Current Landscape of Advanced Imaging Tools for Pathology Diagnostics

Abstract

Histopathology relies on century-old workflows of formalin fixation, paraffin embedding, sectioning, and staining tissue specimens on glass slides. Despite being robust, this conventional process is slow, labor-intensive, and limited to providing two-dimensional views. Emerging technologies promise to enhance and accelerate histopathology. Slide-free microscopy allows rapid imaging of fresh, unsectioned specimens, overcoming slide preparation delays. Methods such as fluorescence confocal microscopy, multiphoton microscopy, along with more recent innovations including microscopy with UV surface excitation and fluorescence-imitating brightfield imaging can generate images resembling conventional histology directly from the surface of tissue specimens. Slide-free microscopy enable applications such as rapid intraoperative margin assessment and, with appropriate technology, three-dimensional histopathology. Multiomics profiling techniques, including imaging mass spectrometry and Raman spectroscopy, provide highly multiplexed molecular maps of tissues, although clinical translation remains challenging. Artificial intelligence is aiding the adoption of new imaging modalities via virtual staining, which converts methods such as slide-free microscopy into synthetic brightfield-like or even molecularly informed images. Although not yet commonplace, these emerging technologies collectively demonstrate the potential to modernize histopathology. Artificial intelligence-assisted workflows will ease the transition to new imaging modalities. With further validation, these advances may transform the century-old conventional histopathology pipeline to better serve 21st-century medicine. This review provides an overview of these enabling technology platforms and discusses their potential impact.

Introduction

For a pathologist to perform histological examination of a tissue sample, the sample typically must first undergo formalin fixation and paraffin embedding (FFPE).¹ Formalin fixation of the tissue ensures that the tissue structure and morphology remain intact during storage and processing and supports subsequent sectioning of the tissue. Thin tissue sections (usually about 4–6 μm) are mounted on microscope slides and stained with dyes that generate optical contrast for distinguishing cell nuclei, cytoplasmic components, extracellular matrix structures, and other microscopic features. The slides are primarily imaged with brightfield (white-light transillumination) microscopy to reveal features based on the localized absorbance of the stains. As is well known, the most commonly used stains are hematoxylin and eosin (H&E), which are almost always used in combination. Hematoxylin preferentially stains nuclei purplish-blue, whereas eosin preferentially stains cytoplasm and extracellular matrix pink. This blue/pink contrast allows for a close inspection of individual cells while also providing a general overview of the structure and distribution of cells and structural tissue components.

Apart from the incorporation of digital scanning in increasingly more numerous clinical settings, this histology workflow has remained the same for over a century, owing to its simplicity and general reliability, lab-to-lab and day-to-day variability notwithstanding. The standard FFPE workflow, from tissue sample to an interpreted slide, incurs delays of many hours or longer (days, weeks, or sometimes never, depending on the accessibility of histology facilities in low-resource settings). Additionally, chemical stains can be expensive and difficult to procure. Furthermore, there is a growing concern over shortages in the histology workforce, with vacancy rates remaining higher compared with other professions, and decreased enrollment in histology programs. However, despite the fact that these standard procedures are neither truly low cost nor enable rapid turn-around times, market forces have not (yet) forced an update—in noted contrast to other medical disciplines such as radiology and surgery that have, over recent decades, adopted now-widespread technically advanced approaches such as positron emission tomograph, magnetic resonance imaging, and surgical robotics.

In addition to the overall inefficiencies noted above, the standard histopathology workflow does not lend itself well to certain indications such as intraoperative surgical guidance or other point-of-care applications. The use of frozen sections, although often a viable alternative for providing rapid diagnostic or adequacy reads, comes with its own drawbacks.

