods required prohibitively long acquisition times for weak intrinsic fluorescence as explained in Supplementary Note 1 and Supplementary Note Table 1, and highlighted by imaging parameters in Supplementary Note Table 5.

High-speed 3D roving acquisition permits image stitching over large motile fields of view.

A key feature of MediSCAPE is its very fast 3D imaging speed, even when imaging weak autofluorescence. This speed can be leveraged to allow exploration of large areas of tissue by ‘roving’ or continuously moving the tissue relative to the system’s 3D field of view. The speed of MediSCAPE can tolerate this translation without substantial artefacts in each individual volume, and since each volume has some spatial overlap with the last, a sequence of volumes can be stitched to generate a fully contiguous 3D strip of data spanning millimetres or more. This feature does not require continuous or motor-controlled movement and can tolerate unavoidable in vivo movements such as breathing, making it ideal for evaluating transitions between tissue types, or exploring heterogenous regions for multiscale spatial patterns at the cellular and mesoscopic level.

Supplementary Movie 6 illustrates this real-time roving and stitching process using high-speed MediSCAPE data acquired on the exposed in vivo kidney of an anaesthetized mouse positioned as shown in Fig. 3a. During imaging, the mouse was manually translated along 3 dimensions to mimic how a MediSCAPE imaging probe would rove over intact in vivo tissue (Fig. 3b). Dual colour volumes of 358 × 798 × 165 μm³ each in xyz, with a sampling density of 2.5 × 1.4 × 1.1 μm³ voxel⁻¹, were acquired at 9.3 VPS, while continually roving 3 mm across the intact kidney cortex surface. This roving data were then computationally stitched to generate a contiguous 3,000 × 798 × 165 μm³ volume using the pairwise stitching plugin in ImageJ²⁸ (see Methods). Although this stitching was performed offline here, real-time processing could be achieved on a field programmable gate array (FPGA). Extended Data Fig. 2e shows similar label-free stitched-roving results acquired at 11.2 VPS using our miniaturized MediSCAPE prototype on fresh mouse colon mucosa.

Fig. 3 |. In vivo mouse, label-free roving acquisition of pancreatic disease and the beating heart.

a, Anaesthetized mice were positioned on a glass bottomed dish with desired organs exposed and imaged with an inverted benchtop MediSCAPE system. b, During high-speed 3D imaging, mice were translated on a manual stage to rove over areas of interest, enabling successively acquired volumes to be stitched together. c, Preparation of a KPC mouse²⁷ with a large PDAC for MediSCAPE imaging (left: abdominal incision, right: view from under coverglass). d, xy slice from an in vivo label-free 4.4 mm stitched roving scan of accessible pancreatic tissue. e, Unregistered H&E histological section from tumour-adjacent tissue, interspersed with inflammation, low-grade pancreatic intraepithelial neoplasia (PanINs) and areas of acinar-to ductal metaplasia (ADM). Similar features are seen with MediSCAPE data in d, in which green patches of acinar tissues are visible, interspersed with disordered inflammation punctuated with bright dots resembling stellate cells. f, Successive depths within d (white dashed box) show the 3D shape of dark voids, suggesting ADM or the lumens of PanINs rather than a continuous duct (arrows labelled 1 and 2) and a possible large pancreatic duct (arrow labelled 3). g, Normal pancreas from a wild-type mouse, imaged with MediSCAPE (fresh tissue, autofluorescence) and H&E shows uniform acini with interspersed stellate cells, bright with droplets of vitamin A. All scale bars, 200 μm. h, In vivo label-free MediSCAPE imaging of the beating heart: xy slices at three depths in a stitched volume, roving primarily along the x axis. White arrows, vessels; blue arrows, elastic fibres in the pericardium. i, xyz cross-sections show an ROI (white dashed box in h) in more detail. Yellow arrows, stitching along z; red arrows, volumes successfully stitched during cardiac pulses at 0 and 2.6 s. To the right, time blocks along x indicate each volume’s time of acquisition. Scale bar, 100 μm. j, A kymograph shows the maximum intensity projection (MIP) over the x and z axes of each volume acquired over the full 15.6 s of imaging. Cardiac motion events can be observed as lateral y movements of the tissue (red arrows). See Supplementary Movie 8 for real-time playback of volume acquisition and stitching.

Figure 3c–f shows another in vivo label-free stitched-roving example acquired on a Kras^LSL.G12D/+; Trp53^LSL.R172H/+; Pdx1-Cre^tg/+ (KPC) mouse²⁷ with a large pancreatic ductal adenocarcinoma (PDAC). PDAC is one of the deadliest forms of cancer, having rapid progression and a 5 year survival rate of just 10%²⁹. Accurate biopsy and diagnosis can be challenging, while surgical resection of pancreatic cancer is complex owing to the anatomic complexity of the area and the frequent involvement of large blood vessels. Data shown in Fig. 3d,f were obtained by in vivo roving over exposed tissues in the mouse’s abdomen during 10 VPS MediSCAPE imaging (galvanometer-scanning over a volume size of 254 × 920 × 184 μm³ with a sampling density of 2.4 × 1.1 × 1 μm³ voxel⁻¹). Images are shown in a native magenta/green colour scheme in which perfused blood vessels show clearly in pink, while vitamin A droplets in stellate cells appear yellow/white and metabolically active cells appear green. Compared with normal pancreas (Fig. 3g), this scan spanning 4.4 mm of complex in situ tissues revealed key signs of pancreatic disease including inflammation, vascularity, bright stellate cells and well-vascularized patches of acinar cells with signs of acinar-to-ductal metaplasia. While relatively low resolution in this form, these images demonstrate how larger-scale stitched roving data can reveal mesoscale disordered structure that would not be visible in small 2D fields of view. Imaging smaller 2D regions of interest in this case could not have captured the mesoscale heterogeneity of this tissue which, itself, is a primary marker of pancreatic disease. The abnormal structures of the bulk of the excised tumour were also clearly distinguishable from normal pancreas (Fig. 3g). These results support the potential value of MediSCAPE for guiding complex biopsy site selection for surgical resection of pancreatic cancer.

To further demonstrate the tolerance of MediSCAPE imaging to inherent in vivo motion, we also imaged the beating intact in vivo heart of a wild-type mouse as shown in Fig. 3h–j. Data were acquired by roving across the exposed heart surface while continuously acquiring dual colour volumes at 12.9 VPS (galvanometer-scanning over a volume size of 305 × 798 × 138 μm³ with a sampling density of 2.5 × 1.4 × 1.1 μm³ voxel⁻¹). Figure 3h shows xy slices within a 3D-stitched field of view created from 15.6 s of data acquired while using the 3-axis stage to manually rove over cardiac tissue. Striated cardiac muscle cells in the myocardium are clearly visualized, while elastic fibres can also be seen on the surface of the myocardium (blue arrows). Granular autofluorescence along the muscle fibres is probably lipofuscin, a lipopigment that accumulates in highly active cells over time.

