For the 1.7 million patients per year in the US who receive a new cancer diagnosis, treatment decisions are largely based on histopathological specimen examinations. Unfortunately, the gold standard of slide-based microscopic pathology suffers from high inter-observer variability and limited prognostic value due to sampling limitations and the inability to visualize tissue structures and molecular targets in their native 3D context. Here, we show that an open-top light-sheet microscope optimized for non-destructive slide-free pathology of clinical specimens enables the rapid imaging of intact tissues at high resolution over large 2D and 3D fields of view, with the same level of detail as traditional pathology. We demonstrate the utility of this technology for various applications: wide-area surface microscopy to triage surgical specimens (with ~200 μm surface irregularities), rapid intraoperative assessment of tumour-margin surfaces (12.5 s cm−2), and volumetric assessment of optically cleared core-needle biopsies (1 mm in diameter, 2 cm in length). Light-sheet microscopy can be a versatile tool for both rapid surface microscopy and deep volumetric microscopy of human specimens.
Pathologic features provide a gold-standard by which diseases are diagnosed, patient prognoses are determined, and treatment decisions are made1. In a pathology lab, the examination of a tissue biopsy or a surgically excised specimen is most reliably performed by the microscopic examination of tissues that have been formalin-fixed and paraffin-embedded (FFPE), and then thinly sectioned, stained and mounted on glass slides (Supplementary Fig. 1a). This is a labour- and time-intensive process, in which only a small fraction of a specimen is sampled for microscopy. Frozen sectioning enables tissues to be rapidly sectioned, stained and mounted on slides. However, this technique can introduce severe artefacts , especially for tissues that do not freeze well (such as breast tissue). Furthermore, similarly to FFPE histology, it is destructive of tissue and suffers from the same sampling limitations2.
Considering the foundational role that pathology plays in patient care, it is concerning to note the large degree of inter- and intra-observer variability among pathologists in diagnosing and grading malignancies 3,4,5,6. A number of factors limit the reproducibility (precision) and predictive power (accuracy) of conventional pathology, including: (i) sampling errors due to the sparse amounts of tissue that may realistically be prepared on glass slides and visualized7, (ii) the inability to visualize structures of diagnostic importance within their volumetric context8, (iii) the destructive nature of routine histology, in which tissues are heavily processed, sectioned and mounted on slides, thereby compromising their value for downstream molecular and genetic analyses, and (iv) the high overhead of labour, equipment and consumables (glass slides, reagents, paraffin, cassettes, and so on), all of which increase healthcare costs9.
The ability to image intact tissues at high resolution over large areas or volumetric fields of view, with the same level of morphological and molecular contrast that is possible through conventional pathology (that is, haematoxylin and eosin (H&E) staining and immunostaining), promises to advance the field of pathology and thereby improve patient treatments and outcomes.
Fluorescence microscopy of intact human tissues
While 3D digital pathology using serial sections (on glass slides) is possible, it is not a viable strategy for routine clinical use due to the labour and time involved in preparing and imaging hundreds of glass slides, and aligning the digital images. Non-destructive fluorescence microscopy of human tissues has been attempted using a variety of approaches including confocal microscopy10,11, microscopy with UV surface excitation (MUSE)12, wide-field structured-illumination microscopy (SIM)13,14, and nonlinear (for example, multiphoton and stimulated Raman scattering) microscopy 15,16,17 . Compared with alternative optical-sectioning approaches, confocal and nonlinear microscopy systems provide superior resolution and contrast. However, these methods are complex and slow, often requiring raster scanning of a tightly focused laser beam in three dimensions to obtain a volumetric data set (for example, a recent study required 30 hours to image a single biopsy specimen18). Alternatively, camera-based optical-sectioning microscopy techniques such as MUSE and SIM are advantageous for clinical applications due to their faster imaging speeds and reduced complexity, but are not intended for deep volumetric microscopy. For example, MUSE relies on the limited penetration depth of UV light to obtain a selective image of the tissue surface, thereby precluding subsurface imaging. SIM relies on tissue illumination with spatial patterns that are imaged with high contrast only at the focal plane of a microscope, thereby enabling digital rejection of out-of-focus background light (primarily lower spatial frequencies) through a demodulation algorithm. Imaging depth is therefore limited by the shot noise of the background light that impinges on the detector array, as well as the finite dynamic range of the detector19,20. Finally, reflectance-based imaging techniques such as optical coherence tomography/microscopy (OCT/OCM) have been successfully translated into clinical settings for label-free imaging of human tissues21,22, with a demonstrable impact on patient outcomes23. However, these methods are not typically designed to achieve subcellular resolution and are not suited for probing multiplexed and molecular targets as desired for standard-of-care clinical pathology.
