3D imaging of aerated emulsions using X-ray microtomography

Transcription

1 3D imaging of aerated emulsions using X-ray microtomography G. van Dalen, M.W. Koster Unilever Research & Development, Advanced Measurement & Data Modelling, Olivier van Noortlaan 120, NL-3133AT Vlaardingen Introduction Stabilisation of air bubbles in liquid food systems is notoriously difficult. They have to be stable during production or preparation. Examples of relative stable aerated products are bread, ice cream and mousses in which the gas bubbles are entrapped within the solid continuous phase. However liquid or soft solid foams are thermodynamically unstable resulting in an increase in bubble size (coarsening) in time and eventually to complete loss of air. These products are subject to destabilisation processes such as coalescence and Ostwald ripening (diffusion of air from small to large bubbles). Stabilisation of aerated soft solid oil in water (o/w) emulsion offers a huge challenge. A better and comprehensive insight into mechanisms responsible for the formation and stabilization of foam structures is required. Quantitative methods are needed for the characterization of aerated products and therefore suitable for the derivation of substantiated structure-function relationships. The amount of gas and the bubble size distribution of these products have to be followed in time. This paper shows how X-ray microtomography (µct)(1,2) can be used for the 3D visualisation and quantitative analysis of bubbles in aerated emulsions during storage. Method Model emulsions were imaged using a Skyscan 1172 desktop µct system. The used image acquisition parameters are listed in Table 2. Tomographic reconstructions were performed on a cluster computer (HP 220c server Blade with quad core Xeon CPU with 16 cores in total) using Nrecon (v ). A smoothing factor of 4, beam hardening correction of 50% and ring artefact reduction of 20 was used. The features in the stacks of the 2D µct images were identified and measured using an image analysis toolbox (DIPlib from the Delft University of Technology). For visualisation in 3-D space, isosurface rendering was used (Avizo 6.2 from the Visualization Sciences Group). Images of time series were aligned in Avizo using affine registration. Before registration a coarse manual alignment was performed using marks on the sample holder wall (small hole visible in Figure 18 ). The quantitative analysis of the bubble size distribution will be reported in a separate paper (3).

2 Table 3 Tube diameter mm μct acquisition parameters Image size, pixels Image pixel size, μm Kv - μa Rotation step (/180 o ) Frame averaging Number scans vertical * *4000 ** * *2000 *,** *3564* * partial width used, ** settings for storage trial Figure 5 Projection image showing the total scanned area (obtained by stitching multiple vertical overlapping scans). Selection of an appropriate sub sample of the whole aerated product is one of the most critical steps in the procedure. The sub-sample should be representative for the total product. For this purpose plastic cylindrical sample holders are used (Figure 6A). The upper part of the sample holder consist of a removable open tube with a length of 20 mm. This tube was filled by pressing the tube gently into the aerated product, preventing disturbance of air bubbles in the sample and not to include additional air in to the holder. Rotation of the tube or suction of the sample into the tube may cause deformation of the bubbles. To follow air bubbles in time during a storage trial and to prevent subsampling at each time point new sample holders were developed. These sample holders are made of Perspex (polymethyl metacrylate, PMMA) and are sealed hermetically using O-rings (Figure 6B). However during first trials, loss of the continuous phase due to evaporation was observed. This in turn influenced the air bubbles in the sub-sample. The evaporation is caused by the permeability of Perspex. Dependent on the relative humidity, Perspex can absorb water up to 2.2%(4). Evaporation can be prevented wrapping the sample holders with several layers of Parafilm (a plastic paraffin film produced by Pechiney Plastic Packaging Company, USA). Alternatively sample holders of polypropylene (PP) can be used (Figure 7). PP sample holders with an inner diameter of 26mm were made from standard 50 ml centrifuge tubes with screw cap from Sarstedt ( Tubes were cut at the 20ml mark and mounted on a special made base (see Figure 6C).

