Comprehensive Surface Histology of Fresh Resection Margins With Rapid Open-Top Light-Sheet (OTLS) Microscopy

Abstract

Objective: For tumor resections, margin status typically correlates with patient survival but positive margin rates are generally high (up to 45% for head and neck cancer). Frozen section analysis (FSA) is often used to intraoperatively assess the margins of excised tissue, but suffers from severe under-sampling of the actual margin surface, inferior image quality, slow turnaround, and tissue destructiveness.

Methods: Here, we have developed an imaging workflow to generate en face histologic images of freshly excised surgical margin surfaces based on open-top light-sheet (OTLS) microscopy. Key innovations include (1) the ability to generate false-colored H&E-mimicking images of tissue surfaces stained for < 1 min with a single fluorophore, (2) rapid OTLS surface imaging at a rate of 15 min/cm² followed by real-time post-processing of datasets within RAM at a rate of 5 min/cm², and (3) rapid digital surface extraction to account for topological irregularities at the tissue surface.

Results: In addition to the performance metrics listed above, we show that the image quality generated by our rapid surface-histology method approaches that of gold-standard archival histology.

Conclusion: OTLS microscopy has the feasibility to provide intraoperative guidance of surgical oncology procedures.

Significance: The reported methods can potentially improve tumor-resection procedures, thereby improving patient outcomes and quality of life.

Fig. 1. Comparison of standard frozen-section analysis vs. open-top light-sheet (OTLS) microscopy for surgical guidance. (a) Schematic depicting the clinical workflow of FSA vs. OTLS microscopy for assessing margin status during surgery. OTLS microscopy is non-destructive of tissue, can potentially be implemented within the operating room, and generates H&E-mimicking images for pathologist review. (b) FSA involves examining 2D cross-sectional images at a limited number of locations along the inked margin, such that a region with a small positive margin could be missed (due to undersampling of the margin). OTLS microscopy enables comprehensive sampling of the entire margin surface.

Fig. 2. Comparison of prior image-processing workflow vs. fast RAM-based workflow. (a) A series of thin 3D volumes of the specimen surface are generated by stage-scanning the sample through the light-sheet in three dimensions. In our previous workflow, the image tiles (3D volumetric bricks) were streamed to the hard drive and saved as an HDF5 file during tissue scanning. A fusion process was implemented to register, stitch, and blend individual image tiles into a continuous and seamless 3D volume before post-processing. (b) With the fast RAM-based workflow, the image tiles are directly stored and stitched in RAM during imaging. The dataset is available for post-processing in RAM immediately after image acquisition. (c) Without multiple I/O bottlenecks and the time-consuming image-fusion process, the optimized workflow is accelerated by ~ 60X compared to our previous workflow. The time required for each processing step on a 1.1 cm x 1cm x 170 μm volume is shown for the previous workflow (top) and optimized workflow (bottom).

Fig. 3. Digital surface extraction of irregular tissue surfaces. (a) Imaging irregular tissue surfaces at a fixed depth plane (left) can lead to incomplete coverage and overly deep regions where image quality is degraded due to light scattering. Our algorithm (right) digitally extracts the irregular tissue surface to provide comprehensive 2D visualization of the surface. (b) Steps for the smooth manifold extraction algorithm. A depth map of the tissue surface is first estimated based on the maximum fluorescence intensity positions in the 3D volume. Then, the high-resolution depth map is smoothed out to achieve a continuous “in-focus” surface. (c) A freshly excised specimen of human tongue (1.1 × 0.8 × 0.3 cm) was rapidly stained with Acridine Orange (AO) and imaged with the OTLS microscope. When surface extraction is not performed (upper right), regions of incomplete imaging (orange arrowheads) and overly deep imaging (blue arrowheads) are seen. In contrast, our modified smooth manifold extraction (mSME) method is able to extract a more complete surface (lower left) from the 3D volume. (d) ROIs (orange box in (c)) after surface extraction are shown at higher magnification. EDF leads to obvious discontinuities in the image, which does not represent a smooth surface manifold. Fast SME is unable to extract surface regions with weak signal. Our mSME method extracts a continuous “in-focus” surface. (e) Depth profile of the tissue surface along the yellow line in (c) reveals significant topological irregularities (up to 150 μm in extent).

Fig. 4. H&E false coloring of a single-channel fluorescence image with AO staining. (a) Steps for generating false-colored H&E-mimicking images of the tissue margin surface with a single fluorescent agent. The hematoxylin channel containing bright and sharp nuclei, and the eosin channel with enhanced cytoplasm signals, are separated from the raw image by a frequency- and intensity-based method. Then, the two channels are digitally stained to mimic the appearance of standard H&E histology through an open-source false-coloring code. (b) En face visualization of a freshly excised specimen of human tongue. Standard fluorescence intensity visualization is shown on the left, with high-magnification views of three ROIs. On the right are H&E-like visualizations of the same ROIs.

Fig. 5. Image atlas comparing OTLS false-colored images with standard H&E histology. (a) A large margin surface on a freshly excised human tongue specimen (1.1 × 0.8 × 0.3 cm) is imaged by OTLS (left) and compared with slide-based H&E histology of the same tissue (right). Squamous cell carcinoma is present at the bottom left and bottom right of the OTLS surface image, indicating a positive margin. (b) – (e) Microarchitectural features are seen in high-magnification views, where OTLS images are marked with solid borders and corresponding H&E histology images have dashed borders. (b) Benign tissue with adipocytes (blue arrowhead) and skeletal muscle fibers (orange arrowhead). Squamous cell carcinoma found at the surface is characterized by (c) an irregular interface with the stroma and (d) prominent nucleoli (green arrowhead). (e) Positive margin at the bottom left is characterized by an irregular interface (yellow arrowhead), prominent nucleoli (green arrowhead), and variability of nuclear size and shape. Various head & neck specimens with squamous cell carcinoma are displayed, including (f) a lymph node metastasis from a tonsil tumor showing tumor nests (blue arrowhead) and (g) a lymph node metastasis from a hypopharynx tumor showing islands of tumors. Irregularly crowded and enlarged nuclei (yellow arrowhead) can be identified from the high-magnification view in (g). Scale bars are 1 mm for (a), 50 μm for (d), and 150 μm for all other images.