Introduction

Acute respiratory distress syndrome (ARDS) can result in loss of life, and is caused by injury to the capillary wall either from trauma or illness1. Fluid from the damaged capillaries escapes into the alveoli and bronchiolar tissue structures, impeding normal lung function. The mortality rate for the disease can be as high as 45% depending on severity2. Patients who survive the disease can be left with serious and long-lasting effects. Few adequate therapeutic modalities exist to treat the disease2. ARDS is also one of the associated complications that can develop in severe cases of COVID-191. ARDS can present in a histological examination as diffuse alveolar damage (DAD) with oedema, haemorrhage and intra-alveolar fibrin deposition3. Researched treatments include stem cell therapies, applied to reduce the effects of structural damage to lung tissue found in ARDS patients. Currently no therapies have been developed to treat the associated diminished functional lung capacity4. It is hoped that novel approaches to imaging and tissue analysis can aid in that process.

The gold standard practice in conventional histopathology for diagnosing and analysing such lung disease is to apply histomorphological assessment. Tissue samples are fixed in formalin, and embedded in paraffin. Typically, 5 µm slices are physically extracted from the larger sample block, and then stained using haematoxylin and eosin (H&E). The process of producing physical tissue slices using a microtome can result in damage to the tissue and deformation of soft tissue structures. This damage and deformation does not occur in histopathology methods that utilise X-ray computed tomography, since microtomes are not employed during sample preparation5.

Radiology was revolutionised through the development of computed tomography (CT) scanning. CT allowed radiologists to view tissues slice by slice throughout a patient’s body, and to then construct volumetric models6. In more recent advances, artificial intelligence and radiomic analysis has been applied to identify pathologies and segment various tissues from the greater sample volume7. This same advancement in diagnostic capabilities may be applied to histopathology through the use of high-resolution X-ray systems, high soft tissue contrast imaging techniques. Similar advances in artificial intelligence and radiomic analysis8.

The purpose of this study is a proof of concept to demonstrate the viability of such an imaging system. Conventional histopathology methods can take a significantly long time to produce an image slice and only allow for a 2-D view of a limited number of tissue slices9. In contrast, the system we are proposing can complete a tomographic scan in minutes and allow for a detailed analysis in both 2-D and 3-D projections throughout the entire sample, whilst maintaining the samples structural integrity.

Conventional absorption-based imaging may not allow for the necessary level of soft tissue contrast to accurately distinguish various tissue structures10. X-ray phase contrast computed tomography (X-PCCT) is used to allow for greater tissue contrast, segmentation and tissue structure analysis due to the inherent increase in soft tissue image signal found with the imaging technique11. To produce a 3-D histopathology imaging modality, it must also achieve a sub-micron resolution to differentiate individual cells. Ideally, the scan time should not be compromised for spatial resolution if the ultimate goal is to produce a system that is conducive to busy workflows found in clinical or research practice12.

X-ray phase contrast imaging techniques have been in development for decades. Research groups have been primarily focused on propagation/in-line and grating interferometry. Grating interferometry requires precision fabricated gratings with large aspect ratios, which can be technically challenging and expensive13. Propagation based imaging requires an X-ray source of high spatial coherence, and it also requires the user to make assumptions about the composition of the samples prior to scanning14. Speckle based scanning utilises a wavefront modulator, such as a sheet of sandpaper, to track phase displaced induced by a sample. This phase retrieval technique is inexpensive to setup, does not require an X-ray source of high spatial coherence or prior knowledge of the sample’s composition11.

The specific type of X-PCCT used during this study was the fly scan X-ray phase tomography technique. Fly scan mode was developed to overcome the challenges found with other speckle-based techniques, such as long scanning times and reduced spatial resolution. The imaging technique was refined through the use of a spatially coherent X-ray source, combined with the prerequisite beamline equipment, sample preparation and end user feedback. This technique was proposed and employed by Wang et al. in 201912.

Results

This proof-of-concept study investigates the soft tissue structure of unstained lung cores from two rat models, one that has developed ARDS and the other a control. The rodents were supplied by collaborators based in the Discipline of Anaesthesia, University of Galway. The first section, ‘imaging results’, assesses the imaging techniques capabilities in comparison to conventional high-resolution tomography and the gold standard in histopathology H&E staining. Assessments of how volumetric reconstructions can be generated displaying the samples gross tissue morphology were also performed. The second section, ‘biology results’, includes the analysis of data retrieved from the fly scan X-PCCT. The scan data was used to derive biological information from the sample and to assess the possibility of accurately identifying cell types visible in the scan (Supplementary information).

