a. CT imaging acquisition and segmentation
All CT images acquired were full rotation (360°) with 180 projections at a 50 ms settlement time, medium magnification (pixel size, 78.81 µm), and a field of view of 8.07 × 16.11 cm, in a 512 × 512 matrix. Images were acquired at 80 kV, 500 µA with 300 ms exposure time. For in vivo images, a preset beam-hardening correction was applied. Images were acquired of ethanol-water vials and ex vivo and in vivo rat livers. Images were processed in 3D Slicer [56].
The average radiodensity of in vitro ethanol-water samples was computed by segmenting a cylinder within each vial. Ex vivo liver images were segmented by selecting the tissue surrounding the injected ethanol distribution without including surrounding buffer. Ethanol was segmented by interpolating between circles of 5–15 mm on each side of the injected ethanol. Overlapping circles were used when necessary. In vivo images were segmented by selecting the tissue surrounding the injected ethanol without including surrounding tissues (intestines, stomach, or lungs). The same method used for ex vivo images was used to segment the in vivo liver and ethanol distributions. Sample pre- and post-ablation segmentation volumes were similar.
b. Determination of ethanol concentration and quantitation of error from radiodensity measurements of ethanol vials
Solution
|
Concentrations Tested (%)
|
Ethanol-water
|
0, 25, 50, 75, 100 (n=20)
|
Iohexol-ethanol
|
0, 2, 5, 7.5, 10 (n=3)
|
Ethyl cellulose-ethanol
|
0, 3, 6, 8, 10, 12, 15 (n=3)
|
Fluorescein-ethanol
|
0, 0,25, 1, 2.5 (n=3)
|
Solutions were imaged with CT and segmentation was performed as described above. Radiodensity data was converted to ethanol concentration with a linear two-point calibration equation (equation (1)), which corresponded to a five- point calibration. Calibration was performed at the beginning of each imaging session. For in vitro experiments, vials of pure ethanol and water were imaged to serve as 100% and 0% ethanol standards, respectively. In the liver studies, the 0% ethanol standard was the average radiodensity of the pre-ablation liver, since tissue has slightly higher radiodensity than water. The radiodensity difference of the 0% and 100% standards represents the radiodensity range for all possible ethanol concentrations (denominator). The radiodensity difference between the sample and the 0% ethanol standard was calculated (numerator) and divided by the radiodensity range for all possible ethanol concentrations (denominator) to determine the sample ethanol concentration. Equation (1) assumes a linear relationship between ethanol concentration and radiodensity, and that sample thickness does not affect measurement accuracy. Supplementary Fig. S2 illustrates application of this equation.
c. Description of animal work
All animal studies were approved by the Duke University Institutional Animal Care and Use Committee and performed in accordance with guidelines and regulations (Protocol Number A160-18-07). Male Fischer CDF rats (Charles River Laboratories) were used for ex vivo (n=36 lobes, n=10 rats) and in vivo studies (n=12 rats). Male rats were used since anatomical imaging and ethanol concentrations should be insensitive to gender and liver cancer incidence is higher in males than females [1]. Rats had ad libitum food and water access and regular 12-hour light/dark cycles.
d. Ex vivo rat liver studies
Rats were euthanized via isoflurane overdose and bilateral thoracotomy. The liver was immediately excised and stored in Krebs-Ringer bicarbonate buffer (Sigma Aldrich, K4002) on ice until injection (within 1–2 h). Individual lobes were placed in a small plastic container (height, 62.6 mm; diameter, 41.9 mm) for injection. A 27-gauge needle was lowered to the approximate center of the lobe using a holder to prevent lateral motion. Fluid was infused from a 3-mL syringe (BD Medical, Columbus, NE) through 10 cm of rubber tubing (1/4” inner diameter, McMcmaster-Carr, Douglasville, GA) using a syringe pump (NE-1000, New Era, Farmingdale, NY) at a flow rate of 10 mL/h. Prior work demonstrated that 10mL/hr is optimal compared to 0.1, 1, and 100mL/hr[21] 100 µL of fluid was infused based on infusion volumes previously optimized to reduce leakage [22] and the rat liver size. The needle was removed 3 minutes after infusion to allow fluid to dissipate. Non-contrast CT images of the samples were acquired pre- and post-injection with EC-ethanol (0%, 6%, 8%, 10%, 12% or 15%).
e. In vivo rat liver studies
Pre-ablation non-contrast CT images of the rat abdomen were acquired. Rats were maintained with 1.5% isoflurane at 2 L/min during the procedure and a heating pad maintained body temperature. Buprenorphine Sustained-Release (1 mg/kg) was administered subcutaneously as an analgesic. The abdomen was depilated and disinfected three times with 10% povidone-iodine followed by 70% ethanol. A laparotomy was performed by creating an incision with a sterile scalpel through the skin and abdominal wall to expose the left lateral lobe of the liver. A sterile cotton-tipped applicator was used to expose the center of the left lateral liver lobe. Injections were performed as described above for the ex vivo studies. The needle was slowly retracted and a cotton-tipped applicator was used to stop any visible bleeding. The abdominal wall was closed with Reli monofilament sutures (VWR, 89219-212). 1-2 drops of 0.25% bupivacaine were applied along the incision as a local anesthetic. The skin was closed with Coated VICRYLÒ (polyglactin 910) sutures (VWR, 95057-014). Post-injection non-contrast CT images of the liver were acquired.