Light-sheet microscopy, whose technological roots may be traced back over a century24, has recently seen intense development for biological investigations in which rapid volumetric microscopy is desired 25,26,27,28 . The light-sheet microscopy approach achieves optical sectioning (rejection of out-of-focus light) by using a thin ‘selective’ illumination plane, which generates fluorescence signal that is imaged in the orthogonal direction. The flexibility of this dual-axis configuration, where the illumination and collection beam paths are decoupled and may be individually optimized, contrasts with conventional single-axis microscopes in which the illumination and collection beams travel along an identical path, and where the depth of focus and resolution are coupled by the laws of diffraction (for standard Gaussian beams). For example, by using low-numerical-aperture (NA) Gaussian illumination, the depth of focus of a light-sheet microscope can be extended to hundreds of micrometres — but at the expense of a moderately thick illumination beam waist (such as 7 μm in our current system). Despite this low NA, in which the light sheet is moderately thick, it is still possible to achieve high spatial resolution (that is sufficient to visualize nuclei) by using a higher NA for the orthogonal collection beam path29. This is similar to conventional slide-based histology, where 5-μm-thick physically sectioned tissues yield high-quality images when viewed with moderate- to high-NA microscope objectives. Finally, light-sheet microscopy systems can be simple and relatively inexpensive to build (as shown, for example, by the OpenSPIM project)30.
In summary, camera-based tissue-microscopy approaches (such as SIM and MUSE) are fast, but generally provide poor imaging depth (background suppression), which limits their effectiveness for volumetric imaging applications. On the other hand, 3D laser-scanning microscopy methods can image deeply but are generally complex and slow. Here, we show that light-sheet microscopy is a versatile technology that provides an elegant and simple means of rapidly imaging the irregular surfaces of surgically excised tissues, as well as enabling deep volumetric microscopy of clinical specimens. A qualitative summary of various trades-offs for selected thick-tissue microscopy approaches is presented in Supplementary Table 1 for both surface-microscopy and volumetric-microscopy applications.
Open-top light-sheet microscope for clinical pathology
In early light-sheet microscopes, high-resolution volumetric imaging was achieved by slowly rotating or translating a sample embedded in an agarose gel31. These configurations were developed for imaging model organisms and tissues for biological investigations (particularly in developmental biology and neuroscience) and were restricted to relatively small transparent objects. Several variations of the original light-sheet microscopy architecture have been introduced, each of which has facilitated a growing number of biological imaging applications 32,33,34,35,36 . However, a versatile light-sheet microscope designed to image large centimetre-sized human tissue specimens, for diverse applications in clinical pathology, has not been reported.
Our light-sheet microscope design uses an open-top architecture37 that enables large human tissues to be conveniently and rapidly placed on top of a flat glass-plate sample holder and imaged from underneath (Fig. 1 and Supplementary Video 1). All optical components in our system are positioned below the glass-plate sample holder, which allows the microscope to accommodate human tissues of arbitrary size and thickness without physically interfering with the illumination and collection optics. However, this geometry also presents significant optical challenges, as the highly off-axis (45°) illumination and collection beams are not easily index-matched into the sample, which can result in severe aberrations (such as coma, astigmatism and spherical aberrations).
To enable aberration-free imaging, our open-top light-sheet microscope uses a hemispherical solid-immersion lens (SIL) that has been used in the past in single-axis microscopes and dual-axis confocal microscopes38,39, but has not been demonstrated for a light-sheet microscope in which a low-NA cylindrically focused illumination beam, and a higher-NA spherically focused collection beam, must be coupled into a tissue specimen at a large angle of incidence (Fig. 1b). The bottom of the SIL provides a curved surface that approximates the wave-front curvature of the illumination and collection beams as they transition from air into glass (or vice versa), which minimizes the introduction of aberrations at those interfaces (Supplementary Figs 2 and 3). A thin oil layer at the top surface of the SIL, which does not evaporate or require refilling, provides an index-matched interface between the stationary SIL and a translating glass-plate sample holder (n = 1.464 ± 0.001). This enables panoramic scanning over theoretically unlimited surface areas with diffraction-limited performance (Supplementary Figs 3–6). In short, tissues are stage-scanned to allow camera-based imaging of serial oblique light sheets within the tissue sample (Fig. 1c and Supplementary Fig 3). By translating the sample in one dimension, a 3D trapezoidal volume may be rapidly imaged. Adjacent 3D image strips are acquired in a serpentine pattern, enabling panoramic 3D microscopy over large areas (Supplementary Video 1). The use of a dual-channel image splitter allows simultaneous capture of two independent fluorescence channels for samples stained with multiple dyes.
