Mechanically sheared axially swept light-sheet microscopy

Abstract

We present a mechanically sheared image acquisition format for upright and open-top light-sheet microscopes that automatically places data in its proper spatial context. This approach, which reduces computational post-processing and eliminates unnecessary interpolation or duplication of the data, is demonstrated on an upright variant of axially swept light-sheet microscopy (ASLM) that achieves a field of view, measuring 774 × 435 microns, that is 3.2-fold larger than previous models and a raw and isotropic resolution of ∼460 nm. Combined, we demonstrate the power of this approach by imaging sub-diffraction beads, cleared biological tissues, and expanded specimens.

1. Introduction

The study of biological processes within intact tissues has gained paramount importance in modern biology and pathology. Advances in sample preparation methods, such as tissue clearing and expansion microscopy [1,2], alongside improvements in optical imaging systems, now make it feasible to investigate sub-cellular biological processes in their native tissue environments [3,4]. This approach is especially valuable in pathology, where volumetric data can reveal unprecedented insights into rare events, like the identification of isolated metastatic breast cancer cells in lymph nodes [5]. Likewise, in cancer biology, the patterns of cancer dissemination and the complexities of the tumor microenvironment are most evident when observed in its three-dimensional entirety, which offers a more comprehensive view of cellular heterogeneity, immune infiltration, and architectural alterations in adjacent tissues [6].

Light-sheet fluorescence microscopy (LSFM) has emerged as a powerful tool for volumetric imaging, offering fast image acquisition speeds and minimal photobleaching. In LSFM, a 3D volume is acquired by illuminating the specimen from the side, and serially imaging adjacent 2D sections within the specimen with a scientific camera. In classical LSFM geometries, one synchronously sweeps the illumination beam and the detection objective, or the specimen, along the optical detection axis (see Supplement 1 ^{(2.1MB, pdf)} Fig. S1(a)). Alternatively, as is common in lattice light-sheet microscopy [7] and open-top [8] LSFMs, the specimen is scanned obliquely relative to the illumination and detection axes (e.g., in the S direction, see Supplement 1 ^{(2.1MB, pdf)} Fig. S1(b)). In this geometry, the thickness of the specimen is limited by the mechanical working distance of the illumination and detection objectives (see dashed lines in Fig. S1(b)). For objectives with sufficiently large working distances, both the scan direction and the direction orthogonal to it (e.g., X) are limited only by the travel range of the stages employed (Fig. 1). As such, when combined with long working distance objectives, oblique sample scanning presents a significant advantage by enabling practically unlimited imaging in two dimensions [8–11].

Fig. 1.

Optical and mechanical shearing of imaging data. (a) In a conventional oblique scan format, the sample is scanned in the S-direction and images are acquired at each adjacent plane in a staggered format (top) but saved in a continuous format (middle). Thus, data must be computationally sheared (bottom) to place it back into its proper spatial context, which introduces empty space above and below the shear axis (see black outline around sheared image). (b) In a mechanically sheared oblique scan format, the sample is simultaneously scanned in S and Z’, thereby placing it in its proper spatial context during image acquisition.

While oblique scanning offers benefits for samples that have large lateral extents—often seen in tissue-derived specimens—the data acquired requires computational shearing. This process leads to data duplication and increases the size of the images in the dataset. Consequently, sheared datasets become significantly larger than the original, raw data. And even with performant CPU and GPU-based software [12], computational shearing of data introduces processing delays and implementation hurdles. To address this challenge, we introduce a multi-axis, or mechanically sheared, image acquisition scheme that eliminates the need for computationally intensive post-processing of the data. This technique is inspired by a recent advance in light-sheet microscopy whereby a high-speed mirror galvanometer within the detection path of an LSFM optically sheared the data in real-time by scanning an otherwise stationary image across the camera [13]. While effective for imaging modalities that produce high-aspect-ratio images, such as those in oblique plane microscopes (OPM) [14], it is less applicable to ASLM, where the entire camera sensor is utilized. Instead, our method mechanically shears the data by simultaneously scanning the sample along both the S and Z’ axes, placing the data correctly in its spatial context from the outset.

2. Materials and methods

2.1. Tissue procurement and preparation

Animal care was conducted in strict compliance with Institutional Animal Care and Use Committee (IACUC) approved protocols at the University of Texas Southwestern Medical Center. For human specimens, no direct interaction or intervention with human subjects was made for biospecimen collection, and all human tissues were provided and deidentified by someone uninvolved in the study. Detailed methods describing the labeling and clearing of tissues are in the supplemental document.

