Polarization resolved deep ultraviolet microscopy for label free imaging with enhanced nuclei and fiber contrast

Abstract

Deep ultraviolet (DUV) microscopy is a rapid, label-free imaging technique widely used in biological applications. However, fibrillar structures, which provide crucial insights into tissue organization, are often overlooked. In this work, we introduce a polarization-resolved DUV microscope capable of extracting both nuclear and fiber features. Illumination at 265 nm enhances nuclear contrast, while polarization imaging reveals fiber orientation. Four images are captured sequentially to calculate the linear polarization properties of the tissue medium. To address pixel misalignment between multiple images, we apply adaptive local thresholds to extract valid features for precise registration. The degree of linear polarization and angle of polarization exhibit significant changes as light passes through tissue samples, revealing variations in polarization states. This additional polarization contrast offers a new dimension of analysis, potentially enhancing the characterization of biological tissues.

1. Introduction

Deep ultraviolet microscopy has recently regained prominence as a powerful technique for label-free biological imaging. Compared to visible-wavelength imaging, it offers several advantages, including higher spatial resolution due to the shorter DUV wavelength and the rapid extraction of quantitative information from biomolecules essential for tissue analysis [1–5]. Illumination at 265 nm enhances nuclear contrast due to its strong absorption peak, making it particularly effective for highlighting nuclear morphology [6–10]. The development of high-efficiency DUV LEDs has made deep-ultraviolet imaging more affordable and practical. However, the limited external quantum efficiency continues to restrict further improvements in the emission performance of AlGaN-based DUV LEDs. By employing tailored multiple quantum wells, a reflective Al mirror, a low-optical-loss tunneling junction TJ, and a dielectric SiO₂ insertion layer, light output powers of 140.1 mW at 850 mA have been achieved [11]. Moreover, the introduction of an ultrathin tunneling junction technology platform effectively mitigates high series resistance and deep-UV light absorption, enabling the lowest operating voltage reported to date for 275 nm deep-UV LEDs [12]. DUV microscopy has been successfully applied to prostate cancer grading, where UV spectral signatures from endogenous molecules create a phenotypical continuum that provides unique structural insights with nanoscale resolution in thin tissue sections. This continuum serves as a surrogate biomarker for prostate cancer malignancy, enabling the identification of glandular phenotypical shifts in patients with aggressive tumors [13]. Additionally, DUV microscopy has proven to be a robust technique for detecting and grading neutropenia. The automated framework enables accurate segmentation and classification of live, unstained blood cells in smears, distinguishing patients with moderate and severe neutropenia from healthy samples within minutes [1,6]. However, current deep ultraviolet imaging modalities predominantly focus on intensity variations, often neglecting the polarization information that could offer additional structural insights [14,15]. The polarization properties of collagen fibers in tumors provide valuable information regarding their orientation, which is crucial for understanding tissue structure. The density and alignment of fibrillary collagen are well-established markers for diagnosing early to late stage of various human cancers [16–19].

Furthermore, polarization imaging-based diagnostic methods offer the potential to minimize the reliance on time-consuming multi-staining processes, providing a more efficient means of obtaining quantitative data that can aid in the accurate staging of conditions such as liver fibrosis. High-dimensional polarization parameter imaging has proven effective in quantitatively characterizing the microstructural features of liver fibrosis tissues, underscoring the utility of polarization-based approaches in medical diagnostics [20]. In this work, we present a polarization-resolved deep ultraviolet microscope that not only captures intensity variations in nuclei but also extracts polarization information from fiber structures. This additional polarization contrast offers a new dimension of analysis, potentially enhancing the characterization of biological tissues. To acquire the signal, conventional polarimetry relies on polarization array cameras [21]. However, due to the limited availability of materials with good DUV transmission, commercially available polarization cameras lack sensitivity to DUV light and are generally unsuitable for DUV imaging applications [22]. Therefore, we utilize two Glan-Laser polarizers to retrieve linear polarization information. As depicted in Fig. 1, the deep ultraviolet polarization microscope employs a 265 nm LED (M265L4, Thorlabs) for sample illumination. A DUV lens (L4052, Thorlabs) is used to collimate the light and a Glan-Laser α – BBO polarizer below generated linear polarized light before the sample plane. The transmitted signal was collected through a DUV objective (M Plan UV 50x 0.42, Mitutoyo), then another Glan-Laser α – BBO polarizer was mounted on a motorized rotator (PRM1Z8, Thorlabs) working as an analyzer. The image was focused using an air-spaced tube lens (ACA254–100-UV, Thorlabs) onto a DUV sensitive sCMOS camera (PCO Edge 4.2UV, Excelitas). We captured polarization images at 0°, 45°, 90°, and 135° for each sample to calculate Stokes parameters S₀, S₁, and S₂, degree of linear polarization (DoLP) and angle of polarization (AoP) [23]. The incident power at the sample plane is 1.2mw. The image exposure time is 200 ms for each frame. The rotator is controlled with Kinesis software (Thorlabs). The total time of 4 frames measurement is 12 s and the total time of 36 frames is about 96 s. The optical fluence is about 0.22 μ j/μ m² [24].