3 Figure 6 Sample holders for imaging emulsions. Figure 7 Influence of sample holder on the storage of gram water (w) and an emulsion (o/w) in sample holders with an inner diameter of 11mm.

4 Results Representative cross sections of µct images of an aerated o/w emulsion with 25% oil imaged at a pixel size of 2.0, 4.0 and 8.4 μm using sample holders with an inner diameter of 6, 11 and 26 mm are shown in Figure 5. The air bubbles are clearly visible within the fat/protein/water matrix by their high grey value (low absorption coefficient). These images show a broad distribution of bubble sizes including very small and very large bubbles. A high magnification (pixel size of 2.0 μm) reveals the presence of clusters of very small bubbles. However at this magnification the field of view is limited, missing the very large bubbles. Selection of the size of a sample holder will always be a compromise between obtaining maximum detail (resolution) and being representative for the total sample. For accurate analysis of the total range of bubble sizes in the sample shown in Figure 5, images could be made using sample holders with an inner diameter of 6 and 26 mm. Larger sample holders will also result in less sampling artefacts and will be less influenced by environmental conditions during storage, representing more the actual shelf life. Figure 1: µct images of an aerated o/w emulsion imaged in sample holders with an inner diameter of 6, 11 and 26 mm, scanned with a pixel resolution of 2.0, 4.0 and 8.4 μm, respectively. Bottom: enlarged view of the area indicated by the yellow box. Quantitative analysis of the size distribution of bubbles in μct images requires segmentation of the bubbles from the background (in this case the matrix is a homogeneous emulsion). For segmentation using global thresholding, the grey values of the matrix should be uniformly distributed over the total area within the sample holder. However beam hardening will result in gradual change in grey level. The edges of the sample appear more dark (higher absorption) than the centre (cupping). Beam hardening is caused by the preferential absorption of low-energy photons from the polychromatic X-ray spectra produced by the X- ray tube. A larger sample holder and therefore a longer path-length through the sample results into more beam hardening (Figure 8). Standard beam hardening correction(5) during tomographic reconstruction (square function in NRecon) reduced the cupping effect but could

5 not remove cupping completely. A better correction was obtained using a 3th order polynomial function (Figure 9B, C & D). Figure 8 Relative grey level profiles 1 through reconstructed images of an o/w emulsion without added air in sample holders with an inner diameter of 6, 11 and 26 mm showing the influence of beam hardening for different sample diameters (with transmissions 2 of 26, 17 and 15% and RSD 3 values of 24, 13 and 18% respectively at 50% beam hardening correction). Images are shown without beam hardening correction. Correction of beam hardening artefacts during reconstruction causes an increase in noise, thus a decrease in quality (Figure 9A&B). An other possible strategy to reduce beam hardening is to change the energy distribution of the X-ray beam. An increase in tube voltage will increase the maximum photon energy of the continuous radiation. The transmission through the sample increases linearly with the tube voltage (Figure 10B). However this has no significant influence on beam hardening and noise (Figure 10A). The lower energy part of the continuous radiation can be blocked by using attenuation filters after or before it passes through the scanned sample. In this way a harder, more monochromatic beam is presented at the sample. As can be seen from Figure 11B, beam hardening can be reduced by using an aluminum filter of 0.5 mm. In this case a beam hardening correction of only 50% will result in a flat profile. The main drawback of using a filter is that it degrades the X-ray signal at all energies to some degrees, leading to greater image noise. Even when using a higher voltage (59>95kV) and a higher acquisition (exposure) time (589>1767ms), a much higher noise level is observed (Figure 11A). The longer acquisition time leads to an increase in total scan 1 The images and graphs were scaled relative to the average grey level in the centre. The profiles show the vertical average over 50 pixels along a line shown in the images. 2 Transmission through the centre of the sample. 3 The relative standard deviation (RSD) of the mean grey level (matrix air) over an area of 200*200 pixels.

