Histological images for TMB index prediction

---

Install dependencies

Run requirements.txt script to install all the dependencies.

# install openslide
sudo apt-get install openslide-tools
sudo apt-get install python-openslide
# git clone software to ${SRC} and then run:
pip3 install -r ${SRC}/requirements.txt

Prepare work

1、Preparation

# annotated slide home
${DATASET}/lung/images
# class label home i.e., reg_tmb.csv
${DATASET}/lung/labels
# template image home i.e., template.png
${DATASET}/template
# get tile image home from raw slides
mkdir -p ${HOME}/tiles
mkdir -p ${HOME}/tiles_color_normalized
mkdir -p ${HOME}/model

A snapshot view of the directory structure is as follows (${DATASET}/lung/images):

2、Prepare annotated svs slide dataset

Compute tile images, more details can be found in config/config.py you should specify ratio_train and ratio_test in your experiments.

# dataset prepared
cd ${SRC}
# transmit parameter
sh ./HE_Auto.sh /tmp/data/lung/images /tmp/data/tiles /tmp/data/tiles_color_normalized /tmp/data/lung/labels/reg_tmb.csv ../asset/Template.png

---

Training

Run the following command to start training more details can be found in config/config.py as well as you should customize those parameters by yourself.

First of all, start a new tensorboard session for training investigation:

tensorboard --logdir=runs --bind_all

After then, run train.py script and you can start training if everything is okay.

python ${SRC}/train.py

Finally, model can be learned in ${DATASET}/model folder.

---

Inference

Run test.py script for inference:

python ${SRC}/test.py