Fig. 1.

Optical layout of the polarization resolved deep ultraviolet microscope, where two DUV GLB crystals are used for DUV polarization Stokes parameter measurements.

2. Methods

The full Stokes parameters S₀, S₁, S₂, and S₃ are defined in terms of six polarized flux measurements performed with ideal polarizers as [23]:

$$S=\left(\begin{array}{c}{S}_0\\ S_1\\ S_2\\ S_3\end{array}\right)=\left(\begin{array}{c}{P}_H+P_V\\ P_H-P_V\\ P_{45}-P_{135}\\ P_{\mathit{\text{right}}}-P_{\mathit{\text{left}}}\end{array}\right),$$

where P_H, P_V, P₄₅, P₁₃₅, P_right, P_{let f} are different states of polarized light at 0°, 90°, 45°, and 135° linear, and right and left circular polarization. DoLP quantifies the extent to which the polarization ellipse is oriented linearly, or equivalently, the degree to which the electric field remains confined within a single plane. When the DoLP is equal to 0, the light is either circularly polarized, unpolarized or a combination of both.

$$\mathit{\text{DoLP}}=\frac{\sqrt{S_1^2+S_2^2}}{S_0}.$$ (1)

AoP for linearly polarized light is the angle of the electric field oscillation measured anticlockwise from reference axis in radians

$$\mathit{\text{AoP}}=\frac{1}{2}{\mathit{\text{tan}}}^{-1}\left(\frac{S_2}{S_1}\right).$$ (2)

At least three intensity measurements are required to determine the linear polarization state. The signal-to-noise ratio (SNR) of the DoLP and AoP can be characterized by the following equations, as previously reported by Chen, et al. [25]:

$$\mathit{\text{SN}}{R}_{\mathit{\text{DoLP}}}=\frac{\mathit{\text{DoLP}}}{{\sigma }_{\mathit{\text{DoLP}}}}=\sqrt{\frac{\left(DoL{P}^2*S_0\right)}{1-\mathit{\text{DoL}}{P}^2*\mathit{\text{cos}}(4*\mathit{\text{AoP}})}}$$ (3)

$$\mathit{\text{SN}}{R}_{\mathit{\text{AoP}}}=\frac{\mathit{\text{AoP}}}{{\sigma }_{\mathit{\text{AoP}}}}=2\pi *\mathit{\text{DoLP}}*\sqrt{S_0}$$ (4)

This represents a closed-form solution, where σ_AoP and σ_DoLP denote the calculated noise levels associated with the AoP and the DoLP, respectively. To fully recover the Stokes parameters, additional DUV-transmissive quartz wave plates will be necessary. Since polarization analysis requires capturing multiple images of a scene using different analyzer orientations, discrepancies can arise due to variations in the field of view for a given pixel, potentially introducing errors in the measurements. Due to the rotational shifts of the DUV GLB crystal, the captured raw polarized images suffer from pixel misalignment as shown in Fig. 2. This misalignment introduces significant errors during data analysis, leading to inaccurate information, especially when calculating the DoLP. As a result, edge artifacts appear in the DoLP map, as shown in Fig. 2(e). To accurately align and shift back all four captured images, we first apply adaptive local thresholding to extract image features without being biased by intensity variations in the actual images [26]. This helps avoid alignment errors caused by large intensity differences between P_H and P_V, as shown in Fig. 2(a) binary mapping. After identifying valid features in the binary image, we use feature-matching information of the binary images to correct pixel-by-pixel misalignment in the real images. That is, we used the features in the binary images to calculate the geometric transformation parameters between all four images. We performed affine transformations to account for translation, rotation, scaling, and slight distortions. These parameters were then applied to register the original images using the scale-invariant feature transform (SIFT) [27]. This method effectively corrects the misalignment of captured polarized images, even in the presence of large intensity variations. To validate its effectiveness, we apply our method to DUV polarized microscopy imaging of a wax sample. Fig. 2(c) presents the error map comparing the raw and corrected polarized images, highlighting the reduction in misalignment. Furthermore, as shown in Fig. 2(e), pixel misalignment introduces significant errors in the calculated DoLP, demonstrating the necessity of precise alignment for accurate polarization measurements. Although residual alignment errors may still exist, especially under special or corner-case scenarios, our method consistently provided significant improvements and robust results within the context of our experiments. Furthermore, subsequent tests on biosamples revealed no noticeable visual artifacts or significant numerical discrepancies resulting from misalignment. Therefore, our methodology effectively minimizes the object-scene dependency related to alignment. Nonetheless, we acknowledge that further detailed comparative analyses and quantification of alignment errors in polarized imaging represent valuable directions for future research.