Periodic cardiac pulses that occurred during this acquisition appear as sudden lateral movements along the y axis in the kymograph shown in Fig. 3j, where a maximum intensity projection of x and z is shown over 15.6 s of imaging. Even during periods of motion (indicated with red arrows in Fig. 3i,j), MediSCAPE was able to acquire volumes that could be successfully stitched together, with minimal visible motion artefact or blur. Supplementary Movie 8 shows real-time playback of the cross-sections of the beating heart, as well as stitching of these volumes as they are acquired. Note that stitching 3D tissue volumes compensates for tissue motion in all 3 dimensions and allows roving laterally and along the depth axis. Compared with stitching traditional 2D fields of view, volume stitching could more reliably reconstruct inherently 3D tissue structures, correcting for out-of-plane motions, which are unavoidable in vivo.

MediSCAPE detection of histologic features of disease states in human tissue.

To test the ability of MediSCAPE to capture disease-related features in human tissues, we obtained fresh human kidney tissue from surgical nephrectomy and compared MediSCAPE imaging results to conventional periodic acid-Schiff (PAS) and H&E histology on the same samples.

Figure 4 shows autofluorescence imaged by MediSCAPE in a nephrectomy specimen from a patient with underlying chronic kidney disease (CKD) (using a yellow/blue colour scheme for visualization). The entire flat face of the fresh-tissue specimen was imaged using a benchtop MediSCAPE system with ‘stage-scanning’, in which the system’s galvanometer mirror was kept stationary and a motorized stage holding the sample was moved at constant velocity across x in strips (then tiling in y) to acquire and stitch a full 13.3 × 10.6 × 0.3 mm³ volume (total acquisition time 196 s). Supplementary Movie 9 shows a depth flythrough movie of the lateral cross-sections in the full stitched volume. From the full stitched volume, a 2.1 × 1.6 mm² xy region of interest (ROI) is shown in Fig. 4a, with the corresponding area on PAS histology shown in Fig. 4b for comparison.

Fig. 4 |. Key diagnostic features in fresh human kidney tissue from a patient with chronic kidney disease.

a, xy slice from a stitched field of view acquired by stage-scanning. Scale bar, 200 μm. Dashed-line bounding boxes indicate key features shown in further detail in subsequent panels. b, PAS histology of the same region taken at a slightly lower depth. Scale bar, 200 μm. c, Artery showing features of moderate arteriosclerosis, including intimal thickening and reduplication of the elastic lamina, the latter highly autofluorescent as imaged by MediSCAPE and positive on PAS stain (white arrows). d, Glomerulus demonstrating the distinction between capillary tuft (yellow arrow), Bowman’s space (grey arrow) and Bowman’s capsule (white arrow). e, Globally sclerotic glomerulus with focal hyalinosis and with an arteriole showing punctate perinuclear autofluorescent signal in myocytes, possibly representing lysosomes (white arrow). f, Intensely autofluorescent tubular casts (white arrows), similar in morphology to PAS-stained Tamm Horsfall casts seen on another level (tissue distortion and tearing during histological processing prevented exact spatial correlation with histology). g, Tubules showing severe atrophy, demonstrated by small luminal diameter with surrounding increased interstitial fibrosis and accumulation of autofluorescent granular cytoplasmic material in tubular epithelium, probably lipofuscin. h, Residual functional proximal tubules showing pseudohypertrophy with increased diameter, no significant lack of cytoplasmic volume, and prominent autofluorescent cytoplasmic signals, possibly attributed to increased metabolic activity of these epithelial cells. Scale bars, 30 μm. See Supplementary Movie 9 for a flythrough of depths in the full stitched volume.

Examples of key diagnostic features identified by MediSCAPE are highlighted in Fig. 4c–h. Clinically relevant vascular changes could be clearly identified, including arteriosclerosis and arteriolar hyalinosis (Fig. 4a,b). Identification of arteries is aided by strong autofluorescence of the internal elastic lamina of arterial walls, which are even more prominent on MediSCAPE imaging in the setting of hypertensive arteriosclerosis in which there is luminal narrowing by intimal thickening, with reduplication of the elastic lamina (Fig. 4c, note arrows). We could clearly identify glomeruli and distinguish those showing global sclerosis (Fig. 4d,e). We could also discriminate subglomerular structural elements including the glomerular capillary tuft, Bowman’s space and Bowman’s capsule, especially when the latter had undergone partial sclerosis (Fig. 4d, note arrows). In addition, we could identify several glomerular features related to CKD including segmental glomerulosclerosis, focal hyalinosis and nodular mesangial sclerosis (data not shown). In the tubulointerstitial compartment, characteristic chronic changes including tubular atrophy and interstitial fibrosis, known to have the strongest correlation with renal outcomes, were clearly evident in MediSCAPE images (Fig. 4f–h). We could distinguish atrophic from non-atrophic tubules, and identify pseudohypertrophy of proximal tubules and tubular casts. Extended Data Fig. 3 provides additional examples of histologic features identified in a label-free human kidney biopsy from a diabetic patient.

Figure 5a–d highlights the advantages of 3D imaging with an example of a clinically relevant lesion that can be ambiguous to identify from 2D thin sections (present within the larger stitched field of view shown in Supplementary Movie 9). In xy planar images, a small cyst-like structure is evident which, on a single image could be either a severely dilated atrophic tubule or a simple renal cyst. However, the fully 3D MediSCAPE dataset reveals a residual compressed and sclerosed capillary tuft, pressed against the internal wall of the cyst-like space, distinguishing this structure as an atubular glomerulus (or ‘glomerular microcyst’) rather than any type of tubular-derived element. Examining the perirenal fat from normal human kidney tissue, we found that MediSCAPE can also capture the 3D arrangement of elastic fibres and adipocytes on the basis of their intrinsic autofluorescence (Extended Data Fig. 4). Evaluating these features in volumetric space could allow more accurate assessment of fat content, as well as the structure, density and identity (for example, elastic versus collagen) of fibres in different tissue compartments.

Fig. 5 |. MediSCAPE 3d views of an atubular glomerulus in fresh kidney biopsy from a patient with CKd.

a–d, yz cross-sections (a) taken from a 3D rendering (b) (ImageJ 3DViewer) of an atubular glomerulus found within the full stitched volume. The origin of each depth and lateral cross-section is indicated with dotted lines in b. c, xz cross-sections. d, Lateral xy cross-sections. All scale bars, 100 μm. A depth flythrough of lateral planes in the glomerulus can be seen within the full stitched volume in Supplementary Movie 9.

MediSCAPE imaging of fresh human tissue using topical stains.