For this reason, a reimagining of the histopathology workflow is needed. Such innovations may use novel imaging technologies to speed up histological examination at a lower cost while potentially providing novel contrast signals (and thus additional, possibly useful information). If they become reliably accessible, novel histology features provided by technologies that go beyond standard slides may actually enhance the value of current H&E histology, potentially yielding a more complete understanding of what can be gleaned from tissue specimens. Here, we provide an overview of recent innovations in imaging tools, on the reagent, hardware, and software sides, which can accelerate and enhance histopathology workflows. The discussion will touch on various approaches that go from highly complex, but feature-rich, to 3-dimensionally enhanced, to rapid methods that provide high-quality histology within minutes, without requiring the prior preparation of a physical glass slide.

Multiomics

In the talk presented at the 2023 Long Course at the USCAP 112th Annual Meeting, however, a brief mention of one flavor of image interpretation was made before the complex field of multiomics was briefly addressed. This, of course, refers to the demonstrated ability of pigeons to diagnose breast cancer from both histology and radiology images.² Although these educable birds were not being proposed as clinician substitutes, they nevertheless struck a nerve in the pathology/radiology plus artificial intelligence (AI) communities; evidence being the appearance of a recent publication entitled, “Pathologists aren’t Pigeons.”³ Skilled as they are, however, they would not be able to ingest the data that is currently being generated by high-complexity and spatially resolved molecular data. This whole field, which encompasses techniques that can resolve 10s to 1000s of molecular analytes down to subcellular resolution, is booming, at least in the research sector, and holds out the hope that novel findings will shed light on normal and disease-related manifestations. The technologies include approaches based on antibody-enabled proteomics, genome- and transcriptome-targeted methods, post-translational phenotypes, and others using imaging-based spectroscopic tools (mass spectroscopy, Raman spectroscopy, and infrared absorbance, for example) for characterizing both macromolecular and intermediate-size molecules such as lipids, carbohydrates, and other components. The field is far too complex to review here, and the reader is directed to some publications that describe the current state of the art.^4,5 It remains to be seen how these tools might find themselves in clinical practice, as they are expensive and complex to deploy, and the results are high dimensional and challenging to interpret. However, it seems likely that actionable insights can be distilled into lower-complexity assays and ultimately incorporated into patient care.

Slide-free Microscopy

Slide-free microscopy (SFM) techniques allow for the direct ex vivo examination of fresh, unsectioned specimens without time-consuming tissue processing, thereby overcoming the challenges of FFPE procedures.^6,7 In the context of intraoperative guidance, SFM can allow resected specimens to be imaged within a few minutes. There is much diversity in the capabilities and functioning of a SFM system, with methods variably involving previous tissue staining or clearing—or highly sophisticated optical setups—or having the advantage of being label-free. Such label-free methods are intriguing in that they simplify the process and the logistics and also have the potential to be translated for use in in vivo (in situ) imaging. Some systems may use low-cost and simple optical setups.

Technical demands on SFM techniques include the following:

1. The ability to obtain optical contrast that highlights diagnostically/clinically relevant microstructures. In standard histology, this is provided by stains such as H&E and absorbance-based imaging.

2. The ability to image a specific, often very thin, in-focus volume, eliminating other out-of-focus contributions. In standard histology, this is provided by physical microtome sectioning.

3. The ability to image a large field of view (FOV) at task-specific resolution. There is a tradeoff between imaging speed, FOV, and resolution. In recent years, standard histology has relied on whole-slide imaging scanners to scan specimens at cm-scale and up to ×40 magnification.

4. Having a high enough depth-of-field to account for the differences in sample surface topology. This is not an issue when working with thin tissue on slides, although dynamic focus is still required even with slide-based methods.

We cover several SFM techniques in development in the following sections. The key aspects of these technologies are summarized in the Table.

Table.