For the prototype developed in this study, the width (w) of each image is 2.0 mm (Supplementary Fig 3), which is dictated by the field of view of the collection objective. The image height (h, that is, the vertical field of view) is adjustable by varying the vertical crop size of the camera, which directly affects the imaging speed (v) (Fig. 1d). As shown in the results (Figs 2-4), this versatility to trade off between vertical field of view and imaging speed (Fig. 1d) allows our open-top light-sheet microscope to be optimized for specific clinical applications and tissue types.
A variety of clinical pathology applications were explored to demonstrate the clinical utility of our open-top light sheet microscope. First, we performed post-operative surface microscopy (for triage inspection) of fresh radical prostatectomy specimens with highly irregular surfaces. Second, we performed rapid intraoperative surface microscopy of fresh breast tissues for guiding surgical oncology. Finally, dual-channel volumetric imaging was achieved of optically cleared core-needle biopsy specimens, demonstrating the potential to improve on the accuracy of traditional slide-based histopathology for the diagnosis and grading of atypical lesions. Collectively, these examples demonstrate that our open-top light-sheet microscope is optimal and versatile for both rapid microscopy of large irregular tissue surfaces and comprehensive volumetric microscopy. A simplified comparison of a conventional histology protocol and our open-top light-sheet microscopy protocol is provided in Supplementary Fig. 1.
Wide-area surface microscopy of human prostate tissue
For the standard-of-care histological evaluation of a prostatectomy specimen, the organ is first manually cut (‘bread-loafed’) into thick slices (3 to 5 mm in thickness) (Fig. 2a) from which only one histology slide is typically generated per slice. As discussed later, certain institutions will triage up to 50% of the gross slices to conserve resources. In this study, 25 large human prostate slices, approximately 3.1 × 3.5 × 0.4 cm in size (Fig. 2b), were stained with acridine orange for 20 s, rinsed for 10 s, and then imaged using the open-top light-sheet microscope (h = 320 μm, v = 50 s cm−2) within an average imaging time of <10 min per slice (Fig. 2c). A digital surface-extraction algorithm was used to display a comprehensive image of the entire tissue surface from the raw 3D data set. This process accommodates both the irregularities of the tissue surface (~200 μm in extent for manually sliced prostate tissues), and the minor tilt of the sample plane across the entire tissue surface (typically up to a 0.25% slope) (Fig. 2d). An average histogram of the tissue surface depth from all 25 samples demonstrates the typical surface irregularity seen in most unfixed prostate tissues (Fig. 2e). Tissue surface irregularities were found to be less than 200 μm, which is within the depth of focus of the open-top light-sheet microscope (zR > 200 μm; Supplementary Fig. 4) as well as the vertical field of view (h = 320 μm; Fig. 1) used in this case.
High-magnification views of relevant prostate pathologies are shown in Fig. 2f–h. Benign ducts are observed with clear lumens lined by a bilayer of nuclei (Fig. 2e) and fibromuscular stroma. Carcinoma is characterized by increased cellularity and crowded glands (Fig. 2f) surrounded by fibromuscular stroma. Finally, glands with high-grade prostatic intraepithelial neoplasia are characterized by stratified nuclei and tufted projections into the lumen (Fig. 2g,h). Surface-extracted open-top light-sheet microscopy images of all 25 prostate samples are shown in Supplementary Fig. 7 in comparison to slide-based histology. Supplementary Video 2 shows an entire prostate surface, and demonstrates the ability to zoom in and out.
Rapid intraoperative microscopy of human breast tissue
Breast tissue is primarily composed of pliable adipose tissue, which is difficult to freeze for intraoperative assessment2. In contrast, our open-top light-sheet microscope can non-destructively and rapidly assess fresh breast tissues with minimal processing. To demonstrate this ability, a large slice of human breast tissue measuring 2.0 × 2.0 × 0.4 cm (Fig. 3a) was stained with acridine orange for 20 s, rinsed for 10 s, and rapidly imaged using our open-top light-sheet microscope (Fig. 3c). It was subsequently submitted for post-operative histology of the same tissue surface (Fig. 3b).
In contrast to firmer prostate tissues40, the increased adipose content of breast specimens allows them to sit flatter on the glass-plate sample holder with reduced tissue surface irregularities (Fig. 3d). Thus, the open-top light-sheet microscope, which has the versatility to operate with a reduced vertical field of view (h = 80 μm), can achieve an increased imaging speed (v = 12.5 s cm−2) that results in a total imaging time of <1 min for a 4 cm2 tissue surface. This increased imaging speed is particularly advantageous for surgical guidance, where intraoperative pathology consultations must typically be completed within 20 min.
