Web Image Processing Pipeline For Interactive Discoveries

Summary

Several advanced microscopes have the capability to acquire overlapping image tiles automatically. Such automated acquisitions provide a way to image a large spatial coverage of a specimen, to take measurements at multiple physical length scales, and to learn the underlying models that describe the specimen. There is a need to assist imaging scientists with computational solutions that convert raw image tiles to calibrated, stitched, and viewable Giga-pixel images with interactive and traceable measurements and modeling tools. In other words, a need for a solution to go From Image Tiles to Web-Based Traceable Measurements and Interactive Modeling in One Stop (denoted as web image pipeline for interactive discoveries).

Description of Web Solutions

The three modules below enable imaging scientists to focus on the domain science while using the web-based computational tools to deal with videos of gigapixel images frequently reaching terabyte-sized volumes. The functionality is accessible via web in order to facilitate geographically distributed collaboration and utilize distributed computational resources for CPU-intensive processing. After uploading image collections, all data are stored in the same database so that they are accessible by all modules.
A user can decide what computations to execute and in which order while visually inspecting intermediate results, performing optimization of parameters, and following the computational and data provenance traces to reproduce final results.

1. Web Image Processing

The Web Image Processing module provides functionality for uploading a collection of image tiles and processing them into calibrated, stitched, and segmented images. In addition, it generates a hierarchical pyramid representation with a Deep Zoom visualization and on-demand processing of gigapixel images.

View details »

2. Web Image Feature Extraction

The Web Image Feature Extraction module enables extracting traceable image features from raw images and their corresponding segmented masks. The image features span intensity, spatial and texture characteristics of objects defined in a segmented 2D mask. The results are a table of image features and a set of downloadable hyperlinked digital artefacts that were used to compute the image features.

View details »

3. Web Statistical Modeling

The Web Statistical Modeling module is under construction.

View details »

Major Accomplishments

The web system prototype has been deployed inside of NIST for the use by NIST scientists. Publications:

• Peter Bajcsy, Antoine Vandecreme, Julien Amelot, Joe Chalfoun, Michael Majurski, and Mary Brady, “Enabling Stem Cell Characterization from Large Coverage Microscopy,” IEEE Computer, 2016: vol. 49, no. 7, pp. 70-79, URL
• Antoine Vandecreme, Tim Blattner, Michael Majurski, Peter Bajcsy, Keana Scott, and John Henry J. Scott, “From Image Tiles to Web-Based Interactive Measurements in One Stop,” Microscopy and Microanalyses, Portland, OR, August 2-6, 2015, URL