Imaging results

The contrast to noise ratio (CNR, see methodology for detail on CNR calculation) is a metric often used in image quality assessments across a range of modalities. The level of CNR for a specific medium has a significant impact on the observer’s/software’s ability to differentiate the tissue of interest (TOI) from the background (BG) tissue. This is of particular importance when generating volumetric renderings. The fly scan X-PCCT images for all tissue types are significantly higher than the CNR present in the corresponding absorption slice, resulting in an increase in CNR values of 20 to 60 times higher when comparing the two imaging techniques, as shown in Table 1. As expected, this is due to the image signal from the lung soft tissue being enhanced when generating an image based on a phase shift induced by the sample, rather than absorption of radiation. Image noise reduction in the X-PCI image is another factor in the increase in CNR. The signal gain for speckle-based X-PCI over absorption-based CT varies depending on the tissue analysed, as demonstrated in Table 1. The soft tissue gain is still significant for all tissue types.

Table 1 Comparison of contrast to noise ratio values taken from X-PCI and absorption slices for the three different tissue types present in the sample.
Full size table

Figure 1 visually demonstrates the variation in image quality across both modalities. The images have been optimised through windowing to best perceive the various tissues present. The difference in soft tissue signal is visually discernible between the two modes of image acquisition, X-PCI (A) on the left and absorption-based imaging (B) on the right. Both images are of the same slice along a core of lung tissue taken from a rat model. The image contains alveolar, bronchiolar and vascular tissue (Fig. 2).

Fig. 1
figure 1

Displays a qualitative side-by-side comparison of phase contrast (left, A) and absorption (right, B) tomographic slices of rat lung. Both slices are at the same position along the same 1mm diameter sample core of rat lung tissue. The tissue labelled with the blue arrow is an artery. The structure labelled red arrow is the lining of the bronchiole. Both are visibly easier to distinguish from the surrounding alveolar tissue in the S-XPCI image.

Full size image
Fig. 2
figure 2

On the left (A), a volumetric rendering of the 1mm diameter core. On the right (B), bronchiole and vasculature structures segmented from the rest of the core, apparent in the vasculature are the remaining red blood cells. Leukocyte’s, red blood cells, type one and type two pneumocytes are dispersed within the alveolar tissue, these are depicted as red in Fig. 2A. Vascularisation and bronchiolar structures were removed to demonstrate the possible observation and analysis of specific tissue structures present within the sample. The Fig. 2B displays the red blood cells that were still present in the vasculature after sample fixing.

Full size image

Qualitative analysis of the overlay of X-PCI and H&E slices in Fig. 3 demonstrates that the gross tissue morphology was generally maintained. Some degree of displacement in tissue is visible in certain areas of the imaged slice, particularly around the edges of the blood vessels. This is likely due to the slide preparation and use of microtome to cut the approximately five-micron slice. This same displacement was found in other studies, where it was mentioned the likely cause was due to the cutting action of the microtome15.

Fig. 3
figure 3

A side by side, comparison of images of histology H&E stained slice, labelled (A) and phase contrast tomographic slice labelled (B). Both slices are of the same area of rat lung tissue. The comparison of images qualitatively demonstrates the ability the phase contrasts technique to accurately represent soft tissue structure.

Full size image

The increase in CNR allows for automated segmentation and removal of specific tissue types using Amira software as demonstrated in the renderings in Fig. 4. 3-D renderings of specific tissue structures within the sample could be quantitatively analysed. The software was able to distinguish between bronchioles, vasculature and alveoli tissue structures as demonstrated in Fig. 4. This is something that was not possible when using this software in combination with the same absorption-based tomographic.

Fig. 4
figure 4

The image above is an overlay of the two separate images in Fig. 4. This image demonstrates how phase values were acquired for the specific cell types. Various cell types are denoted by arrows with different colours representing each cell type. Red blood cells (yellow), type 1 pneumocyte (blue), type 2 pneumocyte (purple) and leukocyte (red). Some distortion in the blood vessels in present in the H&E image, this is likely due to the slide preparation process.