Although MediSCAPE can image clinically relevant features using autofluorescence alone, fluorescent contrast agents can be readily imaged if available. Extended Data Fig. 5a shows an example of MediSCAPE data collected from a sample of fresh normal human kidney stained with proflavine, a topical nuclear dye commonly used in clinical imaging research³⁰. Proflavine and red autofluorescence emission were acquired with 488 nm excitation using stage-scanning to create a 7,500 × 918 × 164 μm volume in 5.6 s. Proflavine staining reveals nuclear size, shape and distribution, while autofluorescence provides complementary structural information. Following recent conventions for visualization of tissue fluorescence in histopathology³¹, in Extended Data Fig. 5b–d we show this dual colour MediSCAPE data with a pseudocolour H&E colour scale, using proflavine as a haematoxylin analogue (purple) and with eosin represented by the autofluorescence signal collected with a 618/50 nm bandpass filter (pink). Fully 3D pseudocoloured MediSCAPE images closely resemble conventional brightfield H&E histology (shown from an adjacent area in Fig. 5d) and can allow easier evaluation of nuclear detail when needed. Supplementary Movie 10 shows a flythrough of the top 30 μm depth of the pseudocoloured 3D MediSCAPE volume.

Extended Data Fig. 6 shows three additional MediSCAPE scans of normal human kidney tissue showing signs of aging from another patient undergoing a lumpectomy of adjacent tissue, stained with nuclear dyes proflavine or methylene blue. We found that all four renal histologic compartments commonly assessed through a combination of H&E and PAS histology could be clearly distinguished in MediSCAPE images, especially with methylene blue staining, including glomeruli, arteries, tubules and interstitium. Supplementary Movie 11 shows a 3D rendering and flythrough of depth and lateral cross-sections of the proflavine-stained human kidney tissue shown in Extended Data Fig. 6d–h, demonstrating the full complex 3D structures of these histologic features.

Extended Data Fig. 7 directly compares staining with proflavine, methylene blue and fluorescein sodium to autofluorescence and H&E in fresh mouse colon mucosa. These results demonstrate the ability of MediSCAPE to rapidly capture 3D images of a variety of exogenous contrasts with high signal to noise, but also highlights the challenges of ensuring dye penetration compared with exploiting intrinsic contrast.

In vivo label-free human imaging of the oral cavity with MediSCAPE.

The demonstrations above confirm the ability of MediSCAPE to visualize a wide range of multiscale tissue features with broad relevance to intraoperative guidance. To further confirm the feasibility of this new approach for clinical intraoperative imaging, we acquired in vivo human data in the oral cavity of a healthy adult volunteer. Both our miniaturized and benchtop MediSCAPE systems were used for comparison, and both were fitted with coverglass-tipped imaging caps as depicted in Fig. 1g. These custom-fabricated caps provide both sterility and essential stabilization of the tissue being imaged, ensuring the optimal working distance for the objective to capture a 200–300 μm depth range into the tissue while maintaining a water immersion interface for the lens.

Label-free roving scans of the tongue, inner and outer lip were acquired by asking the adult subject to position the appropriate tissue onto the imaging cap and to slowly move its position during continuous volumetric imaging for up to 120 s at 3–5 VPS (see example run in Supplementary Movie 13). These roving scans were stitched into a continuous, large 3D volume. Data from both MediSCAPE systems, shown in Fig. 6, consistently reveal features of the layers of the oral tissues, including different types of tongue papillae and transitions between different tissue types that recapitulate standard features of histopathology of the oral mucosa (see Supplementary Movies 13–15 for 3D visualization). Bright fluorescence is visible on the tongue’s filiform papillae, probably from keratin and bacteria (Fig. 6c), while the fungiform papillae’s epithelium is transparent, permitting an unobstructed view of the bright green inner branched structures that match well with the structure of capillaries. Interestingly, one of the major sources of in vivo image contrast was found to be blood vessels, both from green autofluorescence of the vessel wall and red signal corresponding to blood itself (Fig. 6h–j, consistent with in vivo mouse pancreas data shown in Fig. 3d,f). In the lip, a diversity of different vascularized rete peg structures can be seen, progressing from fine and pointy in the inner lip to thicker and more stump-like at the transition from inner to outer lip. The transition from lip to skin captures striking features of hair follicles circled by microvasculature. The ability of MediSCAPE to image the regularity of patterns of these protrusions and the continuity of the basement membrane below the surface epithelium, as well as vascular patterns within the lamina propria, suggests that MediSCAPE could feasibly detect a range of disorders of the oral mucosa, from ulceration and scar tissue to squamous cell carcinoma. Importantly, large areas of the tissue of interest, here stitched to up to 13 mm for demonstration, can be canvassed for early detection of suspicious lesions and non-invasive follow-up and monitoring. The oral cavity was chosen for this first in vivo human demonstration as it is readily accessible in healthy volunteers. However, these data provide valuable evidence that MediSCAPE could be applied broadly to imaging in vivo in situ human tissues in a wide range of clinical settings including dentistry, otolaryngology, ophthalmology, gynaecology, and diverse open and laparoscopic surgeries and procedures.

Fig. 6 |. In vivo imaging of human oral cavity.

Upper panels: MediSCAPE imaging of dorsal surface of the in vivo human tongue. a, Stitched roving data as top-down MIP acquired on a benchtop MediSCAPE system (30 s at 5 VPS, 9.7 mm long). A and B indicate filiform and fungiform papillae, respectively, as shown in exemplar H&E histology of human tongue in b. c, Zoomed filiform papillae top-down MIP (i) and single-plane zoomed detail (ii) on a benchtop MediSCAPE compared to the miniaturized prototype MIP (iii). d, Fungiform papillae shown as top-down (i) and side (ii) MIP on a benchtop MediSCAPE, and a 3D rendering of a vascularized fungiform core from the mini system (iii). Supplementary Movies 12 and 13 show real-time 3D data and the stitched 3D rendering for benchtop and flythrough stitched data for the mini systems, respectively. Lower panels: MediSCAPE imaging of in vivo human inner and outer lip and skin. e, Stitched roving data as top-down MIP acquired continuously on a benchtop MediSCAPE system (74 s at 3 VPS, 13.1 mm long), from the inner (C) to outer (D) lip and transition to skin (E), as indicated in exemplar H&E of whole human lip in f. g, Comparable stitched roving scan from miniaturized MediSCAPE system from regions C to D (25 s at 4.8 VPS, 6 mm long). h, Papillary dermis in inner lip presented as side-view (i) and top-down (ii,iii) MIPs over different z ranges for both systems. Projections show lamina propria with pointy vascularized rete pegs protruding into the overlying transparent epithelial layer, matching structures in the H&E histology (iv). i, Outer lip papillary dermis presented as side-view (i) and top-down projections of the same region over indicated z ranges (ii–iv) for both systems. Denser and more vascularized rete pegs with stump-like shapes are seen in (v). j, Peri-oral skin presented as side view (i) and top-down projections over indicated z ranges (ii,iii). Capillary networks surrounding hair follicles (iv) can be visualized, demonstrating imaging through the epithelium and stratum basale, down to the dermal junction. Supplementary Movies 14 and 15 show 3D rendering of stitched roving scan acquired on the benchtop system and a flythrough of a comparable scan on the miniaturized system. All scale bars, 100 μm unless labelled otherwise.

High-speed 3D in vivo imaging enables functional characterization of in situ tissues.

Demonstrations primarily focused on the ability of MediSCAPE to image 3D cellular structures and patterns analogous to conventional ex vivo histopathology. However, the ability of MediSCAPE to perform real-time 3D in vivo imaging opens up the opportunity to assess aspects of tissue cellular function and physiology that are not accessible in ex vivo tissues.