2.2. Imaging system and control

The ASLM presented here is constructed in a dual-inverted selective-plane illumination microscopy-like configuration on a two-tiered vibration isolation system (see Supplement 1 ^{(2.1MB, pdf)} Fig. S2 for a detailed optical layout) [10,15]. The bottom tier consists of a 36” x 72” x 18” optical table (Performance Series, TMC) with tuned vibration isolators (UltraDamp Series, TMC) that provide greater dampening at low frequencies. The second tier is assembled on top of the optical table and includes 14-inch vibration-isolating posts (DP14A, Thorlabs) that support a damped 24” x 48” x 4.3” optical breadboard (PG-24-4-ML, Newport). This upper tier serves as the platform for all illumination and detection optics. The specimen stage (FTP-2000, ASI), which is used for sample positioning in Z’, X, and S, is directly mounted on the larger, bottom tier. A detailed summary of the microscope’s construction, and a schematic, is provided in the Supplement 1 ^{(2.1MB, pdf)} . Our microscope uses navigate (https://github.com/TheDeanLab/navigate) acquisition software to perform all imaging-related tasks [16]. Stage and filter-wheel operation is performed via serial communication with a Tiger Controller (TG8-BASIC, ASI) equipped with TGCOM, TGFW, and 2x TGDCM2 control cards. Analog and digital tasks are performed with a data acquisition chassis (PXIe-1073, NI) equipped with multifunction input/output and analog output cards (PXIe-6259 and PXI-6733, NI). The acquisition computer (ProEdge SX6800, Colfax International) runs on Microsoft Windows 10 Pro and is powered by an Intel Xeon Silver 4215R CPU @ 3.20 GHz with 96GB of RAM.

2.3. Optical simulations

Optical simulations were performed with Zemax OpticStudio, using manufacturer-provided files or lens specifications. Components were accurately positioned within the virtual model and evaluated with Zemax's sequential mode.

3. Results

3.1. Optical concept

Figure 1 illustrates the fundamental concept of image shearing; when the laser or sample scan axis is not coincident with the optical axes (e.g., as is the case in OPM and diSPIM-like systems), computational shearing of the data is necessary to place it in its proper spatial context (Fig. 1(a)). Shearing of the data is performed in the Fourier domain or with an affine transform, both of which are computationally expensive and laterally shift the data (e.g., in the y-direction) by a factor that depends upon the z-position within the image stack and the angle (α) between the optical and mechanical axes [7,14,17]. By laterally shifting the data in a depth-dependent fashion, empty space is introduced into the image canvas, and the overall image size increases.

Our approach, which we term mechanical shearing, simplifies the imaging workflow by integrating the correction process directly into the data collection stage (Fig. 1(b)). In mechanical shearing, the specimen is simultaneously scanned both vertically and laterally (see more details in Supplement 1 ^{(2.1MB, pdf)} Note 1, Fig. S3), ensuring that each slice is acquired in its proper orientation from the outset. This method maintains the benefits associated with oblique scanning (e.g., interrogation of thin specimens with large widths and lengths), while avoiding interpolation and thus providing superior resolution. As a result, the imaging process becomes more efficient, reducing both the time and computational resources required to achieve accurately aligned volumetric datasets.

3.2. Optical characterization

To demonstrate the advantages of mechanical shearing for tissue imaging, we developed a microscope in an upright orientation that simultaneously provides a large field of view and high optical resolution. LSFMs face a trade-off between axial resolution and field of view, a limitation observed in LSFMs that adopt both Gaussian and Bessel-Gauss illumination schemes (e.g., lattice light-sheet microscopy [18]). Two notable exceptions to this limitation include dual-view selective plane illumination microscopy (e.g., diSPIM), and Axially Swept Light-Sheet Microscopy (ASLM). In the former, the sample is imaged from orthogonal perspectives, and the data is registered and fused via an iterative deconvolution scheme [15]. In contrast, for ASLM, a diffraction-limited beam is axially scanned synchronously with a camera's rolling shutter, enabling high-resolution imaging over a large field of view [19]. Data generated has an isotropic resolution and can be viewed in its raw format from any spatial dimension. Thus, we sought to combine the strengths of ASLM and mechanical shearing, making it possible to achieve isotropic imaging in large tissue contexts in an upright microscope geometry without necessitating data manipulation.