Key aspects of slide-free microscopy technologies

Modality	Preparation time	Contrast mechanism	Staining method	Equipment	Cost
Widefield	Variable	Transmittance, scattering, fluorescence	Intrinsic, optional absorbent, fluorescent dye(s)	Widefield microscope	$
Quantitative phase imaging (QPi), quantitative oblique back-illumination microscopy (qOBM)	0	Refractive index (R.I.; quantitative maps available)	Intrinsic (absorbent component can be subtracted out to another channel)	QPi on microscope, qOBM on microscope; QPi is available as software	$ to $$$
Confocal	Variable	Scattering, fluorescence	Intrinsic or fluorescent dyes	Laser, point, or line detector	$$$
Nonlinear optics (NLO), multiphoton excitation (MPEF)	Variable	Nonlinear effects SHG, THG, and MPEF	Intrinsic or fluorescent dyes	Ultrafast laser, point detector(s)	$$$$
Nonlinear Raman	0	CARS, SERS, and SRS	Intrinsic or Raman (specific bonds, ex. C=C, C-H, C=O)	Laser(s), with selection of wavelengths	$$$$
Photoacoustic		Photoacoustic effect	Intrinsic (absorbent)	Ultrasound transducer, laser	$$
Structured illumination (SIM)		Fluorescence	Fluorescent dyes	Camera	$$$
Light-sheet		Fluorescence	Fluorescent dyes	Laser, camera	$$$
MUSE	2 minutes	Fluorescence	Fluorescent dyes	LED, camera	$
FIBI	2 minutes	Fluorescence and absorption	Absorbent + fluorescent dyes, intrinsic	LED, camera	$
DUET	Already H&E-stained	Fluorescence and absorption	Absorbent dyes, autofluorescence	LED, camera	$

Cost ranges (approximate): $, $5000 to 100,000; $$, $100,000 to $350,000; $$$, $350,000 to 750,000; and $$$$, $750,000 to $2,000,000. In the United States, many biomedical research instruments are funded through the NIH S10 Instrumentation Program, whose Shared Instrumentation Grant (SIG) and High-End Instrumentation (HEI) programs match the latter two funding ranges. https://orip.nih.gov/construction-and-instruments/s10-instrumentation-programs.

CARS, Coherent anti-Stokes Raman spectroscopy; FIBI, fluorescence-imitating brightfield imaging; H&E, hematoxylin and eosin; LED, light-emitting diode; MUSE, microscopy with UV surface excitation; SERS, surface-enhanced Raman spectroscopy; SHG, second harmonic generation; SRS, stimulated Raman spectroscopy; THG, third harmonic generation.

Confocal Microscopy

Confocal microscopy was one of the first microscopy techniques to be explored for slide-free histopathology applications.⁸ It typically uses a point source of light, usually a laser, to illuminate a small focal spot in the specimen and manages to provide optical sectioning by filtering the collected light through a pinhole that rejects out-of-focus light. Alternative methods that use slit-illumination and slit-based out-of-focus rejection are also promising, as they trade minor hits in resolution with greatly enhanced scanning speed.⁹ Confocal microscopy approaches can be realized with at least the following two major modalities: reflectance confocal microscopy (RCM) and fluorescence confocal microscopy (FCM). RCM detects backscattered illumination light and obtains its contrast from the intrinsic optical properties of the tissue. FCM uses fluorophores that are selected to bind to cellular structures. Although confocal microscopy has some disadvantages, including slower imaging speed and higher instrumentation costs, it can provide high-quality cellular-level resolution. Confocal microscopy has seen significant utility in dermatopathology and has been clinically validated for such use cases.¹⁰ Its use with tissue specimens other than skin for histopathology applications is being investigated as well.¹¹

Nonlinear Microscopy

Specialized nonlinear optical effects can be used as well to develop SFM technologies. These include multiphoton microscopy (MPM) and Raman scattering. MPM uses multiple photons, typically from the near-infrared spectrum, which arrive nearly simultaneously at excitable molecules to induce fluorescence, with a potential for less photodamage and deeper tissue observation than is provided by regular one-photon excitation fluorescence. Tissue autofluorescence can provide sufficient contrast to provide useful signals even in the absence of specific staining.¹² There are several MPM microscopy approaches, such as two-photon excitation fluorescence, second harmonic generation, and third harmonic generation. MPM techniques unfortunately require expensive ultrafast lasers to induce such optical effects. Nevertheless, they have been extensively explored for utility in assessing breast pathologies,¹³ brain tumors,^14,15 and more.^16,17