One example is MediSCAPE’s ability to image 3D microvascular perfusion, already noted in label-free mouse and human data acquired in vivo above (Figs. 3d,f and 6). However, contrast can be enhanced by using intravenous injection of fluorescent dyes such as dextran-conjugated fluorescein, as demonstrated in Fig. 7a–c in the brain of a living, head-fixed mouse through a glass cranial window. Roving scans were acquired at 9 VPS and 3D-stitched into a larger volume shown as multiview maximum intensity projections. Supplementary Movie 16 plays the real-time roving data showing clear ability to observe dynamic flow in the vessels, while also capturing clear details of the 3D microvascular architecture without motion artefacts. In addition to neurological surgery applications, this approach could be of value in assessing microvasculature in tumour margins or after tumour embolization, arteriovenous malformations and tissue or organ reperfusion. MediSCAPE could leverage commonly used intravascular fluorophores, such as fluorescein and the near-infrared fluorophore indocyanine green, for deeper tissue penetration.

Fig. 7 |. In vivo functional imaging of mouse brain and kidney.

a, Exposed cortex was imaged through a glass cranial window in a head-fixed, anaesthetized wild-type mouse. b, IV-injected fluorescein isothiocyanate (FITC)-dextran in vasculature. Single volumes indicated in c, are shown as lateral (xy) and depth (yz) MIPs of skew-corrected volumes at timepoints 0 s (i) and 19 s (ii). Vessels can be easily distinguished along the entire depth range of 159 μm. Images are shown on a log scale for clearer visualization of smaller vessels. c, A large field of view was created by stitching contiguous volumes in a pairwise fashion while roving across the exposed cortex at 9 VPS. Supplementary Movie 16 shows a dynamic movie of roving data. d, An extracorporealized kidney in an anaesthetized wild-type mouse was imaged under two conditions: (1) a 10 min ischaemia period induced by clamping exposed renal vessels and (2) a tail vein injection of FNa. e, Plot of the mean intensity of emission channels over a 300 × 300 × 50 μm³ ROI at the kidney surface over a time lapse acquisition, with volumes acquired every 30 s. f, xy and yz slices from time lapse volumes also show that green autofluorescence is visibly reduced during ischaemia (i,ii) and even in local areas of ischaemia (white dashed outlines) after the vessel clamp was removed (iii–vi). A local ischaemic region in baseline, probably from pulling on the kidney during handling, is indicated with a white arrow. For ischaemia and reperfusion (rpf), time is given as seconds after the placement (ii) and removal (iii-vi) of the vessel clip. See Supplementary Movie 17 for a 3-axis view of volumes over ischaemia-reperfusion. g, The same approximate region as in f is shown as xy MIPs over the top 50 μm during a tail vein injection of FNa. i–iii are individual volumes from a 2 min movie acquired at 10 VPS during the injection bolus, as shown in Supplementary Movie 18. iv was acquired approximately 8.5 min after the injection and shows accumulation of FNa in proximal tubules. White asterisks mark the same tubule in f and g for easier comparison. All scale bars, 100 μm.

In another example, we show that ischaemia and reperfusion can be directly visualized by MediSCAPE through changes in autofluorescence alone, which may be especially valuable for transplant and resection procedures where long ischaemic times can lead to cell injury and death. An in vivo extracorporealized mouse kidney was imaged using a benchtop MediSCAPE system before, during and after ischaemia induced by clamping of the exposed renal vessels (Fig. 7d). High-resolution 3D volumes were acquired in 1.6 s every 30 s for 5 min before vessel clamping, for 10 min during clamping and for 10 min after clamp removal. Figure 7e,f shows that MediSCAPE allowed visualization of both global and localized ischaemia, and subsequent reperfusion through a decrease and increase, respectively, in green autofluorescence, probably from the redox–active coenzyme FAD. A clearly delineated border of progressive tissue recovery can be seen receding within the imaged volume (see Supplementary Movie 17 for full sequence).

Figure 7g shows the same kidney region, following ischaemia-reperfusion, during a bolus tail vein injection of fluorescein sodium (FNa). In addition to showing perfusion of the region, the low molecular weight FNa can be seen to gradually extravasate and selectively accumulate in proximal tubules (matching the tubules that exhibit brighter green autofluorescence prior to dye injection). Supplementary Movie 18 shows the full sequence of spatiotemporally dependent dye uptake.

Discussion

In this study we have shown that MediSCAPE allows real-time volumetric imaging of a wide range of intact in vivo and fresh tissues without the need for exogenous dyes. These capabilities could allow simple, yet comprehensive assessment of living tissues in a clinical setting. The main advantage of MediSCAPE over conventional confocal microendoscopy is its ultrafast 3D imaging speed, combined with much higher sensitivity. We also showed that these features permit high-quality in vivo imaging of cellular features and 3D morphologies using only autofluorescence contrast, while tolerating in vivo motion and allowing dynamic surveillance of large areas of tissue in real-time. We also demonstrated that MediSCAPE can capture measurements of in vivo function and physiology, including perfusion and effects of ischaemia, as well as extravasation and cellular uptake of a range of different exogenous fluorophores, extending its utility for broader clinical applications. Near-equivalent imaging performance was achieved with a small-format, relatively simple and low cost MediSCAPE prototype compatible with hand-held intrasurgical use and with strong FDA predicates.

Applications of MediSCAPE.

The primary clinical applications of MediSCAPE that we envisage are surgical guidance for lesion resection and biopsy site selection. The form factor of our miniaturized MediSCAPE prototype is currently compatible with open surgical fields including brain, heart, orthopaedic and abdominal surgeries, tissues within accessible orifices such as the mouth and cervix, and potentially for laparoscopic and robotic surgeries. Our results in a mouse model of pancreatic cancer suggest that MediSCAPE could provide valuable guidance during complex Whipple procedure surgery. A smaller form-factor system, or gradient index (GRIN) lens-based extension of MediSCAPE could permit ‘probe’ type imaging that could guide or be incorporated into needle biopsy procedures. In addition to disease mapping and biopsy guidance, the ability of MediSCAPE to image intact tissues non-destructively makes it suitable for applications in which biopsy might be contraindicated, such as for evaluation of tissue health, tissue typing, nerve localization, cartilage assessment or analysis of wound healing. The ability to acquire images in vivo can also be harnessed to reveal functional, physiological and metabolic changes in tissues as novel disease biomarkers^32–34, and to provide real-time assessments of ischaemia, tissue perfusion, and tissue reperfusion using intrinsic contrast or intravascular dyes for diverse clinical, veterinary and in vivo research applications. MediSCAPE could also prove useful in combination with wide-field imaging of targeted ‘molecular probes’ to visualize cellular-level uptake and disambiguate labelling, particularly during early clinical validation studies^6,7,35.