Full size image

Biological results

A percentage increase of 5.6% in alveolar tissue density was measured between the diseased and control sample. This is in line with assumptions that S. aureus would induce inflammation and thickening of the alveolar walls. Once all other tissues besides alveolar tissue were removed from the 3-D volumetric rendering, measurement of inflammation between the diseased and control samples was possible. In the control, 21% of the scanned volume consisted of alveolar tissue, compared to 26.6% for the diseased sample. Methods of measurement are discussed further in the methodology section.

There were concentrated areas of increased intensity in both the X-PCI and absorption images. These areas were classified as cells of unknown description by collaborators at the University of Galway. To identify these cells and ensure faithful representation of the tissue’s morphology, an overlay of the stained lung tissue image taken under a microscope and the corresponding phase contrast slice were used as shown in Fig. 3 and used to generate the overlay in Fig. 4.

This study also looked at the possibility of distinguishing cell type by phase value alone. The cells were identified by reviewing the H&E stained image at the Discipline of Anaesthesia, University of Galway. Four cell types were predominant in the viewed slides: type 1 pneumocytes, type 2 pneumocytes, red blood cells, and leukocytes. These results were used to suggest whether a particular cell type could be identified by phase value alone. A comparison using the pairwise T-test was performed (no correction method). P value comparisons between the RBCs with the leukocytes, type 1 & type 2 pneumocytes were > 0.05. One can then conclude that there is no statistically significant difference in voxel value between RBCs, leukocytes, type 1 pneumocytes and type 2 pneumocytes. Those results indicate there is no way to decisively differentiate cell type by voxel value alone. Table 2 outlines the measured voxel values for each cell type and the distribution of phase values within each cell type grouping. Measurements from fifteen cells from each specific cell grouping were taken.

Table 2 The refractive index decrement values for identified specific cell types.
Full size table

Table 3 outlines the results found from the statistical analysis for each cell grouping. Grouping and statistical analysis was done to ascertain if the specific cell type observed could be differentiated from one another by voxel value alone. The pairwise comparison found no statistically significant difference between cells. It may be worth noting that the values displayed in each voxel are refractive index decrement values. This is the degree to which the X-ray beam refracts as it passes through the volume of tissue within that voxel.

Table 3 Pairwise comparisons using the t test, no correction methods applied.
Full size table

There is a subtle degree of ring artifact present in Fig. 3. Background measurements were taken in the presence of the ring artifact and areas where no ring artifact was visible on the slice at any window level or width. Five regions of interest were placed in both regions. The variation in mean and standard deviation in refractive index decrement values were measured for both sets of values. The variation between the mean and standard deviation for the sets of measurements were 0.7% and 8.9% respectively. The presence of ring artifact was found not have any significant bearing on measured refractive index decrement values.

Discussion

In summary we have determined that the fly scan X-PCCT technique developed by Wang et al. in 2019 could potentially be used for histopathological assessment of lung disease, namely S. aureus induced ARDS. Work to date on this specific phase contrast technique has focused on technique maturity to reliably and repeatably produce images that could be used for these purposes. This study builds on the work carried out by Wang et al. in 2019, demonstrating for the first time, how the Fly scan 1-D speckle-based X-PCI technique may be used in the field of histopathology. We assess where this technique could exceed or match current standard practice and where additional advancement is required. The aim of this study is to produce an imaging modality that can facilitate micron scale soft tissue structures in 3-D and 2-D rendering allowing for novel pathological tissue examinations and aiding in the development of improved treatment and diagnostic capabilities.

The initial objective was to confirm findings by numerous other studies demonstrating an increase in soft tissue signal with fly scan X-PCCT images when compared to absorption-based X-ray images. Our study replicated the findings of an increase in soft tissue signal and a decrease in image noise. This resulted in greatly improved contrast of the constituent tissues of the rat lung models when compared to conventional absorption-based X-ray imaging. This was demonstrated both quantitatively and qualitatively in Table 1 and Fig. 1. The measured increase in contrast to noise ratio for the various tissue types found in the lung samples are greater than one order of magnitude higher than was measured in the absorption-based images. When visually assessing the sample in a qualitative manner, the various tissues can be clearly differentiated from their surroundings. The significant increase in CNR was of vital importance for 3-D modelling, segmentation and analysis of specific tissue types. Without it, interrogation of tomographic data from different soft tissues becomes infeasible.