Raman spectroscopy techniques probe the vibrational states of molecules, and abundant biomolecules such as proteins and lipids can be distinguished due to their intrinsic chemical signatures. Stimulated Raman spectroscopy (SRS), in particular, has been applied for various histopathology applications, especially for intraoperative histology during brain tumor resections (Fig. 1).^18,19 It was demonstrated that SRS, in combination with AI, can be used to accurately determine the malignancy status and grade of brain tumors from fresh tissue specimens.²⁰ Other improvements to Raman include coherent anti-Stokes Raman spectroscopy and surface-enhanced Raman spectroscopy. Surface-enhanced Raman spectroscopy is a surface technique, interrogating a specimen placed on a metal or nanostructured surface that uses surface plasmons to greatly enhance (10¹⁰) signals. SRS and coherent anti-Stokes Raman spectroscopy are available as extra-cost add-ons to some commercial confocal microscopes.

Figure 1.

(A) Instrumentation setup for stimulated Raman spectroscopy (SRS) microscopy. Reproduced from Ji et al.¹⁹ (B) SRS imaging, with pseudocoloring, of a resected specimen of a recurrent low-grade oligodendroglioma compared with standard hematoxylin and eosin (H&E). Reproduced from Orringer et al.¹⁸

Photoacoustic Microscopy

In addition to optical effects, other optically related physical phenomena can be exploited. Photoacoustic microscopy exploits a phenomenon in which localized tissue components absorb pulsed light at specific wavelengths; targets that absorb the light are briefly heated, leading to thermal expansion. Repeated expansion and contraction (after the pulse dissipates) cause the generation of a wide-band acoustic wave that is detectable with a transducer related to those found in ultrasound devices.^21,22 Although the pulses are nanoseconds in duration, they are repeated at much lower frequencies (in some cases 10 to 40 Hz); hence, the sound wave is readily detectable. As with ultrasound, a high-resolution three-dimensional (3D) image can be generated. Numerous SFM applications have been explored, such as for imaging breast tissues²³ and bone tissues.²⁴ Typical absorbers in the visible range are hemoglobins, melanin, etc., but if the exciting laser is centered in the 260 to 280 nm range, nucleic acids, which absorb in that spectrum, are rendered visible; hence, high-resolution images of nuclei are feasible as well.²³

Computational and Three-Dimensional Microscopy

Advances in computational microscopy have enabled novel approaches to conduct SFM. One such approach is structured illumination microscopy, which projects illumination light in user-defined patterns to enable optical sectioning²⁵ necessary for slide-free imaging (Fig. 2.) It is able to image large clinically relevant FOVs (multicentimeter) like prostate biopsies²⁶ with submicron-level resolution at high imaging speeds (video rate).^27,28

Figure 2.

Virtual histology images of several human pathological specimen types, obtained with the Instapath structured illumination microscopy fresh tissue microscopy scanner.

Such technologies, including structured illumination microscopy, can image tissue in a volumetric manner, enabling novel 3D histopathology applications. 3D spatial information is not available in standard histology, which only examines a single thin cross-section of a specimen or at most a small number of serial sections that are not necessarily well-registered spatially. Consequently, 3D histopathology, to date, has been underexplored; volumetric SFM may enable a more comprehensive and accurate depiction of tissue specimens, especially those that are heterogeneous.²⁹ Light-sheet fluorescence microscopy (LSFM) with optical clearing has been demonstrated to be a promising technique for 3D histopathology.³⁰ Optical clearing refers to sample preparation techniques that reduce light scattering in a tissue specimen and improve penetration.³¹ LSFM achieves optical sectioning by illuminating a thin plane oriented obliquely inside a specimen with a sheet of light and orthogonally collecting a resulting fluorescence image.³² By mechanically translating the sample, a 3D image can be obtained. As a fluorescence-based technique, the optically cleared sample is usually labeled with a nuclear dye such as DRAQ5 and a cytoplasmic dye such as eosin, which is fluorescent (green) and absorbent (red). LSFM was originally explored for imaging of nonclinical specimens, but recent work has demonstrated significant potential for histopathological applications.^32–34

As referred to above, there have also been interesting developments in low-complexity systems that hold out the promise of relatively inexpensive implementations. These have particular appeal for global and remote installations, as both money and technical expertise may be limited in such sites.

