SCOUT

Summary

SCOUT (Scalable Object Understanding Technology) is a scalable workflow system for whole-slide microscopy analysis using Neural Networks.

Description

SCOUT (Scalable Object Understanding Technology) is a software system that finds all objects of interest in an entire microscope slide and provides various quality assurance metrics. It is designed to detect objects of size ~100² pixels in very large (~ 100K x 50K pixels) 10x magnification microscopy images. It consists of several software components, each of which may run on a dedicated machine or on shared hardware.

• A Processor, which performs all computational processing
  • Object detection using a Convolutional Neural Network
  • Image quality assessment
  • Slide fiducial mark detection and registration
• A Monitor, which monitors the microscope and copies images to a shared file system when each scan completes.
• An Executor, which monitors the shared file system and launches the Processor for the scanned images it finds.
• A Database, which stores the metrics and detections produced by the Processor.

The system is designed to support high-throughput, automated imaging on the scale of 100s of slides per day. This required creating a system that is robust and stable and which could properly recover from failures of its various hardware and software components. It also required creating a system with sufficient performance to keep up with the scanning rate of the microscope, both for image processing and data motion tasks.

This system was created in collaboration with NIST MML and OISM to automate manual inspection of large numbers of microscope slides in a laboratory environment. Prototypes of the system are currently deployed in NIST MML labs.

Major Accomplishments

• Developed prototype whole-slide microscopy analysis system
• Deployed system in NIST MML labs