High-magnification views of relevant breast pathologies are shown in Fig. 3e–g. For example, invasive ductal carcinoma is characterized by nests and cords infiltrating through a desmoplastic stroma with an inflammatory response (Fig. 3e). Benign lobules (Fig. 3f) are visualized clearly at both medium and high magnification. The open-top light-sheet microscope comprehensively images benign fibro-adipose tissue (Fig. 3g) at a quality comparable to conventional archival FFPE histology. In contrast, frozen-section histology is poor, as adipocytes and strands of fibrous tissue with spindled stromal cells become distorted and difficult to visualize. Supplementary Video 3 shows an entire breast surface, and demonstrates the ability to zoom in and out.
Volumetric dual-channel imaging of a human biopsy specimen
A prostate core-needle biopsy, ~2 cm in length by 1 mm in diameter (Fig. 4a), was optically cleared (an overnight procedure) using a protocol based on the CLARITY method41, fluorescently stained with DRAQ5 and eosin (D&E), and imaged using the open-top light-sheet microscope. Conventional H&E histology was subsequently performed (Fig. 4b). Due to the translucency of the optically cleared biopsy, which allowed for deep imaging through the entire specimen, the image height (vertical field of view) of the microscope was increased to h = 1.28 mm for volumetric dual-channel imaging at a speed of v = 6.6 min cm−2 (total imaging time = 14 min). To minimize disruptions to longstanding clinical practice, and to facilitate rapid clinical adoption, the two fluorescent colour channels were false-coloured to mimic the colour palette of conventional standard-of-care H&E-stained tissue sections (Supplementary Fig. 8). A volume rendering is shown in Fig. 4d, along with orthogonal 2D cross-sections (sagittal, coronal, and axial) (Fig. 4d). Inspection of a H&E-stained histology slide reveals a region of carcinoma in a limited 2D view of the biopsy specimen (Fig. 4c). In contrast, the open-top light-sheet microscope enables slide-free comprehensive volumetric imaging of the entire specimen (Fig. 4d,e). Analysis of one region at multiple depths demonstrates that the plane of visualization influences whether the carcinoma is classified as Gleason score 3 + 3 = 6 (well-formed glands) or if the region contains the ‘poorly formed gland’ variant of Gleason pattern 4, with an overall classification of Gleason score 3 + 4 = 7 (Fig. 4e,f). When assessed by two pathologists, the grading varies between 3 + 3 (40% for pathologist A and 62.5% for pathologist B) and 3 + 4 (60% for pathologist A and 37.5% for pathologist B), with an inter-observer agreement of 47.5%, which is similar to published studies6. An inspection of the entire volumetric data set by both pathologists (Supplementary Videos 4 and 5) confirms that fully formed glands exist throughout the specimen (Gleason score 3 + 3 = 6), and that any glands that appear to be poorly formed (Gleason score 3 + 4 = 7) are an artefact of tangential sectioning.
We have developed an open-top light-sheet microscope to image freshly resected human tissues and biopsy specimens for non-destructive slide-free pathology. Unlike previous implementations of light-sheet microscopy, which have focused primarily on investigations in developmental biology and neuroscience, we showcase a highly versatile design that can accommodate specimens of large cross-sectional surface area (limited only by the mechanical travel of the microscope stage, in this case 11 × 7.5 cm) and provide value as an enhancement to gold-standard histopathology for a variety of applications (Fig. 1). The scanning area of our open-top light-sheet microscope is larger than any previously reported light-sheet microscope36 and the rapid 3D imaging capability of our device enables straightforward imaging of irregular and tilted tissues surfaces (Figs 2 and 3), as well as deep volumetric microscopy of biopsy specimens in toto (Fig. 4). Furthermore, trading off between the vertical field of view and imaging speed allows the microscope to be optimized for specific applications in clinical pathology (Figs 1-4).
Major advantages of our technology over traditional histology include greater sampling of clinical specimens (for example, comprehensive non-destructive microscopy of tissue surfaces and volumes) and the visualization of tissue microarchitecture in 3D, both of which would improve diagnostic accuracy. In addition, slide-free non-destructive microscopic pathology has the potential to conserve healthcare resources and to enable valuable specimens to be used for downstream molecular and genetic assays. Finally, our methods are compatible with current pathology standards, including the use of fast-acting fluorescence dyes (acridine orange, DRAQ5 and eosin) and optical clearing, which have been shown to not interfere with downstream histology or assays10,11, 42,43,44 .