Since a microtome was not required prior to tomographic imaging at the synchrotron the structural integrity of the sample was maintained. Non-destructive generation of tomographic tissue slices were possible across the entire area of the 1 mm diameter biopsy core using Amira Software. The segmentation allowed for quantitative gross tissue analysis of individual tissue types present in the sample and it was possible to detect a minor difference in alveolar tissue volume of just under 6% between the control and ARDS sample. Breakdown of any tissue structures should be visible and quantifiable through the use of volumetric reconstructions and tissue segmentation demonstrated in this study. Increase in tissue volume due to pathologies such as inflammation and breakdown of tissue structures is strongly indicative of certain lung diseases.

Unfortunately, the sample was destroyed the slide preparation process. This meant a direct comparison between the quantitative gross tissue analysis with the Amira software and any physical quantification of tissue volume was impossible. It is recommended that any future studies use two sets of samples, one to be used for sample prep and another to be used for volumetric comparison. A study carried out by a co-author of this study at the University of Galway, assessed the increase in lung tissue volume in ARDS rat models when compared to sham models. Stereological assessment found a comparable increase in alveolar tissue volume, approximately 4%16.

Figure 2 demonstrated the ability of the imaging modality to generate 3-D gross tissue volume renderings, and the identification and segmentation of specific tissues from the whole sample rendering. Qualitative gross tissue assessment is useful in the identification of disease type and staging of disease.

Cells were visible in the alveolar tissue in the X-PCI CT slices. These were subsequently identified as red blood cells, type 1 pneumocytes, type 2 pneumocytes and leukocytes using the standard histopathology practice of H&E staining and assessment under a microscope. Using the overlay in Fig. 4 it was possible to identify the specific cell types in the phase contrast image. The recorded refractive index decrement values for each identified cell type displayed no statistically significant difference, as shown in Table 3. Therefore, it was not possible to differentiate the cells by refractive index decrement values alone.

Although cell type differentiation cannot be confirmed by visual assessment or measurement of phase values alone, in future studies it may be worth assessing alternative methods to identify cells such as AI and radiomic analysis, which could be coupled with parallel immunohistochemical identification of cell types to aid training of algorithms. The same radiomic and AI analysis tools could also be used for AI driven cell labelling, tissue labelling and segmentation studies, which could all be carried out in parallel. A limitation in this study was the small number of samples utilised; however, our data encourages further evaluation of this imaging technique as an alternative to conventional histopathology practice. It is the belief of this research group that with further development this imaging technique has the potential to greatly assist in the histopathology process and therefore the diagnosis of diseases such as ARDS (Fig. 5).

Fig. 5
figure 5

Boxplot displaying the distribution of the refractive index decrement values for the various cell types identified in the phase contrast slice & H&E stained image overlay, Fig. 4.

Full size image

Methodology

X-ray phase contrast microscopy

The X-PCI CT scans were performed at Diamond Light Source’s B16 beamline. X-rays with an energy of 15 keV were selected from the bending magnet’s source using a double-multilayer monochromator, setup was as shown in Fig. 6. A piece of sandpaper with a grain size of 5µm was chosen as a modulator, which was mounted on a 2D piezo stage. The distance between the modulator and the source was approximately 41.5 m. The modulator was placed 0.425 m downstream of the sample, the distance between the sample and detector was 0.805 m. Images of the speckle pattern were collected using a high-resolution X-ray camera composed of a PCO.edge CCD detector and a microscope objective with a Lutetium Aluminum Garnet doped with Cerium LuAG (Ce) scintillator17,18.

Fig. 6
figure 6

Experimental setup at diamond light source B16 beamline.

Full size image

The camera system was focused with an effective pixel resolution of 0.7 μm × 0.7 μm. The exposure time for each speckle image was 2.0 s. Following the previous high speed X-ray phase tomography approach12, the fly-scan mode was used and controlled with a ZEBRA system12. The rotation of the sample for tomography was performed as a single command, moving the sample from − 92.5° to 90° and programming the trigger signal to be raised as the stage passed − 90°, and at each increment of 0.2°. The wavefront modulator was then moved to its next position with a spiral trajectory over a range of 100 µm19,20, and the same trigger program was repeated at 30 modulator positions. In this way, the tomography rotation is performed in between modulator steps, saving all the time overhead associated with moving and stopping of the rotation stage. The reference speckle images were then collected by moving the sample out of beam and scanning wavefront modulator with the same spiral trajectory. Once the fly-scan tomography scan was finished, the collected sample speckle images were then reorganized in sequence according to the modulator positions for each projection.