Microscopy With UV Surface Excitation

Microscopy with ultraviolet (UV) surface excitation (MUSE), is a recent SFM technique with significant promise for applications in histology.³⁵ This technology relies on the observation that sub-300 nm UV light has a narrow penetration depth in tissue of around 10 to 20 μm, only slightly greater than the thickness of a standard histology slide. Therefore, the excitation volume is limited to a thin layer that fairly closely corresponds to what is seen with standard brightfield microscopy, allowing MUSE to achieve high-contrast subcellular-scale imaging from the surface of thick tissue specimens. Conveniently, many common fluorescence dyes such as Hoechst, 4’,6-diamidino-2-phenylindole, and rhodamine can be excited by this sub-300 nm light while nevertheless emitting in the visible range. Consequently, although the excitation light is not transmitted by typical glass-based optics and must be directed to the sample surface obliquely, the emitted signals can be captured using standard, inexpensive conventional glass-based microscope optics and a color digital camera sensor. Numerous variants of MUSE have been developed in recent years, all relying on the basic mechanism of limited UV tissue penetration for optical sectioning. For example, Liu et al³⁶ implemented a compact and inexpensive version of MUSE, whereas Yoshitake et al³⁷ explored the use of water immersion to improve MUSE’s optical sectioning.

Fluorescence-Imitating Brightfield Imaging

Another SFM method, known as fluorescence-imitating brightfield imaging (FIBI), can also provide cost effective and rapid imaging of thick tissue specimens with results that are highly comparable with standard H&E histology.³⁸ FIBI relies on rapidly (1 to 2 minutes) staining the surface of fresh or fixed samples with the absorbing dyes H&E (conveniently the same stains used in conventional histology [Fig. 3]). The stains provide the color and spatial contrast needed to generate a useful image that is spatially confined to the near-surface region. The unusual aspect is that imaging is accomplished by illuminating the specimen using standard epifluorescence optics. The generated image resembles brightfield rather than typical fluorescence, however, because the excitation light (typically 405 nm provided by an LED) passes beneath the surface to generate diffuse, broad-band autofluorescence, some of which returns to transit the (stained) surface, where specific absorbance-based contrast occurs and, after collection by a standard microscopy objective, is transmitted to a conventional color camera. Because H&E stains are used to label the surface, the resulting images immediately resemble typical histology, as demonstrated in Borowsky et al.³⁸ Interestingly, although standard histology appearances can be captured because the specimens are physically thick, rather than micron-scale thin, it is possible to view interesting surface topology features, also illustrated in Figure 4. Although such images are currently not that familiar to histologists, it is possible that novel insights can be gathered.

Figure 3.

Top panel. Prostate. Thick formalin-fixed but unsectioned specimen was stained with hematoxylin and eosin and imaged with fluorescence-imitating brightfield imaging. Cycle Generative Adversarial Network mode-converting converting artificial intelligence was used to increase resemblance to conventional hematoxylin and eosin histology. Bottom panel. Fluorescence-imitating brightfield imaging can provides increased insight into three-dimensional surface topography. Left: core-needle biopsy of lung, with alveoli prominently visible. Right: large artery in the stomach, with view of interior wall as the vessel penetrates down from the cut surface.

Figure 4.