To experimentally verify the performance of mechanically sheared data acquired with our upright ASLM, we evaluated the point spread function (PSF) using 200 nm beads embedded in 1% agarose, as depicted in Fig. 2(a–e). Figure 2(a) illustrates beads covering the entire camera chip, with zoom-in sections highlighted in Fig. 2(b). An axial view of the beads is presented in Fig. 2(c), with three sub-regions displayed in Fig. 2(d). Figure 2(e) reveals the PSF of a single 200 nm bead in all three dimensions. To evaluate the spatial uniformity of the resolution, bead images spanning the entire field of view (774 µm width x 435 µm height) were evenly divided into nine sections. Computer vision routines were utilized to analyze the lateral (XY) Full-Width Half-Maximum (FWHM) of beads within each section (see Supplement 1 ^{(2.1MB, pdf)} for details). The mean resolution in each section is displayed as a heatmap (Fig. 2(g)), and a slight decrease is observed in the lateral resolution at the image edges compared to the center. To quantitatively assess the isotropy of resolution, data from 200 nm beads spanning the entire field of view were localized and subjected to a 3D Gaussian fit. The resulting resolution values for all beads in each dimension are plotted in a histogram (Fig. 2(f)) and fit as a mixture of three Gaussian populations. We interpret the Gaussian population with the smallest FWHM as representing single, isolated beads, while Gaussian populations with larger FWHMs are indicative of clusters comprising two or more beads. For each dimension, the largest component of the mixture model was the lower resolution feature, with means of 460, 460, and 483 nm, in X (n = 713), Y (n = 713), and Z (n = 713), respectively. These resolutions are consistent with previously published variants of ASLM, despite a ∼3-fold larger field of view and the mechanically sheared image acquisition format.

Fig. 2.

Analysis of 200 nm beads. (a) Displays the XY maximum intensity projection of 200 nm beads in agarose spanning a 20 µm range in the Z dimension. (b) Shows zoomed-in regions of the image from panel a, arranged from left to right. (c) Depicts the XZ maximum intensity projection of 200 nm beads in agarose across a 20 µm range in the Z dimension. (d) Exhibits zoomed-in regions of the image from panel c. (e) Illustrates the maximum intensity projection of a single 200 nm bead in the XY, XZ, and YZ dimensions. (f) Presents histograms of the full-width half maximum (FWHM) of 200 nm beads in the X, Y, and Z dimensions. The number of analyzed beads is 713. (g) Shows a heatmap of the lateral full-width half maximum (FWHM) of 200 nm beads across a 435 µm x 774 µm camera chip. Scale bars: a, c = 100 µm; b, d = 10 µm; e = 1 µm.

3.3. Quantitative comparison of computationally and mechanically sheared data

Next, we assessed whether computational shearing, which involves interpolation, exerts a discernible effect on the spatial resolution of a microscope. To evaluate this, 200 nm beads were prepared in agarose, and imaged under oblique and mechanically sheared formats. All other imaging variables, including exposure time, z-step size, lateral pixel size, and ASLM scan parameters (e.g., remote focusing amplitude and offset), remained unchanged. Data acquired in the classical oblique scanning format were computationally sheared. Beads from both datasets were evaluated identically using a 3D Gaussian fit, and the FWHMs displayed as a violin plot (Fig. 3). Interestingly, the computationally sheared dataset exhibited a slight but statistically significant improvement in lateral resolution (e.g., in X and Y), likely arising from subtle differences in optical alignment or the shearing algorithm. However, a much larger and statistically significant reduction in resolution was observed in the axial dimension for the computationally sheared data (see Supplement 1 ^{(2.1MB, pdf)} Table S1). These findings highlight the critical role of interpolation, which effectively functions as a low-pass filter in frequency space, in influencing image resolution.

Fig. 3.

Analysis of 200 nm beads for mechanically and computationally sheared data sets. The mean resolution for mechanically sheared data was 491 nm (X), 477 nm (Y), and 632 nm (Z), whereas for computationally sheared data, it was 473 nm (X), 468 nm (Y), and 723 nm (Z), respectively. Statistical significance was evaluated with a Mann-Whitney U test, which makes no assumptions about the underlying population statistics. P-values were 0.003, < 0.0001, and <0.0001 in X, Y, and Z, respectively (see Supplement 1 ^{(2.1MB, pdf)} Table S1). All statistical tests were performed with the SciPy toolkit [20].