Fig. 2.

DUV polarized microscopy pixel misalignment correction: a) Raw captured polarized wax images obtained by rotating a GLB crystal at 0°, 45°, 90°, and 135°. b) Corrected polarized image of wax after binary mapping and pixel realignment. c) Error map showing the difference between (a) and (b). d) From left to right, calculated S₀, S₁, and S₂ are shown. e) From left to right, calculated DoLP using the uncorrected raw image, pixel-corrected image, and AoP are shown.

3. Experiment and results

We compared the resolution and contrast improvement between DUV and visible light. The resolution results were measured using objectives with the same numerical aperture: M Plan UV 50 × /0.42 (Mitutoyo) and M 20 × /0.4 (Nikon). As shown in Fig. 3 (a), under DUV illumination, the system resolves Group 10, Element 3 of the resolution target (Ready Optics), corresponding to a lateral resolution of 390 nm. In contrast, with visible light, the resolution is limited to Group 9, Element 3, or 780 nm. In addition, we have included a comparison of image contrast in thin sectioned lungs tissue in Fig. 3 (b). The unstained DUV image shows a significant improvement in nuclear contrast compared to the visible light brightfield image due to absorption peak at 260 nm. Bronchioles, which are surrounded by epithelial cells rich in nuclei, are clearly visible in the DUV image, yielding contrast comparable to hematoxylin-eosin (H&E) stained sections.

Fig. 3.

Resolution and contrast improvement of DUV illumination. a) Under DUV illumination, the system resolves Group 10, Element 3 of the resolution target, corresponding to a lateral resolution of 390 nm. In contrast, with visible light, the resolution is limited to Group 9, Element 3, or 780 nm. b) Comparison between visible brightfield, DUV and H&E lungs image. The unstained DUV image shows a significant improvement in nuclear contrast compared to the visible light brightfield image. Bronchioles, which are surrounded by epithelial cells rich in nuclei, are clearly visible in the DUV image, yielding contrast comparable to H&E-stained sections.

To evaluate the imaging performance in biological specimens, polarization images of mouse tissue samples were acquired. All mice were deeply anesthetized with ketamine/xylazine and euthanasia was performed via cardiac puncture. Fresh tissues were harvested from mice and then embedded in paraffins to create tissue blocks. These blocks were sectioned into 4 μm slices using a microtome. Tissue slices are placed on the high DUV transmitted quartz slides. There are no coverslips and mounting medium because standard thin coverslips do not transmit DUV light. Using a thicker slide as an alternative would introduce significant spherical aberration, which would degrade image quality [28]. To ensure the sectioned tissue adheres to the microscope slide without using a coverslip, paraffin wax was retained for stabilization. All animal care and experimental procedures were in accordance with ethical regulations, conducted according to the National Institutes of Health guidelines for animal research, and approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Arizona. All procedures were conducted in full compliance with the ARRIVE guidelines. For the accordance statement, all methods were carried out in accordance with relevant guidelines and regulations. We compared the imaging performance of tissue sample between DUV and visible light. Fig. 4 a) shows the DUV linear stokes images of mouse tissue and 4 b) shows the visible stokes images. As illustrated in Fig. 4 a), the normal structures of liver tissue are clearly discernible. The dark spots observed in the S₀ images correspond to the distribution of nuclei within the tissue. Upon passing through the scattering tissue medium, the linearly polarized light undergoes state change, which is evident in the DoLP images. Notably, vascular structures, such as veins, exhibited less change in polarization state compared to the surrounding tissue. In the normal liver tissue images, the polarization data revealed minimal differences, reflecting the homogeneity of the tissue in terms of polarization response. We also observed that the residual paraffin wax, remaining from the sample preparation process, exhibited notable angular variation in the images, emphasizing its distinct optical properties relative to the surrounding tissue medium. Compared to visible light images, the DUV images exhibit better contrast across S0, DoLP, and AoP. The vein structure is clearly visible in the DUV DoLP image but indistinguishable in the visible DoLP. We quantitatively estimated the SNR of the DoLP and AoP using Eq. (3) and Eq. (4) for DUV and visible polarized images of mouse liver and lung tissues. For DUV imaging, the SNRs of the DoLP and AoP across the entire liver region were 9.5 dB and 10.9 dB, respectively, while for the lungs they were 9.1 dB and 11.2 dB. In comparison, for visible imaging, the liver exhibited SNRs of 7.6 dB and 8.9 dB, while the lungs showed 8.1 dB and 9.2 dB.

