iseekwonderful / RSNA-2023-Abdominal-Trauma-Detection-4th-Place-Code

RSNA 2023 Abdominal Trauma Detection

Below you can find an outline of how to reproduce our solution for the RSNA 2023 Abdominal Trauma Detection. If you run into any trouble with the setup/code or have any questions please contact us:

• @sheep: [e-mail]

1.INSTALLATION

• Ubuntu 20.04.5 LTS
• CUDA 11.8
• Python 3.9.17
• training requires a 4 card with 32G or 48G VRAM.
• Please install anaconda first and follow the cmd below to setup environment

$ conda create --name rsna_env python==3.9
$ conda activate rsna_env
$ conda install -y pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 pytorch-cuda=11.8 -c pytorch -c nvidia
$ conda install -y matplotlib seaborn pandas scikit-learn tensorboard
$ pip install opencv-python-headless tqdm timm nibabel scikit-image albumentations python-gdcm pydicom SimpleITK path transformers
$ pip install -U pyyaml==5.1.1
$ pip install packages/segmentation_models.pytorch/ 

2.DATA

1. The convertion of dicom to pngs and the prediction of TotalSegmentator takes very long time and could be downloaded from
   • https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/427427
   • https://www.kaggle.com/datasets/steamedsheep/ts-body-and-organ
   • As for the TotalSegmentator's mask prediction, I use following CMD for each series:

#!/bin/bash

# Base directory
BASE_DIR="/media/data/rsna/input/train_images"

# Loop through each subdirectory
for dir in ${BASE_DIR}/*/*; do
  if [ -d "${dir}" ]; then
      # Extract the directory structure to replicate for the output
      RELATIVE_PATH=${dir#${BASE_DIR}/}

      # Create output directory if it doesn't exist
      mkdir -p "segmentations/${RELATIVE_PATH}"
      mkdir -p "segmentations_organ/${RELATIVE_PATH}"

      # Call TotalSegmentator
      TotalSegmentator -i "${dir}/" -o "segmentations/${RELATIVE_PATH}" -ta body
      TotalSegmentator -i "${dir}/" -o "segmentations_organ/$output" --fast --roi_subset spleen kidney_right kidney_left liver small_bowel
  fi
done
  

2. go to pp folder and run all py file

python 0.2d_mask.py  
python 1.segment_train.py  
python 2.body_segment_train.py  
python 3.rescale.py

3.TRAIN

• The organ folder is the project folder and model weights and logs will be save to results folder, plsease run follow cmd to train 4 models.

./train.sh

3.INFERENCE

• Since this is a code competition and the inference env already made public at kaggle, please refer to submission notebook

Directory Structure

rsna_rep/
    ts/
    results/
    organ/
        basic_train.py
        batch_train.sh
        main.py
        eval.py
        train_net.py
        dist_train.py
    pp/
        0.2d_mask.py
        1.segment_train.py
        3.rescale.py
        2.body_segment_train.py
    input/
    packages/