3.4. Overhead associated with computational shearing

Computational shearing of large datasets is associated with significant numerical overhead. And importantly, the sheared data is larger than the input data, which creates additional storage challenges. To evaluate the computational benefits of mechanical versus computational shearing, we conducted benchmarks across a spectrum of CPU and GPU-accelerated computational shearing packages [12,14,21]. The outcomes of these benchmarks, detailed in Table 1, include the duration required to shear the data and the dimensions of the sheared output. As anticipated, our analysis revealed that CPU-based methods are only constrained by the available system RAM. Despite this limitation, these methods exhibited considerable processing times (e.g., > 20s), even when utilizing advanced file format back-ends designed for distributed, chunked data processing [21]. On the other hand, GPU-accelerated approaches demonstrated a marked improvement in processing speed. However, these methods face limitations due to the GPU’s RAM capacity, typically capped at 24 or 32 GB. This restriction rendered GPU-based methods non-ideal for processing larger image stacks (e.g., those with greater than 2500 slices), despite their evident acceleration capabilities.

Table 1. Comparison of computational shearing software packages.

FOV (pixels)	Size (GB)				Processing Time (s)
	Original	LLSM5D	Python	CLIJ	LLSM5D	Python	CLIJ
1024 × 1024 × 500	1.00	1.90	1.30	1.30	7.25±1.73	144.60±3.66	0.97±0.08
1024 × 1024 × 1024	2.00	3.00	3.50	3.50	19.70±0.96	307.00±8.51	1.87±0.07
1024 × 1024 × 1300	2.50	3.50	5.00	5.00	28.56±0.67	572.70±l32.76	2.51±0.12
1024 × 1024 × 1500	2.90	3.90	6.10	6.10	31.59±3.27	776.00±30.32	2.86±0.12
1024 × 1024 × 4295	3.30	4.30	7.50	7.50	36.69±0.53	836.80±98.14	3.25±0.48
1024 × 1024 × 2000	3.90	4.80	9.60	9.60	51.34±0.06	1328.50±25.19	3.92±0.88
1024 × 1024 × 2500	4.90	5.80	13.80	N/A	85.46±4.21	1935±97.66	N/A
1024 × 1024 × 3000	5.90	6.70	18.70	N/A	124.84±9.99	2600.50±226.40	N/A
1024 × 1024 × 3500	6.80	7.70	24.40	N/A	164.98±0.46	3143.75±355.51	N/A
1024 × 1024 × 4295	8.40	9.20	34.80	N/A	188.59±13.68	5964.00±645.80	N/A

3.5. Cleared tissue imaging

Next, we sought to demonstrate the advantages of mechanical shearing on cleared tissue specimens. Figure 4 displays a human kidney section that was stained with FLARE [22] and cleared with Benzyl Alcohol Benzyl Benzoate (BABB) [23]. Given our large field of view, an entire human nephron could be captured in a single acquisition (Fig. 4(a)). Additionally, a zoomed-in section (Fig. 4(b) & 4(c)) reveals detailed features within a glomerulus such as individual erythrocytes with their canonical biconcave disc morphology.

Fig. 4.

BABB cleared human kidney imaged with mechanical shearing. Specimen was labeled with FLARE [22], and carbohydrates are shown as blue, and proteins are shown as red. (a) Maximum intensity projection of a human nephron. (b) Zoom in a single slice of the region highlighted in image (a). Glomerulus and red blood cells in 3 different dimensions. (c) A volume rendering of glomerulus and red blood cells from (b). Scale bars: a = 100 µm; b, c = 20 µm.

3.6. Expanded tissue imaging

Expansion microscopy represents a robust approach for imaging biological specimens at sub-diffraction scales [2,24]. However, unless reinforced with a secondary polymer [25,26], expanded tissues are mechanically fragile and thus difficult to image when mounted vertically in a light-sheet microscope. Placing the expanded specimen on a horizontal surface, where it rests under its own weight, avoids the need for secondary embedding of the specimen, simplifying imaging. To demonstrate the advantages of our mechanically sheared acquisition format for imaging expanded samples, we imaged mouse liver sections. Figure 5(a & b) illustrates the expansion of mouse liver tissue following the protein retention expansion microscopy protocol. In the magnified section depicted in Fig. 5(c & d), the morphology of cancer nuclei is revealed with remarkable detail, including nucleoli. Non-muscle myosin 2A staining, as demonstrated in Fig. 5(e & f), highlights clusters of metastatic cells. Figure 5(g, h, and i) showcase mouse liver tissue stained to highlight nuclei and collagen I, a key extracellular matrix component.

