Skip to content

Home

Jump to bottom
Sneha-Kondur edited this page May 19, 2019 · 14 revisions

Objective:

  • Provide an additional benchmark towards validating automated classifiers, by assessing a sampling method, that lies in the intersection of microscopy and image sampling.
  • Assess how “good” are the microscopy sampling methods, by comparing it against an averaged abundance estimate from the in vitro analysis.
  • Understand any latent variables or noise sources that occur across both sampling methods.

Rationale and background information:

In order to validate automated classifiers, we need to benchmark this sampling method against traditional sampling methods, like microscopy. Due to the large differences in space and time between the two methods, we are conducting an “in the lab” sampling to address any possible sources of errors between the two methods. Using a clone of the in situ imaging system with the same water, that is collected for microscopy will reduce the error. While it is possible that these methods may not agree very well, we can gain confidence that the error comes from something inherent to the imaging system and not just a difference in the way the water was collected.

Design of the Experiment

Experiment consists of collecting bucket sample and conducting both a microscopy and in vitro analysis on the sample to obtain an abundance estimate.

Experimental Condition:

Subsample imaged is consistent with the sample volume counted under the microscope.

Methodology:

  1. Collect a bucket of water sample from the surface of the ocean (Location: Scripps Pier)
  2. Conduct Microscopy Analysis:
  • Concentrate the water sample using Utermohl method
  • Draw a subsample of 2.5ml and place it on a cover slide
  • Examine under an inverted microscope and estimate the plankton concentration in cells/ml
  1. Conduct In Vitro Analysis:
  • Obtain 3 liter bucket sample from the water sample collected
  • Perform the lab imaging on each subsample
  • Deploy the trained classifier on the images and obtain the class predictions for each sample
  • Validate XXX predicted images for ground truth
  • Estimate the plankton concentration for each subsample
  • Calculate the mean plankton concentrations across the subsamples
  1. Automatically generate plot of In-Vitro abundance estimates against the Microscopic abundance estimates and observe the correlation among the data points.

Clone this wiki locally