Variability of Object Measurements from 2D Microscopy Images

Summary

This project addresses object measurements from 2D microscopy images. Object measurements (called image features) vary in terms of theoretical formulas for the same image feature, the physical units used to represent pixel-based measurements, the definitions of objects in images (called regions of interest or ROIs), algorithmic implementations, the number of exposed parameters to a user, and programming languages. Our motivation is to introduce readers to image-based object measurements and to quantify numerical variability of (a) image features and (b) feature-based classification outcomes. The variability is evaluated across widely-used image feature extraction libraries to highlight the sources of variability when deriving image-based scientific conclusions.

Experimental Results

The experimental results were obtained as follows:

• The inputs come from the persistent terascale image collection available for browsing and downloading from here. The images and classification labels were documented in the publication below by Bhadriraju et al.
• Five open-source image feature extraction libraries were considered. The versions of the libraries are documented here.
• The libraries were integrated into a client-server system called Web Image Processing Pipeline (WIPP) and applied to input images. The Docker container of WIPP software is available for installation from here.
• The numerical features extracted by WIPP were analyzed to identify significantly different values and to measure the impact of feature variability on classification outcomes. The description is available in the publication below by Simon et al. and Bajcsy et al. and the Matlab code for the analyses is available here.
• In order to deliver interactive interfaces to traceable results of image feature variability study, we deployed a set of interactive graphs and tables at the web pages provided below. The web pages allow readers to identify interactively image features that differ across open-source feature extraction libraries and to trace the numerical feature values to the original (persistent) images used for quantification.
  
  
  • Variability of Mean Intensity Image Feature Over Regions of Interests (ROI - Cell Colonies)
    
  • Variability of Perimeter Image Feature Over Regions of Interests (ROI - Cell Colonies)
    
  • Variability of GLCM-based Texture Contrast Image Feature Over Regions of Interests (ROI - Cell Colonies)
    
  • Image Feature Variability Results
    
  • Impact of Image Feature Variability on Classification
    

Data Downloads

All numerical features extracted by WIPP are available for downloading below.

• CSV file of all numerical features

All analyzed results presented in the interactive graphs and tables are also available for downloading below.

• CSV file of feature variability
• CSV file of feature variability impact on classification
• CSV file of mean intensity features across five libraries
• CSV file of perimeter feature variability
• CSV file of GLCM-based texture contrast feature variability

Publications:

• Large-scale Time-lapse Microscopy of OCT-4 Expression in Human Embryonic Stem Cell Colonies (pdf)
  Kiran Bhadriraju, Michael Halter, Julien Amelot, Peter Bajcsy, Joe Chalfoun, Antoine Vandecreme, Barbara Mallon, Kye-yoon Park, Subhash, and Anne Plant
  Elsevier Stem Cell Research, Volume 17, Issue 1, July 2016, Pages 122-129, DOI: 10.1016/j.scr.2016.05.012
• Do We Trust Image Measurements? Variability, Accuracy and Traceability of Image Features (pdf)
  Mylene Simon, Joe Chalfoun, Mary Brady, and Peter Bajcsy,
  IEEE International Conference on Big Data, Washington DC, December 5-8, 2016
• Chapter 7: Object Measurements from 2D Microscopy Images, in the book entitled Modern Computer Vision for Microscopy Image Analysis
  Peter Bajcsy, Joe Chalfoun, Mylene Simon, Marcin Kociolek, and Mary Brady
  Editor: Mei Chen, Elsevier Publisher, 2018 (under review)

Contributors:

ITL-Software and Systems Division

Information Systems Group

• Peter Bajcsy
• Joe Chalfoun
• Mylene Simon
• Mohamed Ouladi
• Antonio Cardone
• Antoine Gerardin
• Philippe Dessauw
• Mary Brady

Lodz University of Technology, Poland

• Marcin Kociolek

Contact:

Peter Bajcsy
peter.bajcsy@nist.gov
Phone: 301.975.2958