6 time from 1.5 to 2.5 hours (scan with a rotation step of 0.20 o /180 o and frame averaging of 4). So this will not be a preferred option, especially because multiple scans (oversized scan) are needed to image the total sample holder. A cupping, % Cupping % RSD Beam Correction hardening correction, depth, % % 0 - second order % RSD B cupping, % Cupping % RSD Beam hardening correction Coefficient third order Beam hardening correction, 3e order coefficient a % RSD C relative grey level e-order coefficient: (0=100% BHC) 11 mm D 0.5 Figure profile, pixels Influence of beam hardening correction (BHC) on cupping 4 and RSD of an o/w emulsion without air in a sample holder with an inner diameter of 11mm. A: standard BHC ranging from 0% (no correction = linear function) to 100% (square function), B: BHC using third order function *Intensity 2 + a 3 *Intensity 3 ), C: relative grey level profiles. A B Figure 10 Influence of voltage (power = 10W) on cupping and RSD of an o/w emulsion without air in a sample holder with an inner diameter of 11mm. 4 The cupping is calculated as a percentage of the average grey value at the border of the sample relative to the average grey value at the centre.

7 A B C 0% - 50% - 100% beam hardening correction 11 mm Beam hardening correction, % Profile, pixels Figure 11 Influence of beam hardening correction on cupping and RSD of an o/w emulsion without air in an 11mm sample holder. Beam hardening was reduced using an aluminium filter of 0.5 mm. Transmission = 55%. Beam hardening can also be reduced using post processing of the horizontal cross sections obtained after tomographic reconstruction. Two options were tested, using a cross sectional average of the total stack of slices of 1: the sample (same dataset) and 2: a physical phantom (separate dataset). The phantom should be uniform and comparable density, size and shape, imaged using identical acquisition parameters. As phantom an emulsion without air was used. For both methods the extent of the beam hardening has to be same for each slice. For an aerated emulsion the beam hardening effect is not significantly influenced by the vertical position. The result of the first correction method is shown in Figure The cross sectional average was obtained by averaging a selection of about 40 slices, equally divided over the total stack. These slices were first aligned and clipped at about 90% of the average grey level of the matrix in the centre of the sample holder to reduce the contribution of the low grey value of the bubbles. The cross sectional average was blurred using a Gaussian filter with a sigma of 6 pixels and stretched between 1 and 99% percentile. After angular averaging this image was used to correct the total stack of slices of the sample: each slice was divided by the correction image and stretched between 1 and 99%. The result of the second correction method using a non aerated emulsion as phantom is shown in Figure The procedure is comparable to the first method. However in this case no clipping was used and the corrected image was aligned with the image of the sample. The final results of both methods are more or less comparable. Only method 2 also corrects for some residual ring artefacts (no blurring used as in method 1). For quantitative analysis the μct images have to be smoothed to remove noise. This can be done during tomographic reconstruction using Nrecon or after reconstruction using image analysis software. Figure 13 shows the influence of smoothing in Nrecon (smoothing kernel = asymmetrical boxcar) on the visible quality of the images and on the obtained binary images. A low smoothing factor results in binary images containing many small features and large bubbles which are not filled. A large smoothing factor will decrease the resolution, removing small bubbles. The impact of smoothing on small bubbles is shown in Figure 14 and Figure 15. Bubbles with a diameter of 12 and 17 μm will disappear after using a smoothing factor of 4 and 8, respectively. The selection of the smoothing factor will be a compromise between accurate quantitative analysis of large and small bubbles. For this application a smoothing factor of 4 was selected.

8 2 Profile, pixels Figure 12 Correction for beam hardening using cross sectional averages of the total stack of slices of the sample (top) and a non aerated emulsion (bottom). The images were stretched between 1-99% percentile (BHC=50%).

9 Figure 13 Influence of smoothing on the reconstructed image of an o/w emulsion imaged in a sample holder with an inner diameter of 11mm. Grey level image (1.7mm*1.7mm) stretched at 1% percentile with binary image in red obtained using a fixed threshold value of 140.

