2.2. Representative Sample Preparation/Selection
A maritime pine (Pinus pinaster Ait.) roof beam was collected from a residential building in Lisbon, Portugal after about 60 years of service life. The presence of different levels of anobiid infestation was easily recognized on the sapwood. The main insect responsible for the deterioration was identified as Nicobium castaneum Olivier by the size of the galleries, the characteristics of the frass, and of the several cocoons that were retrieved from the samples.
After the insect identification, timber was divided in 4 segments (3 containing degradation and the sound heartwood) and cut to produce approximately 40 × 20 × 40 mm
3 samples. These samples were then crosscut to obtain 17 “new” paired samples, with approximately 40 × 20 × 10 mm
3, that were adequate for μ-XCT [
5]. These samples (
Figure 2) representing the varying degrees of degradation along the beam were submitted to μ-XCT. The heartwood was not used.
The samples were assigned into 3 levels of degradation (level 1V, 2V, and 3V) through a visual analysis of the deterioration and considering the area of the emergence holes at the exposed surfaces of the degraded samples (
Figure 2). Level 1V corresponds to the lowest degradation level and level 3V to the highest. It should be noted that this visual grading approach had already been proved as not fully representative of the internal degradation [
7,
8]. After determining the tomographic parameters of interest (wood and voids’ volume), the designation levels of the samples were readjusted according to the results of the lost material percentage.
Before the scanning procedure, all samples were conditioned for 2 weeks in a climatic chamber (temperature (T) of 20 ± 2 °C and a relative humidity (RH) of 65 ± 5%) to stabilize and reach maritime pine sapwood’ reference moisture content (12%) [
25]. No previous treatment was required for the samples to be submitted to μ-XCT as this technique requires very little sample preparation and, regularly, a sample can be scanned exactly as provided; however, the best sample size required for the scanning procedure has to be taken into consideration [
26].
2.3. Scanning Procedure (Acquisition)
All maritime pine samples were scanned with a compact desktop with micrometric range resolution, μ-XCT Skyscan 1172 microtomograph (Bruker Instruments, Inc., Billerica, MA, USA), using computer-controlled tomography acquisition, processing, reconstruction, and analysis software packages (CTAn and ImageJ software provided by Bruker (Billerica, MA, USA) and NIH (Bethesda, MD, USA), respectively).
However, prior to each sample scanning procedure, a decision must be made concerning the best sample size, which is determined based on at least the following criteria: the aim of the problem to be studied and the physical, the geometrical, and operating constraints of the scanner. The next step in the preparation for scanning procedure of the μ-XCT methodology is choosing optimal scanner settings. There is a wide range of scan settings that substantially affect image quality such as resolution, scanner voltage, scan time, number of images acquired, angle of rotation, etc.
Resolution selection may be the first major factor that affects a μ-XCT scan. A resolution about 1000 times smaller than the width of the sample is recommended; however, reference standards exist and may be used [
26,
27]. A maximum resolution of 2 µm was used in the present study (
Table 1), although, a generally accepted standard for industrial CT systems does not exist yet.
X-ray scanner voltage is highly dependent on the type of material studied and their composition. For biological samples, a voltage in a range of 30 to 100 kV is recommended [
26].
The scanning time used for each sample was 1.5 h. Scanning time varies according to the system used, for example, due to the detector sensitivity or to the distance from source to detector [
28], while the number of images required varies with the sample size. Furthermore, in our case, the need to obtain the most information from each scan suggests considering at least an oversize and multi-acquisition setting as a constraint. Therefore, a larger sample could require a longer time of acquisition if multiple scans were performed. Moreover, large samples have magnification limitations due to conic beam configuration. A reduction in the angle of rotation as well as an increase in the number of images acquired will, naturally, lead to a production of images of better quality. However, the benefit level may not be justificatory because such option will contribute to increasing scan time, reconstruction time, and data size [
29]. At the end, the best settings solution for each scanning step should be found considering the output parameters that were previously defined.
During the scanning process, the studied sample must be fixed on a support while rotating in steps around a fixed vertical axis [
30]. A shadow projection image at each angular position is taken and 2D digital radiography images sets are obtained. Later, through reconstruction and rendered processes, using Skyscan-Bruker software, a set of 2D reconstructed object slice images can be combined/rendered to produce a virtual 3D reconstructed object model.
A summary of the parameters used in the scanning procedure are presented in
Table 1. Besides these parameters, the distance between the
X-ray source and the sample, as well as the distance between the
X-ray source and the detector was 257 and 350 mm, respectively. The scanning time was 1.5 h and a 0.5 mm aluminium filter was used for absorbing the low-energy
X-rays, thereby reducing noise.
The parameters presented in
Table 1 were applied in the scanning procedure of all studied samples. After the selection of the shape and the size of samples, the
X-ray microtomography operation procedure was optimized to produce the best possible images [
30,
31,
32,
33,
34].
