Surface and volume rendering of large datasets Kevin Mackenzie Institute of Medical Sciences, University of Aberdeen, Aberdeen, A25 2ZD Aims As part of a project I was involved with looking at the development of teeth, I recently visited Skyscan NV, Kontich, elgium to scan some medieval maxilla jaw fragments and a whole mandible (Fig. 1). The scans resulted in extremely large datasets and these proved very difficult to view on a standard Windows XP Desktop computer. It quickly became apparent that I would need a higher spec computer to allow visualization and rendering of these samples. So I purchased a Dell 64-bit Windows 7 PC (XEON 3.20GHZ) with 12G RAM plus an ATI FirePro V7800 (Fire GL) 2G graphics card, and installed the 64-bit versions of CTAn (1.1.11), CTVol (2.1), and CTvox (2.3) on it. The main objective was to analyze these scans and compare the resulting surface and volume rendered images using CTVol and CTvox respectively. The resulting quality of images and the ease with which the new computer could handle these files, especially using CTvox, prompted me to revisit previous scans from our Skyscan 1072 desktop micro-ct scanner that in the past had been difficult to open. I have also been able to demonstrate the ease with which stereo pair images (red/ blue) can be created within CTVol and CTvox, and to create small models that can be viewed on an Apple mobile device. A Figure 1: (A) Mandible taped up and ready to go into scanner, () ack projection image of combined scan of mandible using the Skyscan 1173. Method The maxilla and whole mandible samples were scanned using the Skyscan 1173 high energy, spiral scan micro CT scanner, at 115kV/69uA using a 0.25mm brass filter, pixel size 25um, rotational step of 0.15 degrees and rotated for 360 degree with frame averaging set at 2 and exposure time of 1.5sec.
The smaller maxilla samples took 1 hour 14 minutes to scan and resulted in 1440 images and file size of 13.4G. Reconstruction was then carried out using NRecon (1.6.4) using settings of ring artifact correction 6 and beam hardening of 15%. This resulted in 1769 slices and a file size of 8.26G (2k x 2k MP file format). The larger mandible required an oversized connected scan that took 6 hours 20mins to complete and resulted in 4800 back projection images with a file size of 84.8G. Reconstruction was again carried out using NRecon (ver 1.6.4) with the same settings as above. This resulted in 3188 slices and a file size of 6.03G (this time as 4k x 3k PNG file, (as MP files would have resulted in 50.7G)). These reconstructed datasets were then loaded into CTAn and 3D surfaced rendered models created and then viewed in CTVol. (Fig. 2). This was time consuming and could take a few hours or more to create a model depending on its complexity. These datasets were also opened in CTvox to create volume rendered models. Opacity was adjusted, colour texture file applied and shadow and/or material affect applied (Fig. 3). Red/blue stereo images were also created in both software packages using stereo mode from the Options menu or by selecting the red green glasses icon on toolbar. The time taken to load the dataset was considerably faster than CTVol taking only 5 to 10mins. Results A Figure 2: CTVol images (A) Mandible rendered image in p3g format, () Stereo mode 3D image of maxilla fragment (red/blue stereo pair).
C D Figure 3: CTvox images (A) Mandible volume rendered image. () Stereo mode 3D image (red/blue stereo pair). (C) Use of colour textures and adjustment of opacity. (D) Maxilla - combined colour textures with shadow and surface lighting options. As stated above, some scans from our Skyscan 1072 desktop micro-ct scanner had in the past had been difficult to open. one samples settings of 50kV/187uA and 0.5mm Al filter, 0.68 degrees step for 180 degrees, produced a back projection dataset of 276 images with a file size of 553 M; and, after using NRecon to carry out reconstruction (Ring artifact correction 7 and eam Hardening 20%), 975 images with a file size of 976 M. Sometimes, to allow the creation of 3D models, we had to reduce the reconstructed dataset size by resizing and selecting defined ROI in CTan. The new computer allows us to easily open the datasets at full size and so we are now able to utilize the many functions in CTvox. This has led to re-examination of scans we have done over the years (Fig. 4).
C D E F Figure 4: CTvox examples from 1072 scanner (A) mouse foot, () shell, (C) fish, (D) mouse vertebra (red/blue stereo pair), (E) glass fibres in resin, (F) mobile phone SIM card. We also tried loading the maxilla scans into Imaris 7.0.0 (64-bit) from itplane. (This software is normally used for viewing confocal laser scanning microscope Z stack images). We found that although it took a long time to load the scans, especially when compared to CTvox, once loaded they were easily manipulated and that the selection of normal shading surface improved the image. (Fig. 5)
Figure 5: Imaris generated images from maxilla (A) Maximum Intensity Projection volume, () normal shading surface Conclusion We found that the large datasets could be opened in CTvox without having to resize them. CTvox is very easy to use and performs well with a fast computer and large graphics card. It also has many options to slice through image, change surface texture, add shadow and adjust lighting of model, and to easily make movies. It even includes an option to output a file suitable for viewing on Apple mobile devices. Another option not commonly used in both CTVol and CTvox is the ability to create 3D stereo pairs which can give extra depth detail to the model.