2.1

Thermo-spectral tissue-mimicking phantom development

Our phantoms were developed by iteratively modifying the synthesis method detailed by Negussie and Mikhail et al.^12,13 in order to match the x-ray attenuation curve to that of human liver tissue calculated from well-accepted elemental composition and physical density values¹⁴. These phantoms are particularly useful for thermal therapy experiments since they present comparable thermal properties to those of human tissue. Briefly, 287.5 ml of deionized water in a 1000 ml flask were degassed by purging N₂ for 15 minutes. After degassing, 202.5 ml of 40% (w/v) acrylamide/bis-acrylamide solution was added to the degassed water to achieve 490 ml solution of 16.5% (w/v) acrylamide/bis-acrylamide. While stirring, 6 grams of calcium chloride dissolved in 10 ml of deionized water was added to the acrylamide solution. A single gram of ammonium persulfate in 2 ml of deionized water and a single milliliter of N,N,N’,N’-tetramethylethylenediamine were subsequently added to the solution. After stirring for additional 15 seconds, the final solution was immediately transferred to a 475 ml plastic jar.

2.2

Physical density spectral map generation

A complete description of the development of our spectral physical density model, its optimization, and its verification on a tissue characterization phantom (Gammex Model 467, Sun Nuclear) was provided before¹¹. Briefly, 70 keV VMI and Z_eff voxel values, which are clinically available on any spectral CT platform, are converted into physical density values through a parametrized Alvarez-Macovski model¹⁵.

2.3

Non-invasive mass measurements

Ex vivo bovine muscle physical density quantifications were evaluated to determine the effect of acquisition parameters on the resulting accuracy. The specimen was weighed with a precision balance (Fisher Scientific Education Precision Balance, Fisher Scientific) to provide ground-truth mass values. Next, the sample was placed on a rectangular block of polyfoam within the 20 cm bore a multi-energy CT phantom (MECT, Sun Nuclear) and scanned with a spectral detector dual-energy CT (IQon spectral scanner, Philips Healthcare) (Figure 3A).

Figure 3:

Experimental setup for two potential applications of physical density quantifications with spectral CT. An ex vivo muscle was (A) scanned within a multi-energy CT phantom on a spectral detector dual-energy CT to investigate non-invasive mass measurements, and (B) subjected to a range of temperatures to evaluate the corresponding effect on physical density.

Scans were repeated three times at a tube voltage of 120 kVp for each set of collimations {16x0.625, 64x0.625 mm}, dose levels {15.2, 30.3, 45.5 mGy}, and acquisition mode {axial, helical} combinations. No helical scan was acquired with 16x0.625 mm at 45.5 mGy due to tube output limitations. Images were reconstructed with a clinical standard body kernel, a field of view of 350 mm, and at a slice thickness of 2.5 mm with 2.5 mm slice intervals. 70 keV VMIs and Z_eff maps were reconstructed from every acquisition and corresponding physical density maps were generated using our spectral physical density model. Regions of interest (ROI) with a diameter of 13.6 mm were positioned in the center of the sample on four consecutive image slices to assess physical density quantification and their dependence on scan parameters. Finally, the sample was weighed after scanning to account for any losses in blood or minor changes in temperature.

To calculate the total mass of the specimen, physical density values were summed from non-air voxels, i.e., 70 keV VMI values larger than -950 HU, and multiplied by the voxel volume (0.68 x 0.68 x 2.5 mm³). Calculated mass was compared to the average of the two weights, pre- and post-scanning, to evaluate its accuracy with different scanning parameters.