Fig. 5.

Protein retention expansion microscopy of mouse liver tissues. (a) A three-dimensional volume rendering showcasing mechanically sheared ASLM images of expanded mouse liver tissue exhibiting melanoma micro-metastases. The rendering displays a volume measuring 774.14 × 418.66 × 100 µm. (b) Orthogonal planes from a 287.28 × 261.07 × 100 µm isotropic volume, provide a detailed view of the micro-metastasis. The imaging includes gray for nuclear structures, green for Myosin IIa, and magenta for amines. (c) A 3D projection of the nuclear channel, oriented at a 45-degree angle. (d) A focused view on the melanoma micro-metastasis area of c. (e) Visualization of nuclei with surrounding Myosin IIa signaling in the micro-metastasis region. (f) A merged view incorporating all three channels to illustrate the micro-metastasis area. (g) Volume rendering of M-shearing ALSM images of healthy mouse liver tissue, demonstrating: Gray for nuclear structures and green for Collagen I. (h) A maximum intensity projection offering high-resolution imaging of the sample. (i) Orthogonal planes from the region indicated in h, providing an enhanced view. All images are accompanied by a scale bar measuring 100 µm. The expansion factor was ∼ 4.5 for all samples.

Owing to the limited working distance of high numerical aperture objectives, chemically expanded hydrogels are often difficult to image [27]. Oblique scan geometries offer a unique advantage in this context, as they provide essentially unlimited travel along two of the three dimensions, and therefore are capable of accommodating samples that span 10’s of mm laterally. To demonstrate the advantages of such an approach, we evaluated a large human colon specimen which measured ∼28 × 20 × 0.25 mm after chemical expansion. Figure 6(a) shows a low-resolution overview of the tissue (see more details in Supplement 1 ^{(2.1MB, pdf)} ). Local high-resolution imaging as far apart as 25 mm could be performed throughout the specimen, enabling evaluation of distinct tissue architectures with sub-diffraction resolution (Fig. 6(b-g)).

Fig. 6.

Imaging of expanded, ∼28 × 20 × 0.25 mm human colon specimen. (a) Low-resolution overview image of FLARE-stained and expanded tissue specimen. Image acquired on a widefield microscope at 10X magnification and transferred to the ASLM system. (b-g) Local high-resolution images acquired from regions shown in (a), presented as single 2D cross-section in X and Y. Scale bar: a = 5 mm; b-g = 100 µm.

To further demonstrate the advantages of mechanical shearing, we imaged a ∼2.8 × 3.5 × 0.2 mm volume of a separate chemically expanded human colon specimen. This tissue was captured in a tiling format and subsequently stitched using BigStitcher [28]. Despite a 4X down sampling of the stitched data, fine vascular structures remained distinguishable (Fig. 7(a)). Selected regions from Fig. 7(a) are presented at their full resolution in Fig. 7(b, c, and d). Ortho-slices in XY, XZ, and YZ highlight the isotropic resolution provided by ASLM.

Fig. 7.

Large volume imaging of expanded human colon specimen. (a) A ∼2.8 × 3.5 × 0.2 mm volume imaged in a mechanically sheared and tiled format. Specimen was stained using FLARE and is presented as a slice after 4X down sampling. (b-c) High-resolution images of sub-regions from (a). (d) Ortho-slice of region shown in (a). Scale bar: a = 200 µm; b, c, d = 100 µm.

4. Discussion

Here, we developed an easy-to-adopt technique termed mechanical shearing that circumvents the need for computationally expensive post-processing of the data. Specifically, the oblique stage geometry permits evaluation of samples with large lateral extents such as clinical specimens [29] and expanded tissues [27]. Once a region of interest is identified, image acquisition proceeds by capturing images after stepping the specimen in both the vertical and lateral dimensions simultaneously. This ensures an accurate spatial representation of the specimen, thereby avoiding computational shearing, improving resolution, and greatly streamlining the imaging workflow.

