Exercise: machine learning for segmentation

Duration ~40 min  ·  Session Day 2, Python notebooks

Cell segmentation with a pixel classifier and Cellpose.

Data: data/bbbc020/ field 15min_1, and the Weka export in data/bbbc020/weka/15min_1_cells_probability_2.tif. If you exported your own in E4, use those instead: point WEKA at results/weka/ and the rest will work unchanged.

import tifffile
import os; os.environ["DISPLAY"] = ":1"

from course import DATA, show

Task 1: Load the pixel classifier output and perform a watershed segmentation.

Use a nuclei segmentation to seed the watershed segmentation.

Hint 1: First threshold the Weka output appropriately.

Hint 2: In case the nuclei are not well separated, you could perform a distance transform based watershed segmentation on the nuclei, to then serve as seed for the watershed segmentation of the pixel classifier output.

# --- your turn ---
...

Task 2: Segment the same image with Cellpose, once using only the cell channel, and once using both the cell and nuclei channels.

Try varying the diameter parameter of Cellpose to see whether it improves the segmentation. It should be set to the approximate diameter of the cells in pixels.

# --- your turn ---
...

Task 3: Segment the cell cytoplasm

Hint: Use the result of applying different Cellpose models on different channel combinations for this.

# --- your turn ---
...