As discussed in greater detail below, the potentially transformative diagnostic paradigm of open-top light-sheet microscopy has been explored for three selected clinical applications: (i) post-operative evaluation and triage of large surgical resection specimens, (ii) intraoperative imaging of large freshly resected specimens for margin assessment, and (iii) volumetric dual-channel imaging of core-needle biopsies. However, many other applications of this technology are possible, including high-throughput preclinical imaging (Supplementary Fig. 9), imaging of systemically delivered contrast agents for fluorescence-guided surgery (Supplementary Fig. 10), and volumetric immunofluorescence imaging for diagnostic pathology (Supplementary Figs 11 and 12).
Since traditional pathology is relied on as a clinical gold standard for disease diagnosis and patient-management decisions, the translation of open-top light-sheet microscopy into mainstream clinical use would likely benefit from a conservative staged approach. A relatively low-risk but high-value early application of this technology would be for rapid inspection and triage of large surgical excisions such as prostatectomy specimens, breast tumours and head-and-neck cancers. Currently, these tissues are grossly processed by dividing them into thick 3- to 5-mm-thick slices (bread loafs), which are further processed into FFPE tissue blocks so that a few thin (~5 μm) tissue sections may be mounted on slides, stained and viewed by pathologists. Due to the large number of tissues that must be processed from specimens such as a radical prostatectomy, many laboratories use partial-sampling methods, which triage the number of inspected specimens by random selection or gross inspection40,45,46. Our system can be used for rapid non-destructive triage inspection of fresh specimens to streamline the pathology workflow, where preliminary imaging studies demonstrate the ability to accurately identify prostate tissue slices containing carcinoma (Fig. 2). In particular, we show that our open-top light-sheet microscope is convenient for rapid wide-area imaging of irregular and tilted tissue surfaces (<200 μm for prostate bread loafs) via digital extraction of such surfaces from the 3D data sets, without the need for axial scanning or elaborate tissue-flattening and alignment mechanisms10.
Current re-excision rates for breast cancer patients undergoing lumpectomy procedures range from 20–50%47, a statistic that underscores the need for effective intraoperative guidance techniques. For invasive carcinoma, recent studies and consensus guidelines advocate for lumpectomies to be considered adequate when there is a ‘negative margin’, that is, one in which there is no cancer at the surgical margin surface (‘no tumour on ink’)48,49. Therefore, a method for accurate intraoperative tumour detection at the surgical margins could significantly reduce the costs, emotional trauma and potential complications associated with multiple surgeries. Open-top light-sheet microscopy with digital surface extraction (Figs 2c and 3c) enables rapid wide-area microscopy of large surgical-margin surfaces — unlike traditional histology, in which tissues are sectioned in the vertical direction at large intervals (3 to 5 mm) with minimal sampling of the tissue surface. Open-top light-sheet microscopy is non-destructive and does not interfere with conventional post-operative pathology, which can still be relied on for the identification of close margins (tumour within a few millimetres of the inked margin), if necessary. However, it should be noted that the historical preference to re-excise patients with close margins (such as for ductal carcinoma in situ, DCIS) may have arisen because of the sampling limitations of conventional pathology, and that this safety margin may prove to be unnecessary if a comprehensive surface-imaging approach is used. We have shown that our technology enables comprehensive surface microscopy of lumpectomy margins at high speed (12.5 s cm−2, or <1 min for the specimen shown in Fig. 3) with image quality that is superior to frozen-section histology, which requires 20–30 min to achieve (Fig. 3f). The imaging times for breast tissues are four times faster than for prostate tissues due to the improved compliance of fatty breast tissues40, which sit flatter on the microscope stage and require a smaller vertical field of view to reliably capture the tissue surface.
The diagnosis, grading and subtyping of various cancers, based on core-needle biopsy specimens, provides the basis for patient prognostication and for stratifying patients for various treatment options that are critical for their eventual outcomes and quality of life. Unfortunately, many studies have shown that the evaluation afforded by current histopathology techniques is often neither accurate nor reproducible, with high levels of inter-observer variability between pathologists regarding critical elements such as carcinoma grade and molecular phenotype5,6,50. For example, a recent study examined the inter-observer agreement within a panel of genitourinary pathologists in distinguishing between the ‘poorly formed gland’ variant of Gleason pattern 4 prostate carcinoma versus tangentially sectioned Gleason pattern 3 glands (‘well-formed glands’), and reported relatively poor inter-observer concordance (κ = 0.34)51. The distinction between Gleason pattern 3 and 4 carcinoma can result in dramatically different treatment recommendations: namely, active surveillance (Gleason pattern 3) versus radical prostatectomy (Gleason pattern 4), with its associated iatrogenic risks (incontinence, impotence, and so on). We have shown that the large degree of sampling afforded by open-top light-sheet microscopy in conjunction with optical clearing, along with the ability to visualize complex 3D structures, has the potential to improve the precision (inter-observer concordance) and accuracy (prognostic value) of the diagnosis and grading of prostate biopsy specimens (Fig. 4), with major implications for patient treatments and outcomes.