To demonstrate the practical benefits of mechanical shearing, particularly for tissue imaging, we constructed an ASLM in an upright, diSPIM-like configuration [10,15]. To ensure our imaging system's compatibility with various refractive index solvents and maximize its field of view, we equipped it with high NA multi-immersion objectives and a large format, 12-megapixel CMOS camera. Attempts to maximize the field of view, through integration of a highly corrected 200 mm focal length tube lens resulted in deleterious aberrations at the periphery of the image (data not shown). Thus, guided by Zemax simulations (see Supplement 1 ^{(2.1MB, pdf)} Fig. S4), we opted for a simple yet effective detection path that included only a 300 mm achromatic doublet. The performance of our system was validated by measuring the resolution with 200 nm beads embedded in agarose. Importantly, in aqueous solutions, we achieved a spatially uniform and isotropic resolution of ∼460 nm throughout a field of view of 774.14 × 435.46 microns, which is ∼3-fold larger than previous variants [3,4]. At higher refractive indices, such as BABB, a resolution of ∼330 nm is anticipated.

Owing to the diSPIM-like geometry, the specimen’s thickness is constrained by the extent to which the objectives’ working distances surpass their physical surfaces, with both objectives at 45 degrees from horizontal and converging on the same focal spot (see Supplement 1 ^{(2.1MB, pdf)} Fig. S5). For NA 0.7 multi-immersion objectives, this amounts to a maximum specimen thickness of 2 mm. However, the other dimensions are limited only by stage travel (here, 120 × 75 mm). While we performed mechanical shearing with stepper motors, an improved configuration would also include a 3D piezo. Such a combination would combine the strengths of large travel range stages with the speed of a piezo and also enable multi-angle projection imaging [13]. Likewise, more complex multi-dimensional stage scanning mechanisms could be used to intelligently adapt the illumination to the contours of the specimen [30]. This adaptability is particularly advantageous for imaging complex biological structures, such as the intricate networks of neuronal tissues or the detailed architecture of vascular systems, where traditional imaging methods may fall short due to their inability to accommodate their unique topographical features. While imaging sensors may be rectangular, the resultant image volumes need not be cuboidal.

We applied mechanical shearing to beads, BABB-cleared mouse and human tissues, and chemically expanded mouse and human tissues. Mechanically sheared beads exhibited enhanced axial resolution compared to those subjected to computational shearing. This improvement is likely due to the interpolation inherent in computational shearing, which functions as a low-pass filter in frequency space. Additionally, we performed targeted local imaging in an expanded human colon specimen measuring approximately 28 × 20 mm—a size that poses challenges in traditional light-sheet microscopy setups. Mechanical shearing allowed us to perform high-resolution imaging on any portion of the specimen without the need for dissection into smaller segments. We also demonstrated high-resolution imaging in a tiled format, on a substantial volume (∼2.8 × 3.5 × 0.2 mm) using mechanical shearing, followed by stitching of the acquired data. Due to the imaging configuration, stitched image volumes exhibited a sawtooth-like shape. If one wishes to image the full thickness of a biological specimen, the final image shape must be accounted for in the preparation of the tissue (see Supplement 1 ^{(2.1MB, pdf)} Fig. S6).

In conclusion, by leveraging mechanical shearing in conjunction with a diSPIM-like setup, our approach not only facilitates the exploration of large tissue expanses with enhanced resolution but also significantly reduces the time and computational resources required for post-acquisition processing. Despite the apparent simplicity of multi-dimensional sample scanning (e.g., vector addition), it has surprisingly not been utilized in light-sheet microscopy, as researchers continue to favor more computationally intensive approaches. This methodology eliminates such computational obstacles, thereby reducing the barriers to entry for diverse scientists aiming to understand the molecular origins of tissue function in fields ranging from developmental biology to pathology.

Supporting information

Supplement 1. Supplemental Document.

https://doi.org/10.6084/m9.figshare.26380969

Funding

National Institutes of Health10.13039/100000002 (RM1GM145399, U54CA268072); Narodowe Centrum Nauki10.13039/501100004281 (2020/37/B/ST6/01959); University of Texas Southwestern Medical Center10.13039/100007914 , Simmons Comprehensive Cancer Center Translational Seed Grant and UT Southwestern President's Research Council.

Disclosures

K.M.D. declares that he holds a patent for ASLM that is currently licensed by Intelligent Imaging Innovations, Inc. and subsequently sub-licensed by Life Canvas Technologies. However, all authors affirm that they do not have any investment interests or financial stakes in either of these companies. K.M.D. has an investment interest in Discovery Imaging Systems, LLC.

Data availability

Data underlying the results presented in this paper may be obtained from the authors upon request.

Supplemental document

See Supplement 1 ^{(2.1MB, pdf)} for supporting content.