Fig. 4.

a) DUV linear Stokes images of the mouse tissue. b) Visible linear Stokes images of the mouse tissue. The circular region at the top of the mouse lung has been identified as a vein, which is surrounded by alveolar spaces characterized by thin walls and densely packed nuclei. Vein structures exhibit less change of in polarization state. Normal tissue shows the uniform polarization state. Field of View (FoV), 180 × 200 μm.

We attribute this improvement to the fact that DUV light interacts more strongly with molecular bonds and fiber alignments, making polarization-based contrast more detectable. DUV wavelengths are shorter than the structural dimensions of many biological components, allowing for stronger interactions with subcellular structures, such as nuclei and collagen fibers. Furthermore, DUV light has higher absorption and scattering sensitivity to molecular-level anisotropy, which enhances the detection of polarization-dependent optical properties like birefringence and dichroism. The visible light source used is a halogen lamp (Amscope), which covers a broad wavelength range at least from 350 nm to 700 nm. The incident power at the sample plane was approximately 1.2 mW for 265 nm illumination and 7.8 mW for visible light. Fig. 4 a) and 4 b) were acquired from the same tissue slice to enable direct comparison.

Furthermore, the polarization signal exhibits a strong correlation with the orientation of porcine fibrous structures. To systematically analyze fiber orientation, the analyzer was uniformly rotated, acquiring 36 images at 10° increments over a full 360° rotation, as illustrated in Fig. 5. The DoLP analysis indicated that the polarization state of collagen fibers significantly changed after passing through the sample, whereas the background glass maintained its initial polarization state. Additionally, the AoP images demonstrated that fibers with varying orientations generated distinct polarization angles, highlighting the potential of polarization-resolved deep ultraviolet microscopy for structural characterization of biological tissues. The polarization sensitivity was demonstrated by comparing selected regions with the background. Clear phase and amplitude variations are observed across different polarization rotation angles shown in Fig. 5(b). Fig. 5(c) demonstrates that porcine collagen tissue exhibits strong polarization sensitivity. Additionally, a cross-sectional plot of pixel intensity versus rotational angle is provided, with angles color-coded from 0° to 360°. The 3D plot reveals the detailed polarization dependence of the fibrous structure from pixel to pixel. This approach allows us to capture much richer information that is usually lost when using point measurements or averaging over an area, as done in Fig. 5(b). The cross-section of the 3D plot illustrates how the intensity varies both across pixels and with rotation angle. This clearly shows the phase shift along the blue line on the fibrous structures indicated in Fig. 5(a). Compared to the green line in Fig. 5(b), which represents an average over an area, the 3D plot provides a more complete picture of the fiber orientation. A similar 3D plot, following a dark yellow line, is shown for a background region. This visualization highlights the contrast between the porcine collagen tissue and the background, revealing that while the background follows a sinusoidal pattern with consistent pixel intensity values, the collagen tissue exhibits distinct polarization-dependent variations.

Fig. 5.