While the current system is relatively simple in design and implementation as a versatile general-purpose device for non-destructive pathology, more specialized open-top light-sheet microscopes may be developed for specific clinical applications. For example, in alternative system designs, the use of propagation-invariant illumination beams or tiling of thin Gaussian light sheets could result in improved resolution while maintaining a large depth of focus 52,53,54 . In addition, multi-view imaging methods and confocal slit detection could provide enhanced contrast and resolution55,56, and the parallelized use of multiple cameras could enable increased imaging speeds57. In particular, further improvements to our system aim to improve spatial resolution, which would have value for the definitive diagnosis of diseases.
There is institutional inertia that hinders the adoption of incremental changes in the centuries-old technological framework of histopathology, a practice that is foundational to patient-management decisions. For example, attempts to digitize histology with whole-slide scanners have not necessarily simplified the practice of preparing and interpreting histology slides, but have added time and expensive equipment to the pathology workflow without clear diagnostic benefits. In this study, we have developed an open-top light-sheet microscopy system that circumvents the need to prepare histology slides through non-destructive imaging of fresh clinical specimens that are simply placed on a glass plate (Fig. 1). This easy-to-use technology is highly versatile for a diversity of applications in clinical pathology, enabling both rapid surface microscopy and volumetric imaging of clinical specimens, with implications for streamlining the pathology workflow (Fig. 2), guiding surgical oncology (Fig. 3) and improving the diagnosis and grading of biopsied lesions (Fig. 4). Future studies are needed to investigate the capabilities of this technology to improve treatments and patient outcomes. Furthermore, these technologies will lead naturally to efforts in computer-aided diagnosis and automated image interpretation.
Open-top light-sheet microscope design
An optical schematic of the system is shown in Supplementary Fig. 3, along with ZEMAX ray-tracing simulations of the microscope performance. The illumination beam, emitted from a single-mode fibre with numerical aperture (NA) = 0.12, is collimated with a lens, L1 (f = 25 mm), reflected at 45° with a mirror, M1, and focused into the sample through a cylindrical lens, C1 (f = 100 mm), resulting in an illumination light sheet of NA = 0.03. These illumination optics generate a light sheet that is approximately 2 mm wide, with a full-width at half-maximum (FWHM) thickness of 7 μm (wavelength-dependent) at the beam waist, and a depth of focus of ~400 μm (2zR, wavelength-dependent), as shown in Supplementary Fig. 4. The illumination beam is then transmitted through a custom 15-mm-diameter fused-silica solid-immersion lens (SIL) (BMW Optics), a 0.5-mm-thick index-matching oil layer (Cargille Labs, Cedar Grove), and a 1-mm-thick glass-plate sample holder (fused silica, 10 × 10 cm). Optimal index-matching is achieved by applying a thin layer of saline or oil between the top surface of the glass-plate sample holder and the bottom surface of the irregular tissue sample. Light pressure is applied to the top of the tissue sample to remove any air bubbles in the saline or oil, thereby minimizing aberrations when imaging. The custom SIL is a hemisphere with a radius of 7.5 mm in which the centre thickness has been reduced from 7.5 to 7.375 mm such that the illumination and collection beams are index matched with minimal aberrations when they are positioned at a distance of 125 μm into the sample.
Fluorescence generated from the illumination light sheet within the tissue sample is collected by an objective lens, L2, (f = 43.47 mm, NA = 0.28, Olympus), reflected at 45° from mirror M2, focused through a tube lens, L3 (f = 150 mm), transmitted through an emission filter, F1, and then transmitted through an additional low-power cylindrical lens, C2 (f = 2,000 mm, IDEX Optics and Photonics), to mitigate astigmatism. The total magnification of the collection arm, from sample to detector, is 1.46 × (fL3/fL2) = 5, where the SIL provides a magnification factor that is equivalent to the refractive index n = 1.46. A dual-channel image-splitting module (W-VIEW GEMINI, Hamamatsu Photonics) is used to image two fluorescence imaging channels onto two halves of a high-speed scientific complementary metal-oxide–semiconductor (sCMOS) detector array with 2,048 × 2,048 pixels (ORCA Flash v2.0, Hamamatsu Photonics). Images are transferred at maximum data-transfer rates to a dedicated workstation comprised of a high-speed RAID0 hard disk using a CameraLink interface (Firebird PCI Express, Active Silicon). In addition to correcting aberrations, the SIL also increases the illumination and collection NA by a factor that is equivalent to the refractive index of the SIL material (fused silica, n = 1.46)38,39. This collection NA provides a field of view of 2 mm, and in-plane resolution (in the plane of the light sheet) of 1.5 μm within the sample. The images are acquired by the sCMOS camera at a sampling pitch of 1.25 μm pixel−1 along the plane of the light sheets, and a sampling pitch of x = 1.25 μm along the primary scanning direction (Fig. 1c). Serpentine scanning of serial volumetric imaging strips is achieved with a high-speed motorized stage (MLS-203-1, Thorlabs), where the image strips are acquired with a separation of y = 2 mm (secondary scanning direction).
