Lung Nodule Detection in Pytorch

A One-stage Method for lung nodule detection in LUNA16 dataset

Scripts:

• Would be executed:

  • config_training.py: set filepaths
  • data_detector: generate data loader during training and testing
  • prepare.py: LUNA16 dataset preprocessing
  • main_detector_recon.py: model training and inference
  • GenerateCSV.py: Generate result.csv for computing CPM
  • noduleCADEvaluaionLUNA16.py: Compute CPM of .csv
  • FROC_CPM.ipynb: Plot FROC curve

• Others

  • data_detector.py: generate data loader during training and testing
  • preprocess.py: some preproceesing-related codes
  • layers.py
  • loss.py
  • split_combine.py (At Testing stage)
  • utils.py

Requirements:

• Python 3.6
• torch 0.4.1
• torchvision 0.2.0
• SimpleITK
• scikit-image

Files:

• LUNA.json: Every Case ID stored in .json for make_dataset.py
• json: Directory contains train/val/test.json (Case ID) respectively
• Download LUNA16 dataset from Grand Challeng: https://luna16.grand-challenge.org and save at the following filepaths
  ./data/LUNA16/allset: all .raw and .mhd of LUNA16 data
  ./data/LUNA16/seg-lungs-LUNA16: all .zraw and .mhd of LUNA16 mask

How to Do step by step:

• Preprocessing for LUNA16

  • python prepare.py
  • output file path: config_training -> config[preprocess_result_path]

    Output: id_clean.npy & id_label.npy (for training) ; id_extendbox.npy & id_mask.npy & id_origin.npy & id_spacing.npy (for vox2world) 
  

• Start training and testing

  • training

    python main_detector_recon.py --model OSAF_YOLOv3 -b [batch_size] --epochs [num_epochs] --save-dir [save_dir_path] --save-freq [save_freq_ckpt] --gpu '0' --n_test [number of gpu for test] --lr [lr_rate] --cross [1-5 set which cross_data be used]
    eg: python main_detector_recon.py --model OSAF_YOLOv3 -b 2 --epochs 100 --save-dir OSAF_YOLOv3_testcross1 --save-freq 1 --gpu '0' --n_test 1 --lr 0.001 --cross 1 
  

  • testing

    python main_detector_recon.py --model OSAF_YOLOv3 --resume [resume_ckpt] --save-dir [] --test 1 --gpu '0' --n_test [] --cross []
    eg: python main_detector_recon.py --model OSAF_YOLOv3 --test 1 --cross 1 --resume "./results/OSAF_YOLOv3_testcross1/1.ckpt" --save-dir "OSAF_YOLOv3_testcross1" --gpu 0
  

• Compute CPM (After test all 5 fold)

  • Generate result.csv

    python GenerateCSV.py
  

    output: OSAF_YOLOv3_80_all.csv
  

  • Then compute CPM and save related .png and .npy

    python noduleCADEvaluaionLUNA16.py (Remember to modify the filepath in noduleCADEvaluation.py)
  

    output: print(csv_name, CPM, seven_sensitivities@predefined_fps) and save ./CPM_Results_OSAF_YOLOv3_80/_.npy & _.png
  

• Plot FROC Curve

    Execute FROC_CPM.ipynb
  

    output: FROC_OSAF_YOLOv3.png
  

• How to transform voxel coord of pbb into world voxel of 3D CT ?

  • You can refer to GenerateCSV.py: How to transform pbb -> pos