Three-dimensional (3D) light-sheet microscopy is used to image prostate specimens stained with an hematoxylin and eosin (H&E) analog (left). The images are converted to synthetic immunohistochemical (IHC) images using a 3D deep learning image-to-image translation approach. A simple threshold-based segmentation is used to volumetrically segment the glands (right). Reproduced from Xie et al.³⁴

As with MUSE, FIBI can provide subcellular contrast with remarkable clarity. It is also nondestructive and does not interfere with subsequent standard FFPE histology or molecular analyses. FIBI differs from MUSE in that FIBI images more immediately resemble those encountered using standard brightfield rather than fluorescence-based microscopy, although the use of color mode transformation tools (see below) can be used to generate MUSE images that resemble brightfield.

Virtual Staining and Modality Mapping With Artificial Intelligence

Slide-free and label-free microscopies often provide contrast and spatial signals that are unlike standard histology and may therefore be harder to interpret, making these techniques more difficult for histopathologists to adopt. However, converting these novel modalities to resemble H&E-stained slides (or vice versa) can allow for wider adoption. This general process of converting one microscopy modality into another (usually brightfield microscopy with H&E stains) can be referred to as microscopy mode conversion or virtual staining. Virtual staining is not limited to SFM, and its use with slide-based histology has been widely reported. For example, standard H&E-stained slides can be transformed into other stained images, encompassing histochemistry targets such as collagen or more broadly, fibrosis, as shown with Masson’s trichrome, for example, or used to infer the presence of specific molecular phenotypes, such as HER2 status.³⁹ A few examples of AI-based modality mapping are described here, but a more comprehensive overview is provided by Bai et al.⁴⁰

A pioneering work in virtual staining research is that of Rivenson et al,⁴¹ which investigated the use of deep learning to virtually stain unlabeled tissue-autofluorescence images. They trained neural networks on salivary gland tissue stained with H&E, kidney tissue stained with Jones stain, and liver and lung stained with Masson’s trichrome, using convolutional neural networks of the U-net architecture (a standard architecture for any image-to-image task). The networks were trained via the generative adversarial network framework, an approach that has been quite successful in various image synthesis tasks. The virtually stained images were evaluated by pathologists, and no significant difference between the virtually stained and regular brightfield images was noted. This indicates that virtually stained images can potentially provide similar diagnostic utility, although a number of outstanding challenges were noted.⁴⁰ Variability in the colorized virtually stained images is lower than that of authentic, histologically stained images, resulting in potentially useful stain appearance standardization useful for AI-based interpretations. Overall, such virtual staining approaches may enable histopathology workflows that bypass typically time-consuming and costly staining procedures.

Note that in many cases, this level of pixel-wise supervision (or correlation) can be challenging or even impossible to obtain for SFM virtual staining applications. For instance, if a tissue specimen is imaged with MUSE or FIBI, or other slide-free approaches, and then processed for standard histological sectioning to generate a corresponding H&E slide, the general features will be very similar, but the exact structures at a pixel level will not exactly be the same. This is a major challenge that has hindered the application of virtual staining to SFM applications. To address this issue, unpaired image-to-image translation frameworks can be used, which enables the learning of a transformation from images in some input domain A to images in some input domain B from a collection of domain A images and domain B images that are not necessarily coregistered. For SFM virtual staining, domain A would be the SFM images, and domain B would be the stained images. There are several approaches that have been explored in the computer vision and deep learning literature, but the most commonly used unpaired image-to-image translation algorithm is the Cycle-Consistent Generative Adversarial Network (CycleGAN) model.

One of the first studies demonstrating virtual staining of SFM images was Combalia et al. The authors converted confocal microscopy images to H&E using CycleGANs.⁴² The RCM and FCM images are linearly combined to form an RGB image, by assigning RCM images an eosin-like color, and FCM images a hematoxylin-like color. This naively converted image is then passed to a CycleGAN for final optimization. Importantly, it was demonstrated that deep learning–based “despeckling” (speckles are often encountered in laser-based imaging systems) is important for successful virtual staining. Without it, the CycleGAN generated unwanted artifacts.