A Gaussian filter was first applied to the raw speckle images to minimize the speckle noise and improve the tracking accuracy. A small window (3 × 3) was selected for each pixel for the stack of sample speckle images, and a larger window (11 × 11) was selected for the stack of reference speckle images for each projection. The selected window was then transformed in polar coordinates, and the virtual speckle stack is then generated along the polar direction and the spiral scan direction19. A sub-pixel registration DIC algorithm was then applied between virtual sample speckle image and reference speckle image for multiple polar directions21. The maximum of the cross-correlation coefficient can be precisely located for each pixel position. The speckle displacement induced by the sample was related to the coordinate along the polar direction of the maximum of the cross-correlation coefficient. The horizontal and vertical wavefront gradients can be derived from the sample to detector distance L, detector’s pixel size P and speckle displacement ξ at polar angle θ = 0° and θ = 90°. They can be written as:

$$\left\{\begin{array}{c}\frac{\partial \Phi (x,y)}{\partial x}=\frac{2\pi }{\lambda }{\alpha }^{x}(x,y)\approx \frac{2\pi }{\lambda }\frac{{\xi }^{\uptheta =0^\circ }P}{L}\\ \frac{\partial \Phi (x,y)}{\partial y}=\frac{2\pi }{\lambda }{\alpha }^{y}(x,y)\approx \frac{2\pi }{\lambda }\frac{{\xi }^{\uptheta =90^\circ }P}{L}\end{array}\right.$$
(1)

where αx and αy are the refraction angles in the x and y directions respectively, and Φ is the phase shift. The phase shift Φ induced by the sample can then be obtained from the derived phase gradients by using the Fourier transform relation between a function and its derivative as follows:

$$\Phi (x,y) = \frac{2\pi }{\lambda }{\mathbb{F}}^{ - 1} \left[ {\frac{{{\mathbb{F}}\left[ {\alpha^{x} (x,y) + i\alpha^{y} (x,y)} \right](m,n)}}{2\pi i(m + in)}} \right](x,y)$$
(2)

where \({\mathbb{F}}^{ - 1}\)\(\left( {\mathbb{F}} \right)\) is the inverse (forward) Fourier operation and \((m,n)\) are the variables in the Fourier space corresponding to the real space variables \((x,y)\)13.

Speckle data processing for each projection was performed with the help of a compiled Matlab code22, which was run using parallel processing distributed over 30 nodes of the Diamond Light Source (DLS) computation cluster. It took about six hours to process all 900 datasets using a region of interest of 2560 × 500 pixels. Since the produced phase shift Φ is proportional to the line integral of the refractive index decrement δ, it can be directly used in phase tomography reconstruction with filtered back projection reconstruction methods23. The refractive index decrement δ was reconstructed with Savu by applying filtered backprojection methods using the Ram-Lak filter24. Savu is a software package used for tomographic reconstructions of large image datasets25.

Measurements of materials using this phase retrieval method with a known X-ray refractive index decrement were calculated by the Wang et al. method11. The derived values of refractive index decrement and the values from literature were in good agreement within uncertainty. The derived values for PMMA & PTFE were 1.18 × 10−6 (± 0.02) and 1.95 × 10−6 (± 0.01), while the literature stated values were 1.19 × 10−6 and 1.95 × 10−6 respectively. These values are approximate to the refractive index decrement for soft tissue measured in previous studies at Diamond Light Source11.

The same experimental arrangement was used to acquire the absorption based tomographic scan and utilised the same fly scan technique but it did not require the 30 modulator positions. The scan time, therefore, was 30 times shorter and did not require any use of phase retrieval algorithms. The tomographic scan was reconstructed using the same Savu software and filtered back projection methods.

Figure 7 is a visual depiction of the data acquisition workflow for this study. It outlines sample preparation, both phase and absorption tomographic data acquisition, tomographic analysis, physical slice peroration and imaging, and cross comparison of data sets.