Animals were monitored post-ablation every 6-8 h for 24 h for: mobility impairment; inflammation/edema; bleeding; respiratory distress; loss in body weight; licking, biting, scratching or shaking of procedure site; hair coat changes; posture; and lethality. Rats were euthanized by isoflurane overdose 24 h after ablation and the liver was immediately excised. The left lateral lobe was cut into three 2×2 cm samples. Samples were placed into Peel-A-WayÒ disposable embedding molds (Polysciences Inc., 18646A-1, Warrington PA), labeled, and covered in optimum cutting temperature (OCT) gel (Sakura Finetek, Torrance, CA). The molds were placed in a metal container of 2-methylbutane (Sigma, 277258) and frozen using liquid nitrogen. Samples were stored in a -80°C freezer.
f. Ethanol distribution volume and radial symmetry of ex vivo and in vivo liver tissues
Maximum intensity projection images were produced from the 3D segmentation by projecting the voxel with the highest estimated ethanol concentration onto a 2D image from top- and side-view perspectives. An 20% ethanol concentration threshold was used because a 10-min exposure of 20% ethanol is cytotoxic [25] and imaging were acquired 10-20 min after injection. Supplementary Fig. S3 illustrates that this threshold excludes regions of naturally low radiodensity in untreated tissue from analysis. Ethanol distribution volume was calculated by converting total number of voxels with ethanol concentration 20% to volume. The degree of asymmetry of the ethanol distribution was quantified by the aspect ratio, as in equation (2)—the radius of gyration over the effective radius—for all voxels with estimated ethanol concentration 20%.
g. Pathologic evaluation of ablative extent
To assess the extent of necrosis, two 7-µm sections were cut serially every 1 mm from frozen samples at -15°C with a cryostat microtome (Microm HM 560, Thermo Fisher Scientific, Waltham, MA). Serial sections were adhered to positively-charged, uncoated glass slides (Thermo Fisher Scientific, 6776214). One slide from each pair was stained with reduced nicotinamide adenine dinucleotide (NADH)-diaphorase, a viability stain which distinguishes viable cells (blue) from necrotic cells (unstained). The slides were covered in Tris buffer (0.05 M, pH 7.6) with 8 mg / 5 mL NADH (Sigma, N8129) and 10 mg / 5 mL nitro blue tetrazolium (Sigma, N6876) and incubated for 15 min at 37°C. Slides were then washed three times with deionized water followed by three exchanges in 30%, 60%, and 90% acetone. The slide was covered in 90% acetone until a purple cloud appeared in the solution. Slides were washed three times with deionized water and allowed to dry. Coverslips (Thermo Fisher Scientific, 12540C) were applied with aqueous mounting medium prepared by mixing 21 mL of deionized water, 4 g of store-grade unflavored gelatin, 25 mL of glycerol (Sigma, G2025), and 0.5 mL of phenol (Sigma, P9346) at low heat.
Sectioning was performed beginning with the sample from the distal end of the liver lobe. The second sample was sectioned and stained using the same procedure. If no necrosis was observed in any sections in the second sample, the third sample was not sectioned or stained; otherwise, the third section was sectioned and stained using the same procedure.
h. Image analysis of pathology specimens from in vivo liver tissues
Slides were digitally scanned at 10x magnification with a Zeiss Axio Imager Z2 upright microscope. The region of interest (ROI), or necrotic area, was determined using a custom MATLAB program. Images were cropped to remove portions that did not contain tissue. First the tissue was segmented by applying an entropy filter to the blue channel of the images and binarizing the result using a user-defined threshold. Small regions (<15,000 connected pixels) were deleted to remove noise. The edges of the regions were eroded using a flat structuring element with a 15 pixel neighborhood, holes in the regions were filled, and the edges of the regions were dilated using the same structuring element. The boundaries of the tissue sample were detected and the area was quantified using the MATLAB function ‘regionprops’. A mask generated from the binary image was used to remove background pixels from the original image.
The resultant image was then used to segment the necrotic area. The blue channel was binarized with a user-defined threshold, and small regions <5,000 pixels were removed. All ROIs except for the five largest were deleted from the image. Regions such as large vasculature which may be detected under the same threshold used for necrosis were manually selected for removal. The boundaries of the necrotic regions were detected and the MATLAB function ‘regionprops’ was used to quantify area. Supplementary Fig. S4 shows representative images each step. Necrotic volume was calculated by multiplying necrotic area by the sectioning step size (1 mm) and taking the sum for all samples for each animal. All images were processed by one user. Adjacent H&E slides were used for confirmation. The semi-automated MATLAB algorithm was compared to gold standard manual segmentation using ImageJ software. Digital images from the sections of two samples (1 per group) were manually segmented. For the 13 images assessed, the MATLAB algorithm estimated an average of 0.0465 cm2 more necrotic area than manual segmentation, with an average absolute scalar difference of 0.0049 cm2. Supplementary Fig. S5 shows the comparative analysis of the necrotic areas determined manually and using the MATLAB algorithm.