LTML-MIMIC-CXR

[Paper]

Long-tailed multi-label classification with noisy label of thoracic diseases from chest X-ray.

Requirements

• Pytorch
• Sklearn

Please download extra files from Google Drive: (We will public in future for secrecy) or in BaiduYun: 链接：https://pan.baidu.com/s/1wbCZwOpR7kvL9LN1b4A8xQ?pwd=2023 提取码：2023.

Please download MIMIC-CXR dataset from https://physionet.org/content/mimic-cxr/2.0.0/.

Quick start

Use our dataset

The information of LTML-MIMIC-CXR is saved in ./data/mimicall/LTML_MIMIC_CXR_label_.csv. The path of jpg images has been listed in LTML_MIMIC_CXR_label_.csv, which can be saved in ./data/mimicall/mimic.

The ./appendix contain the split of LTML-MIMIC-CXR in this repo.

appendix
  |--mimic
      |--class_freq.pkl
      |--class_split.pkl
      |--class_name.pkl
      |--img_id.pkl
      |--val.txt
      |--val.pkl
      |--test.txt
      |--test.pkl

Training

Traing ANR_LLA

CUDA_VISIBLE_DEVICES=0 python tools/train.py --config ./configs/mimic/LTML_resnet50_ANR_LLA.py

Traing ANR_LLM

CUDA_VISIBLE_DEVICES=0 python tools/train.py --config ./configs/mimic/LTML_resnet50_ANR_LLM.py

Testing

Test ANR_LLA

CUDA_VISIBLE_DEVICES=0 python tools/test.py --config './work_dirs/LTML_MIMIC_CXR_resnet50_ANR_LLA/LTML_resnet50_ANR_LLA.py'  --checkpoint './work_dirs/LTML_MIMIC_CXR_resnet50_ANR_LLA/latest.pth'

Test ANR_LLM

CUDA_VISIBLE_DEVICES=0 python tools/test.py --config './work_dirs/LTML_MIMIC_CXR_resnet50_ANR_LLA/LTML_resnet50_ANR_LLM.py'  --checkpoint './work_dirs/LTML_MIMIC_CXR_resnet50_ANR_LLM/latest.pth'

Model

The trained models is saved in ./work_dirs, which can be used to inference directly.