Protist-Image-Analysis

Code, data, and firgures for analyzing protistan parasites of diatoms via micrographs taken under fluorescent light with CalcoFluor White (CFW) staining

• Problem (manual analysis): Too many images, not enough time

• Problem (automated analysis): Complicated images characteristics, background noise, strange shapes, inconsistent cell boundaries, infection dynamics

• Goal: Build a computationsl model that can accurately 1. Count cells, 2. Classify cell infection status by proportion of pixels in blue channel versus red channel

• Aproach(es):

  • Normalize RGB color channels so that intensity ranges are comparable
    • Threshold channels
  • Mask individual color channels from thresholded images
    • Combine red and blue channels for a full cell mask
  • Model cells with Hough Line Transform (HLT)
  • Cluster pixels with K-means clustering
  • Apply Sobel edge detection to identify cell boundaries
  • Cluter pixels using Gaussian Mixture Models
  • Combin HLT & K-means methods

• Model features:

  • Count individual cells
  • Calculate individual cell infected proportion via red & blue pixels within individual cell mask
    • Proportion outcomes determine infection classification

• Eventual Outcomes:

  • Extrapolate population dynamics
    • Changes in infection ratios over time series facilitates inference about shifts in mixed-community population dynamics

---

• Repository organization:
  • code:
    • benchmarking: "data_cleaning" includes code for cleaning raw csvs (result of running .ipynb files from ../code/final_models); model_evaluation.py evaluates performance using "./master_csvs/*.csv"; model_evaluation_statistics.csv is output from model_evaluation_F1.py
    • final_models: code for final versions of each 6 models
    • _deprecated: outdated code from various stages of model development
  • data:
    • test_images: set of ".jpg" files that models were trained on (note, models were only evaluated on AS*.csv files)
    • experimental_metadata.csv: metadata collected during lab experiment
    • model_evaluation_statistics.csv: output from "../code/benchmarking/model_evaluation_F1.py"
  • figures:
    • model_evaluation_statistics.csv: output from "../code/benchmarking/model_evaluation_F1.py"
    • video_explision.mp4: downloadable video of active parasitism
  • _deprecated:
    • outdated code and files from various stages of project development