Although the penetration depth of MediSCAPE is limited by the scattering properties of the tissue being imaged, the high-speed 3D data produced are equivalent to 10s–100s of sequential thin histology sections. In many of our demonstrations, this 3D information provided valuable additional information about tissue structures, while also permitting roving and stitching, which would be impossible with 2D planar imaging. While this penetration depth limitation prevents non-invasive imaging of deep tissue structures, we note that the ability to image repeatedly during resection enables flexible interrogation of in situ and residual margins as overlying tissue is removed. Penetration depth could also be improved with light-sheet optimization, or use of red or near-infrared illumination, especially in concert with red-shifted contrast agents³⁶.

Our choice of human kidney tissues as a test sample was driven by a further potential application of MediSCAPE: the rapid, non-destructive inspection of donor organs prior to transplantation. Many donor kidneys are discarded because of difficulties in evaluating their health within the short window of time between donation and transplantation. The ability of MediSCAPE to visualize key diagnostic features in intact human kidney supports this potential application, which could extend to in situ evaluation and biopsy guidance in other transplant organs such as the liver and heart.

As demonstrated by our comprehensive imaging of fresh resected tissues with stage-scanned acquisition, MediSCAPE microscopy also has potential for rapid 3D evaluation of biopsies and resected tissues at the bedside. MediSCAPE far surpasses the 3D imaging speed limitations of point-scanning confocal, two-photon and Raman microscopy approaches^9–12,37,38 while avoiding the need for costly specialized lasers that can be challenging to locate at the bedside. Moreover, since imaging excised tissues removes constraints on utilizing a wide range of fresh-tissue-compatible selective dyes and labels³⁹, our MediSCAPE results in stained fresh tissues (Extended Data Figs. 5–7) show that bedside forms of MediSCAPE could provide a more comprehensive evaluation of biopsied tissue as a complement/cross validation to its in vivo use. Ex vivo tissues can also be chemically cleared to provide a more comprehensive 3D visualization^40–43. Although tissue clearing steps can take excessive time, cleared tissues can also be imaged using our benchtop MediSCAPE designs, offering advantages over dual-objective light-sheet systems, including simplicity of the single-objective light-sheet geometry and the ability to image to the full depth of the primary objective’s working distance¹⁷.

Sources of contrast.

The majority of images shown here were acquired with a single 488 nm laser for fluorescence excitation. However, we also note that a wider range of excitation wavelengths could readily be incorporated into MediSCAPE, including 405 nm, 561 nm and near-infrared ranges. Additional wavelengths could harness autofluorescent molecules such as NADH, collagen or retinol (see Supplementary Table 2), as well as exogenous dyes extending into the near infrared such as indocyanine green.

We note that although we compared autofluorescence imaging to conventional histological contrast as our gold standard, autofluorescence has the potential to reveal additional valuable information beyond what is seen in histology. For example, the autofluorescence detected with 488 nm excitation in the human kidney was particularly strong in the elastic lamina of arterial walls, cytoplasmic lipofuscin deposits and urinary cast material. Also clearly visible were cytoplasmic granular structures within epithelial cells of pseudohypertrophied proximal tubules, and focal arterioles with intense punctate perinuclear autofluorescence suggestive of lysosomal signals. Nearly all of these tissue features appear much less distinctive in routine histology, suggesting the potential for MediSCAPE to gather adjunctive information beyond what traditional histology can provide. New diagnostic features could have great clinical impact, particularly for limited or rare human tissue specimens such as small core needle biopsies.

Our functional in vivo imaging data (Fig. 7) also revealed dynamic changes in autofluorescence in response to ischaemia and reperfusion, as well as dynamic effects such as dye extravasation or uptake. The ability to interrogate in situ living tissues presents exciting opportunities to discover new diagnostic functional biomarkers for tissue identification and surgical guidance. Using MediSCAPE to capture dynamic in vivo tissue responses to dyes and perturbations to the tissue, including pressure, heat, cold, hypercapnia, hyperoxia or application of topical or systemic pharmacological agents, could provide further specificity⁴⁴.

Visualization, display and automated analysis.

A key factor in clinical adoption of MediSCAPE will be the way in which data can be visualized and interpreted in real-time by both the acquiring surgeon and the examining pathologist. All analysis and rendering of MediSCAPE images shown here were performed offline; however, real-time stitching and visualization of depth and lateral cross-sections should be feasible using FPGA technologies that work well for real-time visualization and rendering of ultrasound and optical coherence tomography (OCT) data^45,46. Moreover, the digital nature of MediSCAPE’s data would permit online cloud-based inspection of datasets by remote pathologists (as is common in radiology) who could readily select their preferred views and colour schemes⁴⁷. MediSCAPE’s rich volumetric data are also ideally suited for automated machine learning-based analysis that could automatically classify normal and suspicious areas and pick out key tissue features. Online analysis results could be projected onto the surgical field, or visualized using augmented reality. Where available, MediSCAPE data could be spatially registered to stereotactic coordinates and other imaging modalities such as MRI, and fully archived as part of the patient’s electronic health records.

Technological development and form factor.

Near-equivalent performance was demonstrated with both benchtop and miniaturized MediSCAPE designs. The form factor of our more compact system is compatible with being mounted on a surgical microscope frame and hand-guided within the surgical field. Further miniaturization using custom-built small-diameter high-NA objectives, micro-electro-mechanical systems (MEMs) mirrors, fibre optics, rod and GRIN lenses could all further reduce the form factor of the system to permit laparoscopic, probe-based and even endoscopic use (see Supplementary Note 1 for further discussion)^33,48,49.

Our optically transparent spacer at the tip of the primary objective enabled unrestrained in vivo tissues to be readily stabilized and roved at the optimal working distance without precision positioning. This tip can double as a disposable or sterilizable sheath for patient protection, facilitating intrasurgical use. Features such as microscale stabilization and auto-scanning over fixed distances could further improve ease of use, while the ability to mark, capture or even laser-ablate identified regions in concert with imaging could provide substantial benefits for microscale resection. As demonstrated, the ability of MediSCAPE to dynamically zoom into features of interest would also be beneficial, providing a compromise between covering larger areas via roving and capturing key features of disease in the tissue of interest.

Overall, MediSCAPE represents an approach to in situ histopathology that leverages the benefits of light-sheet scanning to allow for high-speed 3D, label-free imaging of a wide range of tissues. Our demonstrations show that MediSCAPE provides a non-destructive alternative to biopsies and conventional histopathology, for the in vivo assessment of a wide range of functional tissue features in situ. These capabilities could help improve the standard of care while also reducing the time and cost of a wide range of surgical procedures.

Methods

In vivo mouse tissue preparation and imaging.

In vivo mouse imaging was carried out according to protocols reviewed and approved by Columbia University’s Institutional Animal Care and Use Committee (IACUC). For all in vivo imaging of internal organs, each mouse was heavily anaesthetized using isoflurane at a concentration of 3% for induction and 1.5% for maintenance, and its snout was placed in a mouse mask. Body temperature was maintained with a warming pad positioned on top of the mouse and breathing was monitored continuously. Abdominal organs were exposed and the mouse was placed on a 60-mm-diameter glass bottom dish mounted on a 3-axis stage. Organs (for example, kidney in Fig. 1h–k and Supplementary Movie 6) were positioned to be against the surface of the glass coverslip for imaging from below. For roving imaging, the position of the mouse was manually translated during continuous imaging. Warm saline was used to periodically flush tissues to minimize drying and maintain body temperature.