Imaging speed is limited by the camera frame rate for the acquisition of serial light-sheet images spaced by 1.25 μm. For imaging fresh human tissues, a 256 × 2,048 sub-region (crop) of the sCMOS chip is typically used (0.32 × 2.56 mm field of view), which enables a frame rate of 800 frames per second. This frame rate results in an overall imaging speed of v = 50 s cm−2 of tissue. For more pliable breast tissues, a 64 × 2,048 sub-region of the sCMOS chip is used (0.08 × 2.56 mm field of view), which enables a frame rate of 3,200 frames per second and an overall imaging speed of v = 12.5 s cm−2 of tissue. Finally, for deeper volumetric imaging of optically cleared samples, a larger 1,024 × 2,048 sub-region of the sCMOS chip is used (1.28 × 2.56 mm field of view) with a frame rate of 200 frames s−1 and an overall imaging speed of 3.3 min cm−2 of tissue. For dual-channel imaging, the required imaging sub-region on the sCMOS chip is doubled, resulting in half the frame rate and imaging speed (for example, 100 frames s−1 and an overall imaging speed of 6.6 min cm−2 for the dual-channel imaging shown in Fig. 4).
Raw open-top light-sheet microscopy images undergo several image processing steps to render the final images. The processing is written using a combination of MATLAB and ImageJ tools in Miji (a Java package that allows transfer of imaging data between MATLAB and ImageJ).
Each individual image must be flat-field corrected for variations in intensity across the Gaussian light sheet, vignetting of the objective and tube lens collection optics, inter-pixel variations in the sensitivity of the sCMOS camera, and intrinsic aberrations in the open-top light-sheet microscope. To account for all of these effects, before imaging, a drop of a homogeneous fluorescence staining solution is imaged to acquire a reference image. Every subsequent raw image is normalized by this reference image to perform flat-field correction.
The captured images correspond to oblique planes oriented at 45° with respect to the sample surface. However, when being stored digitally, these images form a data cube in which the images are oriented at 90°. Therefore, post-processing, the images must be sheared by 45° in the x–z plane to create a trapezoidal data volume in which the tissue structure is undistorted. The sheared en face images may exhibit minor striping artefacts due to mechanical vibrations. These artefacts are removed in ImageJ using a stripe filter based on a combined wavelet-Fourier filter58.
After stripe filtering, each image strip is registered to adjacent image strips using the ImageJ grid-stitching algorithm59. For digital registration of adjacent image strips, an overlap of 500 μm is used between the image strips, and a linear blending of each image strip is used to generate the final image mosaic. This operation is initially performed for a single en face image mosaic in the x–y plane. The resulting registration and blending operations, determined from this single plane, are then applied to all of the imaging data at each depth, z. The imaging data now represents a 3D volume of multiple registered, blended and stitched image mosaics. For fresh, scattering tissues, to further reduce the dimensionality of the data, and to extract only the irregular surface of the sample, an extended depth of field (EDF) algorithm may be used60. For images acquired using two fluorophores, the cytoplasmic (eosin) and nuclear (DRAQ5) channels were false-coloured to resemble H&E histology as in a recent study42, using an algorithm described previously61.
Collection and analysis of fresh human prostate tissues
De-identified fresh prostate samples (25 samples from 19 patients) were obtained prospectively from an genitourinary biorepository (approved by the University of Washington Institutional Review Board, IRB), with patient consent, and were imaged using the open-top light-sheet microscope. After imaging, the samples were either snap frozen in a slurry of liquid-nitrogen-cooled isopentane (N = 24) or formalin fixed and paraffin embedded (N = 1). Multiple H&E-stained slides were prepared from each tissue sample (en face sectioning near the imaged tissue surface), and the slide that was most representative of the open-top light-sheet microscope image was scanned to produce a whole-slide image (Aperio, Leica). Approval was obtained from the IRB. The descriptive study measuring irregularities in prostate tissue surface topology did not measure an effect size, and thus did not undergo a pre-specified power analysis. We halted the study after collecting 25 samples, when additional samples would have led to minimal changes in the overall standard deviation of the measured tissue surface irregularities.