Several other studies have applied CycleGANs for virtual staining, including one that explored the use of CycleGANs for photoacoustic microscopy of bone tissue.²⁴ Significant agreement was observed between the virtually stained images and the standard H&E images, as determined by statistical analysis of nuclear features and review by three pathologists and one orthopedic surgeon. Martell et al⁴³ also used CycleGANs to transform label-free images from their combined UV-PARS and UV scattering microscopy system into a realistic virtual H&E stain style. To generate the inputs to the CycleGAN, the UV-PARS and UV scattering images were first combined into pseudo-colored virtual images approximating H&E histology using a custom fusion algorithm. The CycleGAN was then trained on unpaired examples of these pseudo-colored images and true H&E-stained histology. This method was tested on human breast and prostate tissues and on freshly resected murine tissues, allowing the visualization of important histopathological features. In a blinded study, pathologists rated the virtual histology images as having superior nuclear detail and overall stain quality compared with frozen section analysis. Engel et al⁴⁴ used CycleGANs to map FIBI images of skin specimens to virtual H&E images. Dermatopathologist evaluation indicated that the relevant diagnostic and morphologic features were present in the virtual H&E images.

CycleGANs were also applied to convert quantitative phase images from quantitative oblique back-illumination microscopy into virtually stained H&E-like images.⁴⁵ A key advantage of oblique back-illumination microscopy is its ability to provide label-free, high-resolution 3D quantitative phase images of thick unprocessed tissue specimens.

Three-Dimensional Image Analysis

Deep learning–based approaches specific for 3D imaging data can be used as well, as demonstrated by Xie et al.³⁴ They applied an image translation method using generative adversarial networks to synthetically convert light-sheet derived 3D images of prostate biopsies stained with a fluorescent analog of H&E into synthetic images mimicking cytokeratin 8 immunofluorescence (Fig. 4). The virtually stained cytokeratin 8 images allowed automated segmentation of prostate gland structures with traditional computer vision algorithms. By avoiding the need for slow and expensive antibody staining, the pipeline provides an efficient way to obtain segmented glands from 3D whole-biopsy data sets stained only with a small-molecule fluorescent H&E analog. This virtual staining approach could be extended to other tissue structures and stains, providing a flexible and low-cost way to extract quantitative biomarkers from 3D pathology data. The authors showed that features derived from virtually stained and segmented glands had prognostic value for risk stratification of prostate cancer patients.

Dual-Mode Emission and Transmission

Finally, we can consider a surprisingly simple method for extracting more information from standard H&E-stained slides by imaging both in brightfield and fluorescence.⁴⁶ The approach, termed dual-mode emission and transmission microscopy, performs almost simultaneous imaging of a slide by switching from standard brightfield (transmission) mode to a single-wavelength excitation, long-pass emission fluorescence mode at every stage-stop during a scan, using the same camera to capture pixel-matched images. Interesting and useful contrast appears in fluorescence mode, including collagen, elastin, and other specimen-dependent components, avoiding the need for special stains. This approach also avoids the need for possibly fragile AI-based conversions as the new signals exhibit robust optical contrast in the fluorescence channel (Fig. 5).

Figure 5.

Dual-mode emission and transmission images showing novel features revealed in the fluorescence mode (panels C and D) compared with standard brightfield (panels A and B) of the identical field. (A and C) Invasive breast cancer with necrosis and the remains of a blood vessel. Details are hard to discern; however, in panel c, it is easy to distinguish remnants of the elastic lamina (bright) from adjacent necrotic areas. (B and D) An arteriole is clearly shown in the brightfield image, which does not show the details of the internal and external elastic laminae. Interestingly, the internal laminae are multicolored (yellow and blue)—the physiological significance is as yet unknown.

Conclusion

The era of simple H&E slides as the bread and butter for pathology interpretation seems to be gradually merging into a more sophisticated world in which novel optics, stains, molecular probes, and advanced computational tools can generate deeper views into what the specimens contain and, one hopes, faster and better clinical care supported by an invigorated discipline of anatomic pathology. The future is bright.