Kidney, heart and brain imaging studies were all performed in wild-type mice. For in vivo heart imaging (Fig. 3h and Supplementary Movie 8), the chest cavity was opened and the heart was rapidly positioned for imaging prior to euthanasia. For in vivo pancreatic imaging (Figs. 2 and 3c, and Supplementary Movies 4 and 7), data were acquired in a KPC mouse model of pancreatic ductal adenocarcinoma (LSL-KrasG12D; LSL-p53R172H; Pdx-Cre). A ~6-mm-tumour-bearing mouse was identified by ultrasound. The abdomen was shaved and a small incision was made over the left side of the abdomen to expose the tumour area in the tail of the pancreas. The mouse was then positioned with this opened area against the surface of the glass bottomed dish for imaging from below. After in vivo imaging, the pancreas was excised and cut to better expose the tumour area and imaged immediately (Fig. 2j and Supplementary Movie 4).

For imaging brain microvasculature shown in Fig. 7a–c and Supplementary Movie 16, the mouse was anaesthetized using urethane at a dose of 1.5 mg kg⁻¹, and sealed bilateral glass cranial windows were implanted over the somatosensory cortices as previously described⁵⁰. A metal head plate was glued to the skull to enable head fixation under the MediSCAPE objective lens in an upright configuration. Imaging was performed following tail vein injection of ~0.1 ml 5% w/v 70,000 MW fluorescein isothiocyanate–dextran. The xyz position of the mouse was translated manually with 3-axis stages during roving imaging.

For kidney ischaemia-perfusion and dye uptake experiments (Fig. 7d–g, and Supplementary Movies 17 and 18), the wild-type mouse was anaesthetized with isoflurane, the dorsal abdomen was shaved and a small incision was made over the left kidney to extracorporealize the kidney. A plastic platform with a millimetre-sized groove was used to hold the kidney above the body and expose the renal vessels. The kidney was kept moist with warm saline and imaged from above with water immersion. Ischaemia was induced by clamping the exposed renal vessels with a vessel clip. After clip removal, the kidney was allowed to reperfuse for 20 min. For imaging of dye dynamics, a tail vein injection of ~0.05 ml 2.5% w/v fluorescein sodium (AATBio, CAS 518–47-8) was administered during real-time imaging.

Fresh mouse tissue preparation and imaging.

Fresh mouse tissue was excised from mice according to protocols reviewed and approved by Columbia University’s IACUC. Wild-type mice were used in all cases, except for the pancreatic cancer model detailed above. Mice were heavily anaesthetized and then euthanized using cervical dislocation. Excised tissue was kept on ice until imaging or until at least 30 min before staining. All data were acquired within 3 h of excision. Tissues were imaged from below in 30-mm-diameter glass bottom dishes, with the benchtop MediSCAPE objective in an inverted setup. Tissue was kept moist with saline and gently pressed down with a coverslip to create a flatter imaging surface when necessary. On the miniaturized MediSCAPE system, tissues were placed in a petri dish and imaged from above (in an upright configuration). Tissues were kept moist with saline and pressed down with a coverslip when needed. For datasets showing stained fresh tissue, tissues were topically stained at room temperature for 1–3 min, rinsed with saline and imaged immediately, as described below.

For mouse imaging of normal and PDAC pancreatic tissue in Figs. 2a–h and 3d–g and Supplementary Movies 4 and 7, one wild-type JRGECO(−) 58-week-old male and one KrasLSL.G12D/+; Trp53LSL.R172H/+; Pdxl-Cretg/+ (KPC) 6-month-old female were used, respectively. In the kidney ischaemia-reperfusion experiment shown in Fig. 7f,g and Supplementary Movies 16 and 17, two wild-type mice were imaged, including a JRGECO(−) 58-week-old male and cdH5(−) 40-week-old male. For imaging normal tissue morphology, a total of nine adult wild-type mice were imaged.

Human kidney biopsy collection and imaging.

De-identified fresh human kidney tissue was acquired through the Tissue Bank at the Columbia University Medical Center Department of Pathology under IRB-reviewed protocol #AAAS3959 (designated not human subjects research). Tissue was imaged within 24 h of excision, stored in a petri dish with saline-soaked cloth at 4 °C and kept on ice before imaging. Tissue was imaged from below on benchtop MediSCAPE in a 30-mm-diameter glass bottom dish with saline to keep moist.

Where indicated, human tissues were topically stained with 0.01% proflavine (Sigma, 131105) in saline, 1% methylene blue (Ricca, 485016) and/or 0.01% fluorescein sodium (AATBio, CAS 518–47-8) in water. Dyes were gently applied with a cotton swab to tissue at room temperature for 1–3 min and then rinsed away 3 times with saline. Stained regions were imaged immediately.

In vivo human imaging.

Human participants procedures were reviewed and approved by Columbia’s Institutional review board (IRB #AAAT6431) and ethics committees. Written informed consent was obtained from the participants according to CARE guidelines and in compliance with the Declaration of Helsinki principles. For data shown in Fig. 6 and Supplementary Movies 12–14, an adult female subject was asked to rinse their mouth with water and then position MediSCAPE’s sterile objective lens cap onto various oral tissues including the tongue top and side and inner lip. Where possible, the subject was asked to slowly move relative to the primary objective lens to produce a roving scan.

Histology.

After imaging, all fresh tissues were marked with tissue marker to clearly indicate the face imaged on MediSCAPE, and placed into histology cassettes with the imaged face laid flat on biopsy paper. Tissues were fixed in 10% formalin for at least 24 h at 4 °C. Subsequent histological embedding, slicing, staining and mounting were done by Molecular Pathology Histology Services at the Columbia University Medical Center Herbert Irving Cancer Center. All mouse tissues were level cut into several 5 μm flat face sections spanning the first 50–100 μm of the imaged face and stained with H&E. Kidney biopsy tissue was cut into 2 μm flat face sections spanning the first 50–100 μm of the imaged face and stained with both H&E and PAS. Histology slides were digitally scanned using a Nikon AZ100 slide scanner. Regions of interest were matched by manually comparing structural features visible in MediSCAPE images with digital histology data.

Data processing.

Both benchtop and miniaturized MediSCAPE datasets were collected using a custom MATLAB-based acquisition software that interfaced with camera software (Andor Solis version 4.30), and controlled the galvanometric scanning mirror and Thorlabs motorized stages. MediSCAPE data processing consisted of background subtraction, skew-correction of data and merging dual colour images with a custom-written MATLAB graphical user interface. The camera background was calculated by taking the average of approximately 100 frames acquired without laser light at the end of each run, providing a measure of the differing baseline offset of each pixel in the sCMOS sensor. In runs where these frames were not taken, background was approximated as the lowest 5% of pixels along the x dimension of the full run. Skew-correction by the system sheet angle was applied with a simple affine transform as described previously¹⁷. Dual colour merging was performed using xyz transform coordinates to overlap the colour channels, which are acquired side-by-side on the camera chip. These coordinates are determined manually during initial setup of the image splitter. A slight magnification and rotation of the green channel were also applied where needed to properly overlap channels. A pseudoflatfield correction was applied to dual colour images along the x and y axes by dividing volumes by a Gaussian-blurred mean intensity z (and t) projection.