10 Figure 14 Relative grey level profiles through an air bubble with a diameter of 17 μm (FWHM) within a reconstructed image of an o/w emulsion in sample holders with an inner diameter of 11 mm showing the influence of smoothing during reconstruction. Grey levels were scaled between matrix (1) and air outside the sample holder (0). Image pixels size = 4 μm. Figure 15 Influence of smoothing on the contrast (relative grey level (matrix-bubble)/(matrix-air) ) and noise (RSD of the matrix) within a reconstructed image of an o/w emulsion (Image pixels size = 4 μm). The stability of aerated o/w emulsions was tested by imaging for each sample exactly the same sub sample in time. For this purpose sub samples were imaged using Perspex sample holders with an inner diameter of 6 and 11mm (Figure 6B). The sample holders were stored at ambient conditions and imaged during 16 weeks. Important microstructural parameters of aerated products are the air content and the bubble size distribution. For this study the volume weighted mean diameter D[4, 3], and the surface area weighted mean diameter D[3, 2] were measured. Examples are shown in Figure 16 and Figure 17. The gas bubbles of this sample were not stable during storage. The number of air bubbles per volume was reduced and larger air bubbles were formed in time. For large bubbles coalescence was observed (bubbles combine into larger ones). For small bubbles Ostwald ripening was observed (Figure 18 and Figure 19). The air phase is transferred from small to large bubbles due to their different Laplace pressures. It is controlled by the solubility of air in the emulsion. At locations were

11 the bubbles disappeared a higher absorption was observed (Figure 20). This can be due to the migration of water (higher absorption than oil/water). Beside a decrease of the total volume of gas bubbles in time, also a decrease of the emulsion volume was observed (Figure 17). This was caused by evaporation (see discussion about sample holders in the method section). Figure 16 μct images of horizontal (middle) and vertical (bottom) cross sections of an aerated o/w emulsion during 16 weeks storage in a sample holder with an inner diameter of 11mm. The top row shows the 3D visualisation of the bubbles (box size = 13.8mmx13.8mmx20.4mm, pixel size = 8.0 μm). Figure 17 Bubble parameters of an aerated o/w emulsion during storage calculated from the images shown in Figure 16.

12 1 day 3 days 1 week 1 month 6 mm Figure 18 μct images of horizontal cross sections of an aerated o/w emulsion during 1 month storage in a sample holder with an inner diameter of 6mm. Enlarged views (2.4mm * 4.8mm) at different storage times with outlines of bubbles detected at T0 in red. Figure 19 3D visualisation of μct images of horizontal cross sections of an aerated o/w emulsion during 3 days storage with outlines (red grid) of bubbles detected at T0.

13 xz 1 day xy 3 days xy xz Figure 20 μct images of horizontal and vertical cross sections (2.0mm * 2.0mm) of an aerated o/w emulsion at 1 and 3 days after preparation. Conclusion X-ray microtomography allows in-situ observation and analysis of bubbles during storage of aerated emulsions. Various coarsening mechanism can be identified and investigated, namely coalescence and Ostwald ripening. References: 1. Dalen G van, Don A, Nootenboom P and Blonk JCG (2009) Determination of bubbles in foods by X-ray microtomography and image analysis, SkyScan user meeting, Ghent, Belgium, April 2009, Dalen G van, Nootenboom P, Vliet LJ van, Voortman L and Esveld E (2007) 3D imaging, analysis and modelling of porous cereal products using x-ray microtomography, Image Anal Stereol 26: Dalen G van and Koster, M. (2011) 3D visualisation and quantification of bubbles in emulsions using μct and image analysis, Proceedings of the 13th International Congress of Stereology (ICS-13), Beijing, China, October Properties of Perspex, Perspex South Africa (Pty) Ltd N.pdf 5. Casteele, E (2004) Model-based approach for beam hardening correction and resolution measurements in microtomography, thesis University of Antwerp, Belgium