Porcine collagen tissue imaging using DUV polarized microscopy with a FoV of 400 × 180 μm. a) From top to bottom, the calculated S₀, S₁, and S₂, DoLP, and AoP for the captured porcine collagen tissue are displayed. b) The polarization sensitivity was demonstrated by comparing selected regions with the background. Clear phase and amplitude variations of the polarization intensity are observed by changing polarization rotation angles from 0° to 360°. A total of 36 images were captured at 10° increments, covering a full 360° rotation. c) A 3D plot of pixel intensity values along a selected blue line across 360° demonstrates that porcine collagen tissue exhibits strong polarization sensitivity. Additionally, a cross-sectional plot of pixel intensity versus rotational angle is provided, with angles color-coded from 0° to 360°. d) A similar 3D plot, following a dark yellow line, is shown for a background region. This visualization highlights the contrast between the porcine collagen tissue and the background, revealing that while the background follows a sinusoidal pattern with consistent pixel intensity values, the collagen tissue exhibits distinct polarization-dependent variation.

4. Discussion and conclusion

In summary, we developed a polarization-resolved deep ultraviolet microscope for label-free imaging, enhancing nuclear contrast and revealing fiber orientation. The DUV illumination offers approximately twice the spatial resolution compared to visible light and provides superior nuclear contrast without the need for staining. Additional, DUV polarization images provide better contrast compared to visible illumination. We attribute this improvement to the fact that DUV light interacts more strongly with molecular bonds and fiber alignments, making polarization-based contrast more detectable. The DoLP and AoP images demonstrated that collagen exhibits polarization sensitivity at DUV wavelengths. This additional polarization information can aid in assessing tissue status, making disordered fibers more easily detectable.

However, this work still has some limitations. Although the polarization response of the fiber can be observed, its detailed polarization properties require further demonstration. Both dichroism and birefringence may contribute to the observed effects, but the dominant mechanism remains to be confirmed. Collagen’s rod-like triple-helix structure gives rise to both linear and circular optical anisotropy. As a result, collagen exhibits linear birefringence, with a refractive index that is higher along the length of the fiber than across its cross-section, due to its highly ordered molecular packing [29]. Nevertheless, dichroism plays an important role in molecules. Deoxyribonucleic acid (DNA) consists of aromatic base molecules stacked vertically and twisted along a helical axis. The absorption starts from near 300 nm and extends into the far-UV range. Linear dichroism (LD) is commonly used to study the orientation of extrinsic chromophores, including DNA-binding drugs [30]. It also serves as a valuable complementary technique for investigating protein fibers. Studies have shown that LD can reveal the orientation of secondary structural elements, aromatic amino acids, and bound nucleotides within F-actin [31,32]. To better characterize the polarization behavior in the DUV range, rotating the analyzer is necessary. Acquiring rotationally varied crossed and parallel polarization image pairs can enable Fourier analysis to quantify birefringence, determine the birefringent axis orientation, and assess dichroism and optical activity [33].

Additionally, the extinction ratio of our setup is only 0.4, as shown in Fig. 5(b). This is largely due to the Glan-type polarizer we used, which is optimized for collimated laser beams at near-normal incidence. These polarizers rely on birefringence and total internal reflection to separate polarization states but have a limited acceptance angle. Since our DUV LED source emits light over a broad angular range, the off-axis rays likely degrade the performance of polarizers, reducing the effective extinction ratio. Inserting a small aperture to narrow the illumination angle could improve the extinction ratio, but it would also significantly reduce the available light, making it insufficient for imaging. In the future, we plan to incorporate deep ultraviolet-transmissive quarter-wave plates to obtain full Stokes parameters and compute the Mueller matrix, potentially uncovering birefringence and dichroism of tissue samples. Inserting DUV waveplates to enable full Stokes parameter measurement would significantly expand the capability of our system. Currently, our work is limited to linear polarization states, which primarily reveal linear birefringence and dichroism in tissue structures such as collagen fibers. By measuring the circular polarization component, the system could also detect circular birefringence (optical activity) and circular dichroism, which are characteristic of chiral biological structures. This would open new possibilities for probing molecular-level asymmetry in biomolecules such as DNA, proteins, and other chiral assemblies that strongly absorb in the DUV range. For example, circular dichroism spectroscopy in the UV is widely used to study protein secondary structure and nucleic acid conformations. Translating such sensitivity into a spatially resolved imaging modality could provide valuable complementary information to linear polarization measurements, potentially allowing us to map structural anisotropy, chirality, and conformational changes within tissue samples. Although the lack of suitable DUV waveplates currently limits us to partial Stokes measurements, we see the integration of these components as an important direction for future work, enabling more comprehensive polarization analysis and deeper insight into biological structure.

Data availability

Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the corresponding authors (zhihanhong@arizona, rliang@optics.arizona.edu) upon reasonable request.