Reducing Artifacts in BF and HAADF-STEM Images of Pt/C Fuel Cells using MBIR-ARAR

Electron tomography is a powerful tool for characterization of three-dimensional (3D) nanoscale materials and devices. Bright-field (BF) and high-angle annular dark-field (HAADF) scanning transmission electron microscopy (STEM) are two widely used imaging modes in electron tomography [1,2]. These imaging modalities have proven to be useful for characterizing the structure of carbon-supported platinum (Pt/C) electrocatalysts used in fuel cells, which are an important class of clean energy conversion systems. BF and HAADF-STEM images are typically simultaneously acquired due to the complementary information they contain, as BF-STEM is more suitable for characterizing carbon due to its ability to detect lighter elements, while HAADF-STEM is more suitable for characterizing platinum due to its sensitivity to atomic number. However, the quality of BF- and HAADF-STEM images is often compromised by various artifacts, such as local blurring due to abrupt contrast changes as well as missing wedge artifacts which can severely impact the accuracy of the reconstructed 3D images. The traditional methods of reconstruction, such as Filtered Back-projection (FBP) [3], Simultaneous Iterative Reconstruction Technique (SIRT) [4], and Model-Based Iterative Reconstruction (MBIR) [5], are often not suitable for removing these artifacts effectively. To address this challenge, we present a novel approach called MBIR with Artifact Reduction and Adaptive Regularization (MBIR-ARAR). This method leverages an adaptive regularization that adjusts the strength of the regularization in real-time based on the image data, and an artifact reduction technique to reduce the artifacts in the reconstructed images. Our approach distinguishes itself from traditional methods by incorporating an adaptive regularization, which allows the method to balance the trade-off between preserving high-frequency and low-frequency information in the image. Additionally, the artifact reduction technique further improves the quality of the reconstructed images. The MBIR-ARAR method is a two-step approach for reducing artifacts in BF-STEM and HAADF-STEM images of fuel cells and characterizing both carbon and platinum. The procedure is as follows. 1. Read the 3D projection data (Scp) 2. Perform MBIR reconstruction with a regularization suitable for high contrast features (Rcp) 3. Segment the platinum regions to get the platinum-only reconstruction image (Rp) 4. Forward project the metal-only image to get the platinum-trace sinogram (Sp) 5. Interpolate the original sinogram in the regions where platinum-trace sinogram is non-zero (Sc) 6. Perform a new MBIR reconstruction with the updated sinogram, Sc, and a regularization suitable for low frequency content of the carbon region (Rc) 7. Fuse the metal-only reconstruction from step 3 (Rp) with the carbon reconstruction from step 5 (Rc) to get the final reconstruction (Rcp) Unlike the standard MBIR reconstruction, the proposed approach will improve the reconstruction, and remove artifacts due to high-Z metal particles (such as shadowing around the dense metals and streaks), by separating the reconstruction of Pt and carbon. Moreover, an additional adaptive regularization step is applied, which adjusts the weighting of the regularization term relative to the fidelity term in the reconstruction algorithm to reduce artifacts while preserving important information about the sample. Simultaneous BF and HAADF electron tomography was performed on the Pt/C catalyst scraped from the cathode of a fuel cell. Example slices from the HAADF-STEM reconstruction are shown in Figure 1. The main result for the proposed MBIR-ARAR approach is shown in Figure 1d and 1h. It is evident that the proposed approach removes the streak artifact from the images and produce high quality reconstruction for both carbon and Pt, allowing for better visualization, segmentation and quantification. In Figure 2, example slices from both modes (BF and HAADF) demonstrate better performance of MBIR-ARAR in producing high-quality reconstruction of both Pt and Carbon with reduced streak artifact and noise. The reconstructed data was used to extract the average internal Pt size in the data which was 6.6 ± 1.45 nm, which is close to 6∼nm measured independently. In addition, the high quality of the reconstruction allows for accurate characterization of the internal porosity of the carbon support. In conclusion, the MBIR-ARAR approach offers a promising solution for improving the quality of reconstructed images in HAADF-STEM and BF-STEM data of Pt/C catalysts used in fuel cells. The proposed method can significantly improve the quality of the reconstructed 3D images and provide more accurate and reliable information about the distribution of Pt nanoparticles and pores within the carbon support. This information is essential for understanding the degradation mechanisms of fuel cells, which is critical for the development of more efficient and sustainable energy conversion and storage systems. The adaptive regularization and artifact reduction technique, along with its superior performance compared to traditional methods, make MBIR-ARAR a valuable tool for the reconstruction of tomographic tilt series for a variety of materials [6]. Example slices from the reconstructed HAADF volume from different perspectives. a-d) XY perspective; and e-h) XZ perspective. Comparison of a, e) FBP; b, f) SIRT; c, g) MBIR (no correction); d, h) MBIR with Artifact Reduction (AR) and adaptive regularization (MBIR-ARAR). Example slices from the reconstructed BF and HAADF volume from different perspectives (a-d, from XY and e-h from XZ) using MBIR (a,c,e,g) and MBIR-ARAR (b,d,f,h). The high quality of MBIR-ARAR for both BF and HAADF modes allows one to use either method independently to perform reconstruction and characterization of both pt and C, without requiring the other method