2.4

Non-invasive temperature monitoring

To assess the correlation between changes in temperature and changes in physical density, relative to the physical density values at baseline temperature, optical fiber temperature probes were inserted into the same sample used in the non-invasive mass measurement experiment, as well as three of our dedicated thermo-spectral tissue-mimicking phantoms, using 13-gauge medical trocars to continuously record local internal temperatures (Figure 3B). The sample or the phantom was placed in a plastic container, and pre-heated water was poured in to completely submerge it, consequently subjecting it to a wide range of temperatures. After the sample, or phantom, reached an equilibrium temperature, ice was added to cool the water. During heating and cooling, scans were performed approximately every minute with a spectral detector dual-energy CT at a tube voltage 120 kVp, a 16x0.625 mm collimation, a revolution time of 0.75 seconds, and three different radiation dose levels {15.2, 30.3, 45.5 mGy}. For each scan, physical density maps were generated from 70 keV VMI and Z_eff spectral results using the same reconstruction parameters detailed in the section above.

The locations of optical fiber temperature probes were determined by thresholding 70 keV VMIs at 90 HU. 4.1 mm diameter ROIs were placed adjacent to the tip of the optical fiber to measure physical density. To elucidate thermal volumetric expansion, physical density values were normalized by dividing the last temporal physical density value with the physical density at a given timepoint. Similarly, the change in temperature was determined relative to the last temporal scan. Linear regressions were fit to the data, where the slopes were associated with the thermal volumetric expansion coefficient. R-values were determined to characterize the correlation between normalized physical density and temperature change.

3.1

Thermo-spectral tissue-mimicking phantom development

Attenuation curves that were measured on multiple VMIs, at energies between 40 and 200 keV, with increments of 10 keV, are presented in Figure 4. The curves present the iterative developmental process. It enabled us to achieve a maximum error of 4 HU compared to human liver tissue across the entire energy range (Iteration #0 is the scale-down formulation from Negussie and Mikhail et al.). Since human livers differ in their elemental compositions, e.g., different fat contents, we deduce that these attenuation errors were sufficiently small for the intended purposes of these phantoms, i.e., the development and testing of spectral CT thermometry approaches.

Figure 4:

Iterative development process of a thermo-spectral tissue-mimicking phantom, achieving errors below 4 HU across all x-ray energies.

3.2

Non-invasive mass measurements

Spectral physical density quantifications from scans at different doses, collimations, and axial/helical scans revealed that with dose matched scans and different collimations, physical density decreased approximately by 0.003 g/ml with increased collimation. In addition, we observed no effect of dose and axial/helical scans on physical density quantifications.

Non-invasive “whole-organ” mass estimations of the ex vivo bovine muscle from scans of varying collimation and dose levels illustrated extremely high accuracies (Figure 5). Similar to the physical density quantification, dose levels did not impact the mass value. Between the five different combinations of collimation and dose, estimated mass values were within ±1.1 grams of ground-truth mass measurements with a scale. For 16x0.625 and 64x0.625 mm collimations, the errors in mass were -0.34% and -0.04%, respectively. The accuracy of estimated mass further validates our physical density results and demonstrates a clinical application of non-invasive mass measurements for determining the presence of pathology, which is currently utilized only in post-mortem autopsies.

Figure 5:

Mass estimation from spectral physical density maps, in excellent agreement with measured mass. The dashed lines represent two independent weight measurements, and the dotted line represents their average.

3.3

Non-invasive temperature monitoring

In addition to non-invasive mass measurements, physical density quantifications also enabled non-invasive temperature monitoring as temperature changes are reflected in physical density changes (Figure 6). Specifically, a linear fit between normalized physical density and change in temperature demonstrated a slope of 0.00042 ± 0.00001 °C^-1 and an intercept of 1.000 ± 0.0003 for temperatures between 22.0 and 45.5 °C. These fit parameters correspond to a 0.42% decrease in physical density with an increase of 10 °C. High linear correlation (R = 0.9781) between normalized physical density and change in temperature recapitulated the theoretical relation.

Figure 6:

Normalized physical density changes with heating and cooling of ex vivo bovine muscle. The linear relationship between normalized physical density and changes in temperature reflected thermal volumetric expansion.