Freeze-fixation of bubbles for micro-ct imaging of liquid aerated food emulsions G. van Dalen 1, M. Koster 1, J. Hazekamp 2 1 Unilever Research & Development, Imaging & Spectroscopy, Olivier van Noortlaan 120, 3133AT Vlaardingen, the Netherlands 2 Unilever Research & Development, Colworth Science Park, Sharnbrook, UK Introduction X-ray micro tomography (µct) is used for the 3D visualization and quantitative analysis of air bubbles in aerated food products 1. μct allows in-situ observation and analysis of bubbles during processing and storage 2. Various coarsening mechanisms can be identified and investigated, namely coalescence and Ostwald ripening. However the air bubbles in the sample should be stable and static during imaging to prevent movement artefacts in the final reconstructed volume. Laboratory µct scanners need a relative long scan duration in order to obtain good quality images suitable for quantitative analysis. General acquisition times up to 90 minutes are used on static systems to generate full resolution data series. For a mobile sample, this may even be reduced to several minutes at the cost of image quality (lower resolution and poor signal-to-noise ratio). We found that many liquid foams need much faster acquisition times which are only possible using Synchrotron µct. To avoid this limitation, samples can be immobilised using freezing methods 3. In this study liquid emulsions were fast frozen and imaged using a special cooled sample holder adequately retaining the sample in a frozen state. The method was validated using aerated dairy desserts. Method Commercially available liquid aerated dairy desserts were used containing water, protein (1.8 g/100ml), fat (5.0 g/100ml), sugar (7.9 g/100ml), starch, thickeners and emulsifiers. The bubbles in these products are relatively stable making it possible to image them in non frozen state. To compare images before and after freezing, images were made directly before freezing to prevent bubble coarsening in time. The samples were frozen in dry ice (solid carbon dioxide). The samples were imaged using a Skyscan 1172 desktop µct system. The image acquisition parameters used are listed in Table 1. Tomographic reconstructions were performed on a graphical card (GPUReconServer, Engine version=version: 1.6.9) using Nrecon (v 1.6.9.4). A smoothing factor of 4, beam hardening correction of 80%, ring artefact reduction of 20 and constant histogram settings (0-0.035) were used. For quantitative analysis and visualisation in 3-D space, isosurface rendering was used (Avizo 8.0.1 from the FEI - Visualization Sciences Group, http://www.vsg3d.com/avizo/fire). To compare images before and after freezing, the images were aligned in 3D space by using the AffineRegistration module in Avizo. The frozen samples were cooled during the X-ray scanning using a special sample mount containing dry ice on top and bottom (figure 1). The dry ice creates a cold blanket around the tube. In addition the tubes were also encased in polystyrene foam (Styrofoam) for isolation (Figure 2). The sample holders with an inner diameter of 26mm were made from standard polypropylene (PP) 50 ml centrifuge tubes with screw cap (available from Sarstedt, www.sarstedt.com or Corning www.corning.com). Tubes were cut at the 20ml mark and mounted on a special made base. Freezing of a sample in a 50ml tube covered with dry ice pellets (upright) took about an hour to reach a temperature of -80 o C (Figure 3A).
Without isolation the sample remains frozen for about 1 hour and with the stryrofoam isolation for about 2 hours (Figure 3B). For image analysis the following steps were used: a) noise reduction using a 3D median filter, b) masking of the sample holder, c) thresholding using a constant threshold level (105 for all images in this study), d) removing particles with a volume smaller than 27 pixels, e) separation of touching particles, f) removing particles touching the lower and upper edge, g) measuring size per particle, h) generation of a size distribution and calculation of mean particle diameters 4 and i) generation of a size labelled image (classifying measures with sieves). For segmentation a watershed transform of the Euclidean distance map was used. In Avizo Fire a standard function was used combining both procedures. For the noisy (scan A, Table 1) additional steps were used after thresholding: filling holes in 2D and closing in 3D. The volume fraction of air bubbles was determined from the total stack of binary images by dividing the number of pixels identified as air bubbles by the total number of pixels inside the sample holder. Figure 1 Sample tubes modified for µct imaging of frozen emulsions with a sample holder assembly containing dry ice on top and bottom. Figure 2 Sample tube with and without a 1 cm polystyrene foam layer.
Figure 3 Temperature as function of time during freezing (A) and defrosting (B) of an aerated sample containing about 35% air in a 50ml sample tube. The temperature was measured inside the tube near the wall (A and B) as well as in the centre (A). The tubes were measured both with and without a 1 cm polystyrene foam layer as isolation. Table 1 μct acquisition parameters (59kV/167 μa) Scan duration min/subscan Image size, pixels Image pixel size, μm Rotation Rotation step (/180 o ) Frame averaging A 8 2000*2000 16 180 1.0 1 B 30 2000*2000 16 180 0.25 1 C 174 4000*4000 8 180 0.20 3 D 56 2000*2000 16 360 0.25 1 Figure 4 shows the difference in image quality between the scans outlined in Table 1. From the images of two different aerated dessert samples the amount of air and the mean bubble diameters were analysed (Figure 5). For aerated products often the surface-weighted mean diameter 3,2) and the volumeweighted mean diameter 4,3) are used. The number-weighted mean diameter 1,0) gives a less representative value because a few small bubbles can skew the distribution. The definition and nomenclature of the mean particle diameters p,q have been reported by Alderliesten 5. The volume fraction of air analysed from scan A and B is slightly lower than obtained from analysing scan C. The mean bubble diameters obtained from scan B and C are comparable. For the fast scan (A) lower mean diameters are obtained due to the noisy images. These fast scans were used to investigate the homogeneity of air bubbles in the total tube by stitching 4 scans together and covering the whole sample tube in a short as possible image acquisition run. Scan B, at intermediate scan duration, was used to investigate the influence of freezing on the bubble size and amount of air, measured on a non isolated sample holder.