For better visualization of details, MediSCAPE data shown in Fig. 2 were processed with an unsharp mask (radius 1, amount 0.3) and contrast-limited adaptive histogram equalization (CLAHE) (block size 75, slope 2) using lmageJ (version 1.53c). Fig. 6c(ii) bench, 6c(i) mini and j(iii) bench were processed with unsharp masking (radius 3, amount 0.5). CLAHE was used for Fig. 6e (blocksize 512, slope 1.5) as well as h(i,iii) bench and i(i) bench (blocksize 128 slope 2). Data were visualized using either lmageJ (version 1.53c) or lmaris (version 9.2.1).

For H&E pseudocolouring of MediSCAPE datasets acquired with proflavine staining (Extended Data Fig. 5 and Extended Data Fig. 6), the virtual H&E algorithm developed by Giacomelli et al.³¹ was adapted into a MATLAB script and used to create brightfield H&E colour channels from fluorescence data on the basis of Beer-Lambert’s law. Autofluorescence emission collected through a 618/50 nm bandpass filter with 488 nm excitation was used to indicate general non-nuclear background structure on a log scale (eosin), while proflavine fluorescence excited at 488 nm was used to indicate nuclear structure (haematoxylin).

For in vivo ischaemia-reperfusion and dye uptake data shown in Fig. 7 and Supplementary Movies 17 and 18, volumes were registered over time using the (2D/3D) descriptor-based registration plugin in Fiji⁴¹ and cropped to show the same region over time. Volumes showing major up–down motion from heartbeats or respiration were automatically discarded in MATLAB before registration. The time lapses shown in Fig. 7f and Supplementary Movie 17 were acquired as two consecutive volumes every 30 s and only one volume was used for analysis. For better visualization of the colour channels in Fig. 7g, the red channel was scaled and subtracted from the green channel to better separate green autofluorescence from the fluorescein sodium.

Data stitching (roving).

To stitch consecutively acquired overlapping volumes from a roving scan, a custom ImageJ macro was written using the existing Pairwise Stitching plugin in Fiji^51,52 to act on volumes already saved as background-subtracted, dual colour tiff stacks in MATLAB. Volumes were stitched pairwise to simulate real-time stitching, which could be implemented using an FPGA. Because the volume rate was generally much higher than the speed at which tissue was translated during acquisition, every nth volume (where n = 1–5) acquired was used for stitching to decrease total processing time and reduce stitching errors. Stitched volumes shown in Figs. 3, 6 and 7, and Supplementary Movies 6–8 and 13–16 were created by fusing approximately every 1st to 4th consecutively acquired volume in a pairwise fashion. During each stitching step, volumes were downsampled 2× along the y and z axes and aligned coarsely given the alignment positions found in the previous successful stitching step. If the alignment r-value was over a given threshold (~0.8), fine alignment of the raw volumes was performed using the initial coarse alignment values and raw volumes were fused using linear blending with a 10% overlap. If the coarse alignment r-value was under the threshold due to excessive motion, the next consecutive volume was loaded and aligned and so on, until both volumes could be aligned accurately. In datasets acquired on the miniaturized MediSCAPE system, the downsampling/coarse alignment step was skipped. Data were skew-corrected in MATLAB after stitching. To create the ‘real-time stitching’ movies shown in Supplementary Movies 6 and 8, each stitching step was skew-corrected and positioned on a blank canvas of the same 3D size as the final fully stitched volume.

Most datasets were acquired and processed on a dedicated workstation with 192 Gb RAM and a 2 TB SSD drive. A typical 30 s long roving scan acquired at 1,000 fps and 10 VPS is about 7.5 Gb and required ~40 s Gb⁻¹ for processing in MATLAB, which consists of background subtraction, colour merging, flatfield correction, deskewing and writing to tiff. Pairwise stitching of a typical 30 s dataset ranged from 1 to 2 h depending on tissue translation speed and signal to noise (SNR). However, these steps can be parallelized and performed as data are acquired using FPGA and/or graphics processing unit (GPU)-based computing to massively speed up processing times. Furthermore, cloud-based analysis and remote display, and automated analysis of datasets can ease local computing power requirements.

Data stitching (stage-scanning).

The Bigstitcher plugin in Fiji^52,53 was used to stitch stage-scanned data (Fig. 4 and Supplementary Movie 9). A custom MATLAB and ImageJ pipeline was implemented to automatically save background-subtracted, skew-corrected dual colour tiff stacks in MATLAB, convert and load data in hierarchical data format 5 (HDF5) format into BigStitcher, pre-align volumes with stage coordinates and stitch data using linear blending with default stitching wizard presets and fine iterative closest point (ICP) alignment. The very large stitched dataset shown in Fig. 4 and Supplementary Movie 9 had a final size of 60 Gb and required around 3 h to process and stitch. However, processing and stitching steps can be parallelized, optimized and performed as data are acquired to reduce processing times.

Statistics and reproducibility.

All images and movies depict single imaging datasets collected on single samples to show proof of concept; however, all datasets shown are representative of multiple repeated measurements that showed similar features. In vivo and fresh normal mouse tissue data shown in Figs. 1, 2a–h, 3h–j and 7b,c, Extended Data Figs. 1, 2 and 7, Supplementary Note 2 Fig. 1 and Movies 1–3, 5, 6, 8 and 15 were collected from 9 different wild-type mice. In addition, the kidney ischaemia-reperfusion experiment shown in Fig. 7f,g and Supplementary Movies 16 and 17 was repeated on 2 wild-type mice with similar results in both animals. Pancreatic imaging shown in Figs. 2a–h and 3d–g and Supplementary Movies 4 and 7 was performed in 1 wild-type and 1 KrasLSL.G12D/+; Trp53LSL. R172H/+; Pdxl-Cretg/+ (KPC) mouse. Human kidney tissue data shown in Figs. 4 and 5, Extended Data Figs. 3–6 and Supplementary Movies 9–11 were collected from 6 individual biopsy pieces from 4 different patients. Human in vivo data shown in Fig. 6 and Supplementary Movies 12–14 were collected from 1 individual over 5 different imaging sessions on 2 different systems. Supplementary Note Fig. 1b shows statistical analysis of 6,300+ beads performed for determining the microscope resolution at 29 z depths by averaging and finding the standard deviation of the full width half maximum (FWHM) of beads in 3 dimensions.

