Add readme.

Renamed several variables.

Scripts/prepare_label-mps.py

@@ -11,8 +11,8 @@
 error_json = Path(r"\\mega\homesall\clwang\Data\jsph_lung\YHBLXA_YXJB\error_dict.json")
 SAVE_PATH = Path(r'\\mega\yliu\Data\pn_cls_data\MPS')
 
-refer_train_path = Path(r"\\mega\yliu\Data\medlp_exp\classification\jsph_mps\resnet50-BCE-BN-sgd-plateau-0622_1914-lr0.1-CTRList\train_files.yml")
-refer_valid_path = Path(r"\\mega\yliu\Data\medlp_exp\classification\jsph_mps\resnet50-BCE-BN-sgd-plateau-0622_1914-lr0.1-CTRList\valid_files.yml")
+refer_train_path = Path(r"\\mega\yliu\Data\strix_exp\classification\jsph_mps\resnet50-BCE-BN-sgd-plateau-0622_1914-lr0.1-CTRList\train_files.yml")
+refer_valid_path = Path(r"\\mega\yliu\Data\strix_exp\classification\jsph_mps\resnet50-BCE-BN-sgd-plateau-0622_1914-lr0.1-CTRList\valid_files.yml")
 
 _FEATURE_KEY = ['lobulation', 'spiculation', 'Relation to bronchus', 'Relation to Vessel']
 FEATURE_KEY = [

Scripts/prepare_label_mvi_MB.py

@@ -4,7 +4,7 @@
 import nibabel as nib
 from ..utils import check_dir, get_items_from_file
 from tqdm import tqdm
-from medlp.data_io.base_dataset.classification_dataset import BasicClassificationDataset
+from strix.data_io.base_dataset.classification_dataset import BasicClassificationDataset
 from monai.data import CacheDataset, PersistentDataset, Dataset
 from monai_ex.transforms import *
 

datasets/__init__.py

@@ -1,4 +1,4 @@
-from medlp.utilities.registry import DatasetRegistry
+from strix.utilities.registry import DatasetRegistry
 
 CLASSIFICATION_DATASETS = DatasetRegistry()
 SEGMENTATION_DATASETS = DatasetRegistry()

datasets/jsph_mps_datasets.py

@@ -2,8 +2,8 @@
 import numpy as np
 from datasets import CLASSIFICATION_DATASETS
 from scipy.ndimage import generate_binary_structure, binary_erosion
-from medlp.data_io.base_dataset.classification_dataset import BasicClassificationDataset
-from medlp.configures import config as cfg
+from strix.data_io.base_dataset.classification_dataset import BasicClassificationDataset
+from strix.configures import config as cfg
 
 
 from monai_ex.data import CacheDataset, PersistentDataset

datasets/jsph_mvi_datasets.py

@@ -2,7 +2,7 @@
 import numpy as np
 
 from datasets import CLASSIFICATION_DATASETS
-from medlp.data_io.base_dataset.classification_dataset import BasicClassificationDataset
+from strix.data_io.base_dataset.classification_dataset import BasicClassificationDataset
 from monai_ex.data import CacheDataset, PersistentDataset
 from monai_ex.transforms import *
 

datasets/lidc_idri_datasets.py

@@ -2,8 +2,8 @@
 import numpy as np
 
 from datasets import CLASSIFICATION_DATASETS
-from medlp.data_io.base_dataset.classification_dataset import BasicClassificationDataset
-from medlp.configures import config as cfg
+from strix.data_io.base_dataset.classification_dataset import BasicClassificationDataset
+from strix.configures import config as cfg
 
 from monai_ex.data import CacheDataset, PersistentDataset
 from monai_ex.transforms import *

main.py

@@ -3,11 +3,11 @@
 
 from train_v2 import train_core
 from utils import check_dir, get_items_from_file, PathlibEncoder
-from sklearn.model_selection import train_test_split, KFold 
+from sklearn.model_selection import train_test_split 
 from datasets import CLASSIFICATION_DATASETS
 
