Kasumi benchmark

R scripts used to analyse publicly available data, benchmark Kasumi against baselines and related methods and generate plots for the Kasumi paper.

For package requirements see utils.R. igraph version 1.5.1 is required to reproduce the exact results from the paper. Aternatively, a Dockerfile is provided for convenience.

Requires a development version of mistyR (>= 1.99.10) that can be obtained from jtanevski/mistyR. The Kasumi R package can be obtained from jtanevski/kasumi.

To reproduce the analysis of the DCIS dataset, run the following Python code in the Image Data/Segmetation_Outlines_and_Labels_Mendeley/ folder in order to generate the required csv files

import glob
import re
from skimage import io
from skimage.measure import regionprops_table
import pandas

for f in glob.glob("*_labels.tiff"):
    mask = io.imread(f)
    rp = regionprops_table(mask,  properties=('label', 'centroid'))
    pandas.DataFrame(rp).to_csv(re.findall("\d+", f)[0] + ".csv", index = False)