Disclaimer

This is not an official Verily product.

HIPI

HIPI: Spatially Resolved Multiplexed Protein Expression Inferred from H&E WSIs

Installation

• Create conda environment:

conda env create -f environment.yml

• Install SimpleITK

Data preprocessing

• Download H&Es and CyCIF files from Lin et al. 2023

bash download_CRC_Lin23.sh

• Align H&Es and CyCIF images using the preprocess/align_hne_cycif.sh script

bash align_hne_cycif.sh

• Extract image tiles using the preprocess/split_hne_to_tiles.sh script

bash split_hne_to_tiles.sh

• Create a tile feature dataframe table using the preprocess/create_tile_measurement_table.py script

python create_tile_measurement_table.py

All scripts use the current directory unless otherwise specified.

Training

• Configuration files for model and training paramaters are in configs folder
• To train HIPI with default parameters run the following (after adjusting the log dir and number of gpus):

python cycif_train_main.py --base configs/ssl_vit_mlp8_16channels.yaml --logdir logs -t --gpus 0,1,2,3,

Prediction

Run the cycif_eval_main.py script with the model configuration and checkpoint to evaluate. You can choose which dataset to evaluate from the config file (train, validatiob, test) or overwrite in the command line.

python cycif_eval_main.py --cfg_file configs/ssl_vit_mlp8_16channels.yaml --ckpt_file "${ckpt_file}" --test_csv "${test_csv}" --num_workers 4 --out_path "${out_path}" --batch_size 512 --device cuda:0 --datasets "${dataset}"

Trianed model

Weights for the tranied model are in models/HIPI_model.ckpt