Fig. 7
figure 7

A visual depiction of the experimental workflow for this study.

Full size image

Pneumonia lung sample preparation and scanning

All animal experiments were approved by the Animal Care in Research Ethics Committee (ACREC) at the University of Galway and licensed by the Health Products Regulatory Agency (HPRA) Ireland. All experimental methods were performed in accordance with ARRIVE guidelines. All methods were carried out in accordance with relevant guidelines and regulations.

Adult male Sprague Dawley rats were anesthetised by inhalational of isoflurane and received a bolus of 1 × 109 colony forming units (CFU) of S. aureus Newman strain (obtained from Department of Microbiology, University of Dublin) in 300µL PBS, or vehicle, and allowed to recover. Animals were returned to their cages for monitoring for 48 h and then reanaesthetised, a tracheostomy was carried out and lung physiology parameters such as oxygenation and static compliance performed. Post-mortem, the left lobe of the rat lungs was isolated, inflated with and fixed in 4% paraformaldehyde (PFA) to prevent decay and preserve the structural integrity, as previously described26. This lung lobe was then cut into five sections and embedded in paraffin wax. In this study two cores 1mm in diameter were taken from larger samples embedded in paraffin wax. These cores belonged to two animals, one a control/uninjured animal and the other, an animal infected with S. aureus pneumonia.

After the scanning took place at Diamond Synchrotron, each core was re-embedded in paraffin wax to enable microtome sectioning. The embedded cores were sectioned into 5μm thick sections using a microtome (Leica Microsystems) and placed onto microscope slides (VWR International) for later Haematoxylin and Eosin staining (Sigma Aldrich). The slides were then fixed with DPX mountant (VWR International) and microscope images were taken at 20 × magnification (BX43 Olympus microscope). The samples were taken from a previous study conducted by J. Devaney et al. The sample collection method was approved by the Animal Care Research Ethics Committee of the National University of Ireland, Galway and conducted under license from the Department of Health, Ireland.

Volumetric analysis

When analysing the phase contrast images a thresholding algorithm could be employed to identify alveolar bronchi and vascular tissues which were present in the reconstruction. A thresholding algorithm was used to identify the different signal phases i.e., air, tissue and background. The lung tissue signal was processed using mathematical operation such as Closing and Fill Hole. This was carried out to completely enclose the full lung tissue volume found on the Amira software. Finally, the paraffin wax within the alveolar was subtracted from the lung tissue volume by using the Masking function available in the software.

In this proof of concept study two samples were used, one infected sample and one control. A volumetric representation of a core of 1 mm in diameter taken from the control sample is displayed in Fig. 2. In order to accurately measure inflammation in the diseased alveolar lung tissue, any non-relevant tissue had to be removed from the 3-D model. These discounted tissues included bronchioles and nearby vasculature, as shown in Fig. 2. Segmentation of macrophages, pneumocytes and red blood cells from the underlying epithelial cells was possible due to the relatively high phase signal of those cells when compared to the background alveolar tissue. The regions within the range of voxel values of 4.5 × 10−7 to 5.7 × 10−7 were identified as these cells and then highlighted as red in the volumetric rendering again using Amira software as shown in Fig. 2.

A quantitative image quality parameter, CNR was measured using ImageJ software27. Five square regions of interest (ROIs) of sizes ranging from 90 µm (120 pixels)–197 µm (262 pixels) were used to retrieve values for CNR measurements. The TOI measured were bronchiole, vasculature and alveolar tissue. Each tissue CNR was calculated as the difference between the absolute value of the TOI average signal, and the average signal of the tissue surrounding it or the BG tissue, all divided by the standard deviation (SD) of the BG tissue, i.e.:

$$CNR=\frac{{TOI}_{average}-{BG}_{average}}{{BG}_{SD}}$$
(3)

H&E stain was used to stain the physical slices later taken from the scanned cores. H&E staining agent is considered the gold standard of staining in lung histopathology, and used as the ground truth in this study. Once the H&E stain and the phase contrast tomographic slice were overlayed it was possible to identify cells in the phase slice. The phase values cells corresponding to different cells were recorded and tabulated to run statistical analysis on the results. A Pairwise comparisons using Wilcoxon rank sum exact test, P value adjustment method Bonferroni was used for statistical analysis.