Extended Data

Extended Data Fig. 1 |. Dual colour autofluorescence visualization.

xy image planes acquired by MediSCAPE in fresh mouse brain cortex, including a prominent blood vessel. Contrast corresponds to autofluorescence excited by 488 nm light. Dual colour emission images are acquired simultaneously using an image splitter in front of the camera and positioning each colour channels side-by-side on the camera chip (along y). a and b show grayscale raw emission channels acquired with 525/50 nm and 618/50 nm bandpass filters respectively. These channels are converted to c, ‘yellow hot’ and d, blue colormaps and then merged, as shown in e. This colour-scheme was chosen to enable the highly overlapping colour channels to be more easily distinguished when merged than conventional RBG scaling, while also being more colour-blind accessible. Scale bars, 100 μm.

Extended Data Fig. 2 |. Label-free imaging of fresh mouse tissue with miniaturized MediSCAPE.

xy (top) and yz (bottom) cross-sections in various fresh mouse tissue acquired with the miniaturized MediSCAPE system with 488 nm excitation and dual colour emission channels labelled as shown. Cross-sections show a, tubules in the kidney cortex, b, capsule and underlying cords of hepatocytes in the liver, c, cardiac muscle in the heart and d, crypts of Lieberkühn in the colon mucosa. White dotted lines in the xy planes indicate the location of corresponding yz planes shown. e. 3D rendering (Imaris) of a stitched field of view created from 16 seconds of continuous volume acquisition at 11.2 VPS while roving across the colon mucosa. Insets to the right show single planes from the ROI indicated (yellow dotted box). Yellow dotted lines in the xy cross-section show the locations of corresponding depth cross-sections. Only emission at ~525 nm is shown here, in grayscale. All scale bars, 100 μm.

Extended Data Fig. 3 |. Autofluorescence in diabetic human kidney tissue imaged with MediSCAPE.

Autofluorescence captured by MediSCAPE reveals features common to and beyond those seen on routine histology. a, PAS histology image of kidney cortex tissue from an older, diabetic patient with features of mild diabetic nephropathy. Scale bar, 200 μm. b, a MediSCAPE xy slice from a stage-scanned volume of the same piece of tissue (while fresh) showing autofluorescence excited at 488 nm. Yellow arrows point to the kidney capsule, white arrows indicate urinary casts, seen in both SCAPE and PAS images. Dotted line squares indicate regions of interest shown in more detail in c, highlighting features more evident by SCAPE than PAS histology. Scale bar, 200 μm. c(i), A focal subcapsular collection of tubules with autofluorescent cytoplasmic granules (teal arrow). Urinary cast material (white arrows) is also evident in the xy plane, and further evidenced by characteristically strong autofluorescence in the yz plane. c(ii). Tubules with accentuated peritubular autofluorescence (red arrow). c(iii) Glomerulus with focal autofluorescent granules. Scale bars, 50 μm.

Extended Data Fig. 4 |. MediSCAPE label-free imaging of elastic fibres and fat cells in human perirenal fat.

a, 3D rendering (ImageJ 3DViewer) of a section of normal human perirenal fat showing highly fluorescent elastic fibres and fat cells. b, a yz cross-section from the plane indicated shows layering of fibres over fat cells, which can be distinguished as circular yellow droplets. c, lateral cross-sections with corresponding tissue depth indicated on the bottom left corner. Fat cells are indicated with red arrows and intersecting vessels also labelled. All scale bars, 100 μm.

Extended Data Fig. 5 |. MediSCAPE images of proflavine-stained fresh human kidney biopsy including pseudocolour H&E.

a, xy slice in a stage-scanned field of view showing proflavine, a nuclear dye (green), and red autofluorescence emission (blue), both acquired simultaneously with 488 nm excitation. Autofluorescence is shown on a log scale for better visualization. b, Pseudocolored xy slice from panel a, approximating H&E histology with proflavine shown in purple and autofluorescence shown in pink. c, Inset showing a glomerulus (white arrow) and tubules in greater detail with true H&E histology of an adjacent region shown in d for comparison. See Supplementary Movie 10 for a depth fly-through. All scale bars, 100 μm.

Extended Data Fig. 6 |. Stained human kidney tissues imaged with MediSCAPE.

Fresh human kidney tissues showing features of arterionephrosclerosis were stained with nuclear dye, either b, 1% methylene blue (exc. 637 nm, em. >685 nm) or d-k, 0.01% proflavine (exc. 488 nm, em. 525/50 nm). maged by MediSCAPE, then processed for histology where the same tissue block faces were stained with PAS and/or H&E. a-c, demonstrate how the 4 main renal histologic components which must be routinely evaluated with both PAS and H&E histology appear in a, PAS histology, b, an xy slice of a MediSCAPE volume stained with methylene blue and c, H&E histology. These components are indicated as following: glomeruli (white arrows), arteries (red arrows), tubules (black arrows), and interstitium (blue arrows). Methylene blue in the MediSCAPE image defines cellular cytoplasmic, nuclear and extracellular compartments, similar to H&E but better highlights arterial elastic lamina and tubular and interstitial compartments, similar to PAS histologic sections. A second biopsy piece from the same patient shows scarred tubulointerstitium in a focal area of fibrosis (green arrows) in both d, MediSCAPE and e, corresponding H&E histology. f, a 3D rendering (Imaris) of the larger stage-scanned volume acquired on MediSCAPE, shows the 3D structure of fibrosis (green arrow), arteries and glomeruli (white arrows). The lengthwise white dotted line indicates the origin of the xz depth section shown in g. h, A non-sclerotic glomerulus is shown in more detail across 20um in depth. See Supplementary Movie 11 for a moving 3D render and depth fly-throughs of lateral and depth cross-sections of the full 3D volume. A second region of the same tissue shows i, globally- sclerotic glomeruli (yellow arrows) and urinary casts (light blue arrow) in a xy slice and depth slice taken at the white dotted line. j, H&E histology of a deeper section of the same region and k, MediSCAPE xy cross-section of the entire stage-scanned volume also shows globally-sclerotic glomeruli (yellow arrows) and normal glomeruli (white arrows). All scale bars, 100 μm.

Extended Data Fig. 7 |. Comparison of topical dyes applied to fresh mouse colon mucosa.

Single xy (top) and yz (bottom) slices in samples of fresh mouse colon mucosa imaged with MediSCAPE. Contrast is derived from a, 0.01% proflavine, a nuclear dye (exc. 488 nm, em. 525/50 nm), b, 1% methylene blue, a clinically-used nuclear dye (exc. 637 nm, em. >685 nm), and c, fluorescein sodium, an FDA-approved topical and IV dye (exc. 488 nm, em. 525/50 nm). Yellow dotted lines indicate locations of corresponding cross-sections and yellow arrows indicate nuclei in insets. Depth penetration of topically applied dyes is both stain- and tissue-dependent, as shown in yz depth sections. Also shown for comparison is d, autofluorescence (exc. 488 nm, em. 525/50 nm) imaged with MediSCAPE with e, corresponding en face and depth-sectioned H&E histology of a similar region in the same colon tissue. White arrows, crypts of Lieberkühn. Red arrows, goblet cells. Scale bars 100 μm.

Data availability

The main data supporting the results in this study are available within the paper and its Supplementary Information. The raw and analysed datasets generated during the study are too large to be publicly shared, yet they are available for research purposes from the corresponding authors on reasonable request.