Collection of fresh human breast tissue
Human breast samples were collected through the University of Washington Northwest BioTrust (NWBT) under an IRB-approved protocol with patient consent. Imaging was performed immediately after surgery, after which H&E slides were obtained. Frozen sectioning was also performed on one human breast sample.
Fresh tissue processing
For fresh tissue imaging, samples were stained with 1 mM of acridine orange for 20 s, and then rinsed in phosphate-buffered solution (PBS) for 10 s, blotted dry, and immediately imaged.
Collection of human prostate core-needle biopsies
Core-needle biopsy specimens were obtained from fresh ex vivo prostatectomy specimens using an 18-gauge (approximately 1 mm inner diameter) needle biopsy device (Bard Max Core, Bard Biopsy). The prostate core was immediately placed in 10% neutral buffered formalin, where it was maintained at room temperature for 48 h. All samples were obtained from an IRB-approved genitourinary biorepository with patient consent.
Human prostate core-needle biopsy processing
The prostate core-needle biopsy was removed from formalin and rinsed for 10 min in 1× PBS. Next, the biopsy was processed using an automated electrophoretic tissue-clearing system (X-CLARITY, Logos) for 5 h at 33 °C. Following X-CLARITY processing, the biopsy was rinsed at room temperature in PBST (0.2 % Triton ×100 in 1× PBS v/v) for 6 h, with agitation. The biopsy was then stained by submerging it in a 1:1,000 v/v solution of eosin Y (Surgipath SelecTech Alcoholic Eosin Y 515) in 1× PBS for 20 min. After removal from the eosin solution, the biopsy was rinsed for 6 min in 3 changes of distilled water (2 min per rinse). The biopsy was then incubated in a 1:1,000 v/v solution of 5 mM DRAQ5 (Thermo Scientific) in 1× PBS for 20 min. Finally, the biopsy was rinsed in 1× PBS for 1 min before being transferred into dilute 2,2'-thiodiethanol (Sigma Aldrich) with a refractive index of n ≈ 1.46 for 30 min (for refractive-index matching). All steps listed in this section were performed at room temperature.
Blinded evaluation of biopsy regions of interest
Forty images (in 5 μm increments to a depth of 200 μm) were obtained from five independent regions of interest for a total of 200 images. The images were randomized and de-identified for a blinded pathology analysis. Pathologists assessed each image for the presence or absence of carcinoma and grade (if applicable). The final analysis consisted of only the single region shown in Fig. 4. Images from the additional four regions of interest were included in the randomization to reduce bias during blinded pathologist assessment.
All raw and processed imaging data generated in this work, including the representative images provided in the manuscript and Supplementary Information, are available from the authors upon request.
The custom computer code used in this study is available at GitHub https://github.com/nicholasreder/alpenglow/tree/v0.1.
How to cite this article: Glaser, A. K. et al. Light-sheet microscopy for slide-free non-destructive pathology of large clinical specimens. Nat. Biomed. Eng. 1, 0084 (2017).
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The authors thank N. Sanai of the Barrow Neurological Institute, St. Joseph’s Hospital and Medical Center (Phoenix, Arizona) for providing the human glioma sample, and B. Najafian for providing the human kidney tissue sample. Human breast specimens were provided by the NorthWest BioTrust, which is supported in part by the NCI of the National Institutes of Health (P30CA015704). Human prostate specimens were provided by the GU Specimen Biorepository, University of Washington, which is supported by resources of the Department of Defense Prostate Cancer Research Program (W81XWH-14-2-0183), the Pacific Northwest Prostate Cancer SPORE (P50CA97186), a PO1 NIH grant (PO1 CA163227), and the Institute for Prostate Cancer Research of the University of Washington. This work was also supported by resources from the NIH/NCI (R01 CA175391 and F32 CA213615), the NIH/NIDCR (R01 DE023497), the University of Washington Royalty Research Fund, and a University of Washington CoMotion Innovation Award.
Ease-of-use and scanning trajectory of the open-top light-sheet microscope. The sample is a fresh prostate tissue slice, stained with 1 mM acridine orange for 20 s, and rinsed for 10 s in 1× PBS.
Open-top light-sheet microscopy imaging dataset from a piece of fresh human prostate tissue. High-magnification regions of benign glands and stroma are shown.
Open-top light-sheet microscopy imaging dataset from a piece of fresh human breast tissue. High-magnification regions of benign lobules and invasive ductal carcinoma are shown
Volumetric visualization of a human prostate core-needle biopsy with two representative zoomed-in (high-magnification) depth-varying image stacks (sagittal) revealing 3D glandular morphology.
Zoomed-in region of a human prostate core-needle biopsy, in which a stack of depth-varying ‘sections’ is shown (at ~5-μm increments) to mimic a stack of conventional H&E–stained tissue sections on glass slides.