Synchrotron microtomography reveals insights into the degradation kinetics of bio-degradable coronary magnesium scaffolds

Abstract

Bioresorbable magnesium scaffolds are a promising future treatment option for coronary artery stenosis, especially for young adults. Due to the degradation of these scaffolds (<1 year), long-term device-related clinical events could be reduced compared to treatments with conventional drug eluting stents. First clinical trials indicate a return of vasomotion after one year, which may be associated with improved long-term clinical outcomes. However, even after decades of development, the degradation process, ideal degradation time and biological response in vivo are still not fully understood. The present study investigates the in vivo degradation of magnesium scaffolds in the coronary arteries of pigs influenced by different strut thicknesses and the presence of antiproliferative drugs. Due to high 3D image contrast of synchrotron-based micro-CT with phase contrast (SR-μCT), a qualitative and quantitative evaluation of the degradation morphology of magnesium scaffolds was obtained. For the segmentation of the μCT images a convolutional network architecture (U-net) was exploited, demonstrating the huge potential of merging high resolution SR-μCT with deep learning (DL) supported data analysis. In total, 30 scaffolds, made of the rare earth alloy Resoloy®, with different strut designs were implanted into the coronary arteries of 10 domestic pigs for 28 days using drug-coated or uncoated angioplasty balloons for post-dilatation. The degradation morphology was analyzed using scanning electron microscopy, energy dispersive x-ray spectroscopy and SR-μCT. The data from these methods were then related to data from angiography, optical coherence tomography and histology. A thinner strut size (95 vs. 130 μm) and the presence of paclitaxel indicated a slower degradation rate at 28 d in vivo, which positively influences the late lumen loss (0.5 and 0.6 mm vs. 1.0 and 1.1 mm) and recoil values (0 and 1.7% vs. 6.1 and 22%).

Keywords: Artificial intelligence; DCB; Degradation; Inflammation; Magnesium scaffold; Paclitaxel; Resoloy.

Fig. 1

Illustrating the workflow from μCT data collection, phase retrieval and reconstruction, and image analysis.

Fig. 2

Illustration of the 2D U-Net architecture similar to Ronneberger et al. [18] The 2D U-Net used in this paper consists of three encoding stages. 3 × 3 convolutional filters were used, followed by non-linear activations (ReLU) and down-sampling by max-pooling. The number below each convolutional block denotes the number of feature maps. In the decoding stage, the volume was up-sampled by up-convolution. Finally, by a 1 × 1 convolution, the number of feature maps was reduced to 1 again. Skip connections between the encoding and the decoding path were realized by concatenations.

Fig. 3

Illustration of the performance of the segmentation: Virtual cross section through the 3D image obtained through synchrotron μCT. Left: grey values as obtained by tomographic reconstruction. Right: segmented data, with metallic Mg in light grey, slightly degraded Mg in red, and severely degraded Mg in blue.

Fig. 4

Illustration of the elemental distribution of two groups (G130_10% and G95_5%_DCB). For each group, one severely and one slightly degraded strut are shown. The detected elements (Mg, Ca, O, P, F and Dy) are depicted in different colors. Higher color intensity stands for more detected counts of this element. Scale bars: 50 μm.

Fig. 5

Carbon distribution (in blue) of a severely degraded strut according to G95_5%_DCB of Fig. 4.

Fig. 6

Segmented 3D-renderings of all scaffolds analyzed with SR-μCT. The remaining Resoloy material, the degraded and the severely degraded regions, are marked, respectively, in grey, red, and blue.

Fig. 7

A: SR-μCT degradation analysis for all groups. Hatched bars: total volume of degraded material (right y-axis); filled bars: proportional degradation with respect to the total scaffold volume (left y-axis); dotted bars: volume severely degraded material (right y-axis). B: Spatially resolved degradation analysis of scaffold struts (see Fig. 8 of appendix). Hatched bars: luminal degradation volume; filled bars: lateral degradation volume; dotted bars: abluminal degradation volume. C: Results of Recoil and LLL derived from angiography for all groups. Bars: LLL values. Circles: Recoil values (error bars in red), statistically significant differences are marked with an asterisk (*). D: Correlation between the volumes of severely degraded scaffold-parts and the corresponding inflammation scores for each scaffold of all groups. E: Inflammation scores per scaffold-group.

Fig. 8

Left: Division of each strut into segments which have a certain orientation towards a reference point, calculated as the center of mass of all struts in the 2D plane. The lines connecting the reference point with the centers of mass of each individual struts crosses the region which is assumed as the luminal region. From this starting line, different segments are defined, which have the same angle range. Right: Example showing how each strut is divided into 6 different segments. Here, all segments having the numbers 1 and 2 are approximately in the luminal region whereas segments 5 and 6 are rather in the abluminal region. By using a smaller angle range, each strut can be divided into more segments, allowing a closer approximation to the luminal and abluminal regions. In this paper, each strut has been divided into 18 segments.

Fig. 9

Overlay of a laboratory processed histology image (colored) and the 2D-image of the SRμ-CT (grey).Please note, that the contrast has been adjusted to show the grey-values corresponding to the soft tissue.