2.4. Reconstruction
Reconstruction process follows the scanning procedure, and here the 3D volume is reconstructed from the 2D digital radiography image stacks that were previously acquired. For the reconstruction process, the NRecon software, provided by Bruker, was used. This software allows the adjustment of three reconstruction parameters: smoothing, beam-hardening factor, and ring artefact reduction. These parameters will obviously affect the quality of the acquired 3D object. However, there are no generalized universal optimal parameters, and the best solution for each case study depends on the type of scanner used as well as the type of material studied. It is also possible to choose the type of output file: 16-bit or 32-bit. Though both options are valid, the set of images exported in a 32-bit file has the best quality. The area to be reconstructed also needs to be selected and should correspond to the area of interest. To save time and to have a manageable size of the data sets, some less relevant areas should not be considered [
29], and a slightly larger area than that required should be reconstructed, as this ensures that critical data are retained.
Table 2 shows the reconstruction parameters used in this study. These parameters were used for the reconstruction of all studied replicates. The data was exported to 32-bit files.
Next, DataViewer software was used to adjust the grayscale of images to enable a better visualization (
Figure 3).
It is also necessary to choose the images viewing plane that are going to be examined. There are 3 possible options (3D objects): coronal images (radial plane)—axis xy; sagittal images (tangential plane)—axis xz; trans axial images (transverse plane)—axis yz. In this study, the coronal plan was chosen because it was the plan where the images had less visible artefacts and, because of that, it was easier to distinguish between wood and voids (tunnels formed by beetles).
With the acquisition of the 3D object, it is now possible to visualize it using a visualization program (e.g., CTVox software, provided by Bruker). The reconstructed 3D object was visualized using CTVox software (
Figure 4).
2.5. Analysis
After the reconstruction procedure, the images can be analyzed and parametrized to obtain quantitative information (the values of the parameters of interest). This process is highly dependent on the software used. For this study, CTAn software, provided by Bruker, was used together with ImageJ software (
https://imagej.nih.gov/ij/, accessed on 13 April 2021), provided by NIH, for image processing and calculation. In this case study, ImageJ software allows complementarity with CTAn, enabling easier and more flexible definition of regions of interest (ROI) in image stacks and an auto interpolation of a subset of images, which can bring some advantages linked to the elimination of local noise in stacks and sub stacks of tomographic slices, for instance. In this case study, the segmentation step is one of the most important imaging treatments that is made in this phase and consists in validly transforming the original images, in a scale of 256 gray levels, into simpler binary images (
Figure 5). With these images, it is now possible to separate the features of interest (wood and voids). These features must be separated as good as possible for the calculation process. Note that those images continue to have noise, and it is impossible to remove it completely during the entire μ-XCT process. For this reason, this step must be carried out with special attention and by a single user, so that the study is not misrepresented, mistaking, for example, wood (e.g., early growth) with voids or voids with noise. It is very important to remove the noise while keeping the structural and morphological information, that may be at the same order of scale as the noise [
35]. The accuracy of the estimated parameters’ values is strongly dependent on the quality and robustness of segmentation process [
36]. A generalized image processing algorithm must be defined (
Table 3) to segment three-dimensional images of complex materials to extract the different phases (in this study, there are 2 phases: wood and voids).
The area to be analyzed from the reconstructed images, i.e., region of interest (ROI), must be selected using a free-hand tool. Then, an auto interpolation among different ROI levels will produce the total volume of interest (VOI) for all selected frames.
Thresholding is one of the first and most important parameters to be selected. Threshold converts the set of images, originally in a grayscale, in a binary set. From the application of this parameter, sets of binary discrete images are established, with the two structural parameters of interest defined and separated. However, the choice of threshold can vary from user to user, and therefore result in differences that are larger than the experimental differences themselves [
36]. Due to that fact, global thresholds must be chosen by one user and kept during the entire study. This reinforces the importance of having a defined basis algorithm (
Table 3). Before thresholds, images can and should be filtered to remove noise. Bruker’s CTAn software allows the adjustment of analyzing parameters such as filter, threshold, despeckle, and additional morphological operations. Each one of these parameters has as a function of treating image sets to get a final set of images, from which interest parameters can be easily calculated.
Finally, with the images treated according to the interests of the study and with the defined volume of interest (VOI), it is possible to calculate the parameters of interest (wood and voids) values. These parameters’ values are obtained through a parametric 3D analysis of the object.
The results (output parameters) are obtained as a txt file exported from CTAn. From the estimated output parameters values, two of them can be highlighted: tissue volume, referring to the volume of interest (VOI), and bone volume, referring to the wood itself (skeleton). As this technique has been developed to study human bones, several less obvious terms were kept even for different uses.
The tissue volume and the bone volume will correspond to that of interest defined previously, as voids volume will correspond to the difference between VOI and bone volume. Besides these parameters, it is also possible to extract a set of other parameters, such as the length of the tunnels inside wood, for example, that provide a more thorough object analysis. However, by using these values one must be cautious, in the sense that the used algorithm was defined for the study of two particular interest parameters (wood and voids volumes).
After the scanning procedure, all 17 samples were conditioned for 2 weeks in a climatic chamber (T = 20 ± 2 °C and RH = 65 ± 5%) to stabilize and reach a moisture content identical to that of reference (12%). After that, all the samples were weighed using a scale with a precision of 0.001 g.
Original apparent density (reference density) is calculated by Equation (1):
in which m is the sample weight in kg and o.v is the object volume (wood) in m
3.
Residual apparent density (after degradation) is calculated by Equation (2):
in which m is the sample weight in kg and t.v is the total volume (wood and voids, corresponding to the VOI) in m
3.