Figure 4 Micro-CT images of aerated dessert sample 1 obtained using different scan parameters outlined in table 1 resulting in scan durations of 8min (A), 30 min (B) and 174 min (C) showing representative total horizontal cross sections (I) with enlarged views of grey level image (II), binary image (III) and labelled image (IV).
Figure 5 Volume fraction of air and mean bubble diameters of 2 aerated dessert samples calculated from images obtained using the scan parameters outlined in table 1. The dotted lines in left graph represent the percentage of air calculated from the density of aerated and non-aerated product. Results Representative micro-ct images of two aerated dessert samples before and after freezing are presented in Figure 6. For this experiment the non-isolated sample holder was used. Exactly the same sub volumes were imaged and aligned in 3D (Figure 2). Freezing affects the microstructure of the sample showing an increased ice artefact towards the centre of the specimen. Especially at the centre of the tube most bubbles were slightly shifted after freezing. Near the tube wall most bubbles remain on the same location as a result of a higher cooling rate, hence less ice formation, at the periphery of the sample. Also the shape of many bubbles changed from spherical to a more elongated. The crystallization of water compresses and moves the bubbles. Results of the quantitative analysis of the amount of air and bubble size distribution are shown in Figure 7 and Figure 8. After bubble size measurement a label image was generated showing the size classification of each bubble (Figure 9 and Figure 10). The volume fraction of air is about 12 % lower after freezing. This is mainly caused by the expansion of the water phase during freezing. Ice is approximately 8.3% less dense than liquid water. The bubble size distribution is comparable before and after freezing with slightly (2-3 %) smaller bubbles after freezing.
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 2100 2200 2300 2400 Figure 6 Horizontal (top) and vertical (bottom) micro-ct cross section of 2 aerated dessert samples before and after freezing (non-isolated sample holders with an inner diameter of 26mm, scan parameters B). r e l a t i v e v o l u m e 100 90 80 70 60 50 40 30 20 10 0 1 -ambient 1 - frozen 2 -ambient 2 - frozen Figure 7 diameter (class end point) Cumulative bubble size distribution of two aerated desserts before and after freezing. Figure 8 Volume fraction of air and mean bubble diameters of 2 aerated dessert samples before and after freezing.
Figure 9 Size labelled μct images of the aerated dessert sample 1 before and after freezing showing from top to bottom a horizontal and vertical cross section and a 3D volume rendering before and after clipping (box size = 29mmx28mmx14mm, pixel size = 16 μm). For grey level images see Figure 6).
Figure 10 Size labelled μct images of the aerated dessert sample 2 before and after freezing showing from top to bottom a horizontal and vertical cross section and a 3D volume rendering before and after clipping (box size = 29mmx28mmx14mm, pixel size = 16 μm). For grey level images see Figure 6).
Representative micro-ct images of two aerated dessert samples before and after freezing are presented in Figure 11. For this experiment the isolated sample holder was used (Figure 3). This sample holder enables longer scanning times resulting in higher quality images. Figure 11 Horizontal (top) and vertical (bottom) micro-ct cross section of aerated dessert samples 2 before and after freezing using an isolated sample holder (tubes with an inner diameter of 26mm, scan parameters D). Conclusion Air bubbles in liquid aerated food emulsions can be quantitatively analysed using μct after freezing of the samples. The frozen matrix facilitates the aquisiation of high resolution datasets by reducing mobility of the specimen. However freezing will affect the microstructure of the sample (size shape and location of bubbles). The relative volume fraction of air is about 12 % lower after freezing either as an effect of schringkage of air or compression by the expansion of water freezing. The bubble size distribution is comparable before and after freezing with slightly (2-3 %) smaller bubbles after freezing. With a special sample holder containing dry ice and polystyrene foam scan times up to 2 hours can be obtained.
References: 1. Dalen G van, (2012) A study of bubbles in foods by X-ray microtomography and image analysis, Microscopy & Analysis, S8-S12 2. Dalen G van, Koster, MW (2011) 3D imaging of aerated emulsions using X-ray microtomography, SkyScan micro-ct user meeting, Leuven, Belgium, 13 15 April, 31 3. Cavalier, A, Speher, D. and Humber, B. (2008). Handbook of Cryo-Preparation Methods for Electron Microscopy. CRC Press. 4. Dalen G van, Koster, MW (2012) 2D & 3D particle size analysis of micro-ct images, Bruker micro-ct user meeting, Brussels, Belgium, 3-5 April, 53999 5. Alderliesten, M. (2008). Mean particle diameters, From Statistical Definition to Physical Understanding, PhD Thesis, Technical University Delft, the Netherlands.