Sediment cores
During the recently performed research cruise M125 with the German research vessel Meteor, a 5.83-m-long sediment core was retrieved from a 25-m-high CWC mound in 860-m water depth off the coast of Brazil (station M125–34-2; 21°56,959′ S, 39°32,031′ W; Fig. 1) [4]. This CWC-bearing sediment core has the potential to provide an insight into the origin, growth, and demise of these mounds as well as to carry out paleoceanographic reconstructions [5, 6].
On board of the vessel, the sediment core was split into sections of 1 m, but remained unopened as a prerequisite for subsequent CT imaging at the Clinic of Diagnostic and Interventional Radiology (DIR) of Heidelberg University Hospital, Heidelberg, Germany. For paleoceanographic analysis, the 1-m sections are cut by a diamond saw (blade thickness 2–3 mm) along the longitudinal axes while still frozen in order to avoid sediment and coral disturbances (see Fig. 1). As described in detail below, CT acquisitions of the sediment core sections containing CWCs were performed for sampling planning. However, despite taking great care to align cores with the main axes of the CT scanner system using the system’s laser sight, shape, length, and mechanical deformation of cores may influence accurate core alignment, thus leading to deviations in expected and actual location of coral clasts within the sediment cores, as detailed in Fig. 2. To allow for the calculation of geometrically accurate sagittal reconstructions of CT acquisitions of sediment core sections containing CWCs, an alignment correction algorithm was developed for restoring alignment of imaged cores with the main axes of the CT scanner’s coordinate system. The developed algorithm was evaluated for its performance using CT acquisitions of a standard image quality phantom.
Image acquisition
Phantom
For evaluating the algorithm performance, a CT acquisition of a standard image quality phantom was used (ConeBeam Phantom, QRM Quality Assurance in Radiology and Medicine GmbH, Möhrendorf, Germany). The phantom was scanned at 120 kVp tube potential and 150 mAs tube current-time product with a pitch of 0.95 (SOMATOM Definition Flash; Siemens Healthineers, Forchheim, Germany). Images were reconstructed with a soft-tissue kernel (B30f), a slice thickness/increment of 1 mm/1 mm, and an isotropic voxel size of 0.39 mm by 0.39 mm.
Sediment cores
Six sediment core sections from cold-water coral mounds taken during research cruise M125, as described above, were imaged using the abovementioned CT scanner system (SOMATOM Definition Flash). The cores consist of sediment embedded in the core liner. Individual core sections had a length of up to 1 m and a diameter of approximately 12 cm. For image acquisition, core sections were manually aligned with the main axes of the CT scanner coordinate system using the integrated laser sight system. Image acquisitions were performed at 140 kVp tube potential and 570 mAs tube current-time product with a pitch of 0.4. Images were reconstructed with iterative reconstruction (ADMIRE, Siemens Healthineers) using a sharp kernel (I70h level 3), a slice thickness/increment of 0.5 mm/0.3 mm, and an isotropic voxel size of 0.35 mm by 0.35 mm.
Alignment correction algorithm
A previously developed algorithm for positioning of regions of interest (ROIs) in image quality phantoms was adapted for the task of correcting sediment core alignment [7]. The algorithm was designed to restore alignment of a scanned cylindrical object with the longitudinal axis of the scanner as defined by the axes of the image stack (Fig. 3). The algorithm performs the following steps:
-
1.
Detecting three points on the boundary of the cylinder by one-dimensional edge detection on each individual image
-
2.
Calculating circle equation for the detected object on each individual image, determining centre position and radius
-
3.
Dividing image stack into segments based on changes in centre position and radius
-
4.
Determining orientation of the longest segment based on centre positions relative to the CT coordinate system
-
5.
Applying alignment correction based on determined orientation by translating and rotating images in the inverse direction
First, three points on the boundary of the cylindrical object are determined by one-dimensional edge detection for each image of the image stack. Edge detection is performed twice along the x-axis of the reconstructed images from the outside to the inside of the field of view (once in each direction). The third line search is performed along the y-axis, from top-to-bottom, thus omitting the CT scanner’s patient table (see Fig. 3). The line search is performed along the central rows of the image, using a bisection algorithm that minimises the variance of grey values of the two sections [7].
Second, using the three detected points on each image, the centre coordinates and radius, x0, y0, and r, defined by the circle equation, are determined for each image:
$$ {\left(x-{x}_0\right)}^2+{\left(y-{y}_0\right)}^2={r}^2 $$
(1)
Third, image segments are defined based on the changes in centre position and radius between images. Here, a segment is a continuous number of images, which have a very similar centre and radius: Images belong to the same segment, if the relative change in radius between two images is below 0.5% and if the change in position is below two voxels in x- and y-direction. Segments shorter than five images are removed. Ideally, this step will yield one segment containing the entire core (Fig. 4).
Fourth, the longest segment is chosen as the one most likely belonging to a major section of the object. A straight line is fit by regression through the determined circle centres of the chosen segment (see Fig. 4). Based on the line equation, the offset relative to the centre of the reconstructed image stack (and thus the CT coordinate system) and the rotation around x- and y-axis can be determined.
Fifth, the image is translated and rotated using the inverse of the determined offset and rotation, using the “Insight Segmentation and Registration Toolkit” (ITK, Kitware Inc., Clifton Park, NY, USA) [8].
Phantom evaluation
For evaluating algorithm performance, the CT acquisition of a standard image quality phantom was randomly rotated and translated, alignment correction was performed, and the algorithmically determined misalignment was compared to the ground truth.
Parameters for rotation and translation were randomly determined according to a uniform distribution. The maximum translation along x- and y-axis was up to 15 mm each. The maximum rotation around x- and y-axis was up to 2.86° each (corresponding to 0.05 rad). For larger translations and rotations, parts of the phantom may be outside of the reconstructed field of view, making a successful alignment correction impossible. Rotations and translations were applied using the abovementioned tool. For evaluation, the process was repeated 1000 times, algorithmically determined rotation and translation were compared to the ground truth of the generated parameters, and descriptive statistics were calculated.
Furthermore, four ROIs were placed in the low-contrast section of the phantom and mean CT numbers of the ROIs were compared for the initial acquisition, before and after alignment correction (see Fig. 4).
Statistical analysis
Statistical testing was performed using SAS 9.4 (SAS Institute, Cary, NC, USA). Differences between detected parameters and ground truth and differences in measured CT numbers in ROIs were analysed using a one-sample two-tailed Student t test. The null hypothesis was that mean differences were equal to zero, which was rejected at a significance level of α = 0.05. Normality of data was assessed visually, using histograms and Q-Q scatterplots.
Application to sediment cores
Images of sediment core sections were corrected for their alignment using the algorithm described above. Detected misalignment, given as translation and rotation, was recorded. After alignment correction, sagittal reconstructions were calculated, showing images in parallel to the plane where the core is cut for sampling, thus allowing for planning of core sampling [7]. Sagittal images were reconstructed at a slice thickness of 5 mm with an increment of 5 mm, adding a digital ruler as an overlay to the image. Using the sagittal images, corresponding to a different depth below the surface, CWC fragments were identified and located, aiming to remove the samples from the sediment core with minimal destructive effect on neighbouring CWC fragments and surrounding sediment.