Stingray : A Practical Guide

 


Introduction

What does it do?

  • Stingray is designed to find and classify objects or particles from holograms
  • Specifically, Stingray was designed using oceanic phytoplankton samples collected with 4Deep’s instruments. However, Stingray can be used for identification and classification for microorganisms in any liquid environment, including yeast in beverages, or particles in river water quality testing

Why would I use Stingray?

  • Stingray will be crucial for anyone who needs to distinguish objects from one another. This includes distinguishing different species within a sample. For example, if a user takes a sample from the ocean, Stingray will be able to group objects into different groups

How does it work?

  • First, Stingray analyzes holograms, which are 2D images that contain 3D information. Essentially, a hologram contains information in a volume, so think of a hologram as a cube (or a cone/pyramid). Stingray cuts through the volume in slices or planes (the number of slices is inputted by the user) and checks to see if there are any particles in the plane. If the particle meets the criteria (again, inputted by the user), it is added to the database as an object
  • These particles are then sorted into groups or taxon manually by the user
  • Good particles from these groups are then used to train a classifier. Essentially, by telling the classifier these are taxon A and these are taxon B, it can differentiate between two taxon

Definitions:

Database: located in the main window, it holds all of the objects that Stingray extracts from holograms
Object: loosely refers to any particle selected by Stingray that is added to the database
Taxon: group. It does not have to be species specific, it can be as simple as “good particles” and “bad particles” or “circle” and “rod”. Essentially it is a sorted object
Sample: an object that has been assigned a taxon and will be used to train classifiers
Classifier: the algorithm that is trained by samples. The classifier will sort objects into taxon


 


The Steps:

1. Get some Holograms
  • First, holograms are needed to analyze
    • Collect holograms using a 4Deep microscope and Octopus. Refer to their respective User Guides for full details
2. Get Objects from Holograms
  • Run holograms through Stingray to add objects to the database
    • You’ll have to input settings such as particle size and threshold
    • For the S6, these settings will work:
      • From: 7000 μm
      • Planes: 400
      • To: 22000 μm
      • Particle sizes: 20 μm – 1000 μm
      • Threshold: Auto
3. Sort Objects into groups

TIP: Create a “Noise” particle group for out-of-focus particles

  • Assign taxon to objects
    • You’ll need to pick objects from the database, and assign it to a taxon or group
    • In order for Stingray to know how to classify objects, it will need to know which objects belong in each group
4. Pick Objects to Train Stingray
  • Save objects as samples
    • From the newly grouped objects, save good objects as samples
    • The samples will be used to train the classifier
5. Train

TIP: You can stop training if it is failing or if it has passed instead of waiting until the end

  • Train classifier
    • Using the samples, the classifier will need to be trained
    • The default settings to train a classifier include:
      • Iterations: 200
      • Batch size: 32
      • Learning rate: 0.001
6. Use Classifier on Objects

TIP: You may need more than one classifier

  • Use classifiers
    • Use the classifier on objects in the database
    • Use the classifier on holograms:
      • Simultaneously sort objects into taxon as Stingray finds objects