Deep learning-based quantification of eosinophils and lymphocytes shows complementary prognostic effects and interplay in patients with colorectal cancer

Code to generate the results from the published paper:

Deep learning-based quantification of eosinophils and lymphocytes shows complementary prognostic effects and interplay in patients with colorectal cancer

How to use:

calculate_scores.py \
    --wsi_path "list_of_cases.txt" \
    --nuclei_results "/path-to/hover_next_results/cohort/" \
	--segmentation_results "/path-to/srma_results/wsi" \
    --nprocs 16 \
    --output_path "/path-to/output/cohort.csv" \
	--array_tasks 8 \
	--array_id $SLURM_ARRAY_TASK_ID

The output file will be one or multiple CSV files with a row for each WSI containing scores for each cell type in tumor front and center (if estimation is available and worked correctly).

This work heavily relies on HoVer-NeXt and the C2R adaptation (unpublished) of SRMA.

Citation

If you are relying on results from the paper or use the code to generate similar results, please consider citing the our work:

@article{baumann2025deep,
  title={Deep learning-based quantification of eosinophils and lymphocytes shows complementary prognostic effects in colorectal cancer patients},
  author={Baumann, Elias and Lechner, Sophie and Krebs, Philippe and Kirsch, Richard and Berger, Martin D and Lugli, Alessandro and Nagtegaal, Iris D and Perren, Aurel and Zlobec, Inti},
  journal={npj Precision Oncology},
  volume={9},
  number={1},
  pages={1--11},
  year={2025},
  publisher={Nature Publishing Group}
}