Rapid Image Segmentation Pipeline for Scanning Transmission Electron Microscopy (STEM) Analysis

NU 2024-111

INVENTORS
Alexandra Lauren Day
Ankit Agrawal*
Carolin Barbara Wahl
Roberto M dos Reis
Alok Nidhi Choudhary
Chad A Mirkin
Vinayak P Dravid

SHORT DESCRIPTION 
This invention introduces an automated image segmentation pipeline for fast, high-resolution 4D-STEM nanoparticle analysis.

BACKGROUND
Nanoparticle characterization via advanced scanning transmission electron microscopy (STEM) produces vast, high-resolution grayscale images that contain intricate details and substantial noise, making it challenging to isolate regions of interest efficiently. Current approaches often rely on uniform segmentation that divides each image into a fixed grid, creating an excessive number of sub-images regardless of feature distribution and thereby overwhelming storage and computational resources. Moreover, the absence of reliable ground-truth segmentation labels and the subtle intensity variations among nanoparticles of mixed compositions require methods that can adaptively differentiate relevant signal from background while maintaining interpretability. These challenges hinder both data acquisition speed and analysis accuracy, underscoring the need for more intelligent and computationally efficient segmentation strategies.  

ABSTRACT
A multi-step automated image segmentation pipeline is implemented for high-resolution 4D-STEM nanoparticle analysis, starting with downsizing STEM images to pixels and isolating particle regions via bounding box cropping. Subsequent steps include image sharpening, Gaussian blurring, adaptive Gaussian thresholding to generate a binary mask, and further cropping and intensity-based cutoff calculations that lead to additional binary masking. This automated pipeline aligns with visual inspection standards in >95% of instances and is projected to perform >9 times faster on average than current benchmarks.

APPLICATIONS 

  • Automated 4D-STEM Image Analysis Software
    • Rapid segmentation pipeline accelerates nanoparticle characterization in research laboratories

ADVANTAGES 

  • High Speed
    • Projected to be 9.81 times faster compared to traditional methods.
  • Enhanced Precision in Region of Interest Identification
    • Achieves 96% success rate in accurately segmenting particles and separating regions with different pixel intensities
  • Data Reduction Efficiency
    • Reduces number of segmentation boxes by factor of ~10, significantly reducing data to analyze.
  • Unsupervised Learning
    • Directly addresses segmentation challenges in grayscale nanoparticle images without the need for extensive ground-truth labeling

IP STATUS
Patent Pending
 

 

Patent Information: