From the 5th to 9th of September, a few Stellenbosch University (SU) and University of Pretoria (UP) students had the opportunity to attend a workshop on scan stitching at the South African Nuclear Research Corporation (Necsa). The workshop was hosted by Dr Lunga Bam and Kobus Hoffman, two Necsa employees who have been working tiresomely to digitise the skeletal collections of SU, UP and Sefako Makgatho Health Sciences University (SMU).
A bit of technical background before we get into the workshop content. The skeletal material is scanned using a METRIS X-TEK XT H225L Rayons-X X-Ray scanner (pictured below). Once scanned, we obtain a set of 1000 two-dimensional (2D) radiographic projection images that needs to be transformed into a three-dimensional (3D) volume set. During the Bakeng se Afrika project, smaller long bones (e.g., humeri and radii) were generally scanned in two parts (A and B), whilst larger long bones such as the femur were scanned in three parts (A, B and C). During the workshop, we learned how to "stitch" these parts together and thus obtain one final merged scan for each long bone. Dr L. Bam and Mr J. Hoffman, with the help of Dr M. Cazenave, designed a protocol to perform these steps in a reproductible way. Day one of the workshop consisted of training on the VGStudio Max 2022.2 software, while day two focused on using another software, ImageJ. Two students, Thando and Aimee, stayed on an extra day and an extra 3 days, respectively, to continue and progress on the merging of humerii.
For the purpose of this blog post, I will only focus on the steps involved in the merging of radii bones, as the femur scan data required a few more intricate steps that I feel would be as confusing to explain as it was for us students to comprehend it during the workshop. Once both datasets (parts A and B) have been imported as an image stack in the software, one will see that both scans (upper and lower parts) overlap one another. We need to first align them correctly.
Before scanning, a little bit of Prestik was placed on the shaft of each bone in order to have a point of reference, a marker. The first step is thus to locate ‘the Prestik’. Yes, you read that correctly, we were tasked with locating either one blob of Prestik on one of the bones, or sometimes both bones have a Prestik marker; in the latter case the elusive ‘floating 3rd Prestik blob’ is what needs to be located. Once located, the coordinates of the point at which the Prestik is just barely visible (either the top or the bottom thereof) is recorded and the distance with the extremity of the scan is measured in millimetres. This measurement needs to be taken on both halves of the bone (upper and lower parts) and the absolute total noted. Then comes the transformation process (not as complex as the Transformers movie transformations!). The previously calculated absolute total is simply used to shift the upper (or proximal, if you want to use your anatomy lingo) portion of the bone upwards, so it automatically aligns to the other part of the bone. Now the bones are merged, yay!
However, as you can notice in the above image, long bones were often scanned in pairs, particularly to optimise beam time. We still need to separate the bones from each other!
The separation process, still performed in VGStudio Max, requires creating Regions of Interest (ROIs). First, a region of interest is created to encompass the entirety of the ‘Prestiked bone’. Then this ROI is inverted, in order to encompass the other bone in a separate ROI. The process of creating ROIs needs to be done for both scans A and B. Once ROIs have been created, they are then extracted. Make sure to export both A and B parts for both bones. And voilà, you have two separate long bones!
On day two of the workshop, we embarked on the same goal of merging scan parts A and B, but in ImageJ. This entails a similar sequence of steps. Firstly, the entire image stack is imported into ImageJ. Then, the step of locating the Prestik blob ensues on scan A. This does not need to be a specific part of the Prestik as it was in VGStudio, any image including the Prestik will do. The exact same image then needs to be located in the B scan. Note down the image number, and go find these images in your reconstructed folders. Drag and drop them into ImageJ. Under the Analyse tab of ImageJ, Histograms need to be selected (from both scans), which will display the minimum and maximum grey values of each scan. These grey values will be used to calibrate the scans to one another (calibration value = max A/max B). Essentially, scan B needs to be calibrated to scan A. Once this step is performed, the scans can now be saved as a 2D merged dataset. ImageJ has several perks, it is free open-source software and is thus readily available, even to students. It also takes a lot less RAM to run ImageJ software in comparison to VGStudio Max. However, as there is a step of grey values calibration, the end product cannot be used for inner bone parameter analysis. On the opposite, when using the VGStudio Max method, specific bone parameters can be analysed (such as bone density) should it be required.
Overall, the workshop was an insightful ‘behind the scenes’ experience of the Bakeng se Afrika project. As a group, we would like to extend our appreciation to Dr Bam and Mr Hoffman for taking the time to share their knowledge with the next generation. We would also like to thank UP (Dr Charlotte Theye, Ms Pearl Bothma and Ms Meg-Kyla Erasmus) for arranging this opportunity for us. Personally, I am grateful to have stayed on an extra three days post-workshop, in order to practice these newly learnt skills. What seems a daunting process at first, becomes second nature of practice. I also have to say that the cute baby monkeys and zebra that roam freely within the Necsa property, were a definite highlight as it made every day feel like a safari tour haha!
A little summary of the stitching process... at x2 video speed!
Aimee Welmans
Bakeng se Afrika team
BARU (Biological Anthropology Research Unit)
Stellenbosch University