Extracting Morphological Changes in Nanocrystals Using in Situ Liquid Cell Microscopy
The goal of this project is to develop a Processing-Structure linkage for nanocrystal nucleation and growth. Using in situ liquid cell STEM imaging, nucleation and growth events can be tracked in real time with nanoscale resolution. The high energetics of the beam induce the nucleation and growth of the nanocrystals, so beam dosage is an important processing parameter. Another important processing parameter is the solution chemistry. We hope to develop a predictive linkage between these processing parameters and the resulting nanocrystal structure.
Our workflow plan is the following: 1) Use existing work in the literature to evaluate current analytical methodologies 2) Create Python scripts to create an automated image segmentation and binarization routine 3) Perform spatial statistics and PCA on the data using the PyMKS toolbox 4) Use the automated Python pipeline to develop the P-S linkage 5) Compare Python pipeline to analytical methods from the literature to show the strength of the methods presented in class
Having very little background in programming, we searched the literature for some help. We found two recent journal articles of groups doing similar work: Ievlev et al. (ACS Nano 2015) and Schneider et al. (Adv Struct Chem Imag 2016). Schneider et al. used a Mathematica script to perform their image processing, segmentation, binzarization, and analysis. We have been trying to run their routine using our data set to see how the results would look, but there have been many issues. The primary issue has been the binarization step. Their problem involved the growth of a single nanocrystal, which makes binarization quite simple. Our data set, however, shows the growth of many nanoparticles and therefore we need to make the routine more sophisticated in order to differentiate all of the nanocrystals.
Examining the above image from our data set, you can see regions where several particles are close together and the image is somewhat saturated. This makes simple global thresholding ineffective, as demonstrated by the image below. A more sophisticated binarization method will be required.
We are still working through steps 1 and 2. We hope to get the binarization figured out soon so that we can move forward, and we are optimistic that spatial statistics and PCA will follow much more easily once the binarization is handled.