 @click.command('train')
-@click.option("param-path", type=click.Path(), default=Path.home()/"Code"/"ExPN-Net"/"param.list")
+@click.option("--param-path", type=click.Path(), default=Path.home()/"Code"/"ExPN-Net"/"param.json")
 def train(param_path):
     confs = get_items_from_file(param_path, format='json')
     random_seed = 42
@@ -17,7 +17,7 @@ def train(param_path):
     else:
         save_dir_name = f"{time.strftime('%m%d_%H%M')}-{confs['net']}-lr_{confs['lr']}-{confs['loss_name']}-{confs['optim']}{confs['postfix']}"
     confs['out_dir'] = check_dir(f"/homes/yliu/Data/pn_cls_exp/{confs['dataset_name']}/{save_dir_name}")
-    with open(confs['out_dir']/'param.list', 'w') as f:
+    with open(confs['out_dir']/'param.json', 'w') as f:
         json.dump(confs, f, indent=2, cls=PathlibEncoder)
 
     dataset_type = CLASSIFICATION_DATASETS[f"{confs['dimensions']}D"][confs['dataset_name']]['FN']

nets/HESAM.py

@@ -17,7 +17,7 @@
 from monai.networks.nets.basic_unet import Down
 
 # from monai.networks.nets import DynUNet
-from medlp.models.cnn import DynUNet
+from strix.models.cnn import DynUNet
 from monai_ex.networks.layers import Act, Norm, Conv, Pool
 from blocks.basic_block import TwoConv, UpCat, ResidualUnit, SimpleASPP
 from nets.utils import save_activation

nets/hesam_aag.py

@@ -9,9 +9,9 @@
 from monai.networks.nets.basic_unet import Down
 
 # from monai.networks.nets import DynUNet
-from medlp.models.cnn import DynUNet
+from strix.models.cnn import DynUNet
 from monai_ex.networks.layers import Act, Norm, Conv, Pool
-from medlp.models.cnn.layers.anatomical_gate import AnatomicalAttentionGate as AAG
+from strix.models.cnn.layers.anatomical_gate import AnatomicalAttentionGate as AAG
 from monai.networks.blocks.dynunet_block import get_conv_layer
 from torch.nn.modules.activation import ReLU
 from blocks.basic_block import TwoConv, UpCat, ResidualUnit, SimpleASPP

nets/resnet_agg_sam.py

@@ -5,9 +5,9 @@
 import torch.nn as nn
 from typing import Callable
 import numpy as np
-from medlp.models.cnn.nets.resnet import ResNet, BasicBlock, Bottleneck
-from medlp.models.cnn.layers.anatomical_gate import AnatomicalAttentionGate as AAG
-from medlp.models.cnn.utils import set_trainable
+from strix.models.cnn.nets.resnet import ResNet, BasicBlock, Bottleneck
+from strix.models.cnn.layers.anatomical_gate import AnatomicalAttentionGate as AAG
+from strix.models.cnn.utils import set_trainable
 from monai_ex.networks.layers import Act, Norm, Conv, Pool
 from blocks.basic_block import UpCat, ResidualUnit
 from nets.HESAM import MultiChannelLinear

nets/unet_cls.py

@@ -7,9 +7,9 @@
 import numpy as np
 
 # from monai.networks.nets import DynUNet
-from medlp.models.cnn import DynUNet
-from medlp.models.cnn.blocks.dynunet_block import UnetBasicBlock, UnetUpBlock
-from medlp.models.cnn.layers.anatomical_gate import AnatomicalAttentionGate as AAG
+from strix.models.cnn import DynUNet
+from strix.models.cnn.blocks.dynunet_block import UnetBasicBlock, UnetUpBlock
+from strix.models.cnn.layers.anatomical_gate import AnatomicalAttentionGate as AAG
 from monai_ex.networks.layers import Act, Norm, Conv, Pool
 from monai.networks.blocks.dynunet_block import get_conv_layer
 from monai.networks.nets.basic_unet import Down

readme.md

@@ -0,0 +1,25 @@
+## Official repo of the paper "Towards Reliable and Explainable AI Model for Pulmonary Nodule Diagnosis"
+---
+
+### For the model training:
+`python main.py --param-path /your/parameters_file.json` \
+An simple example of parameter file is shown in [param.json](param.json).
+
+### For the model testing:
+`python test.py` \
+Notice: change the hard-coded variable `model_path` and `test_json` for your project. 
+
+### Requirements
+You need to pre-install the following packages for this program:
+- pytorch
+- tb-nightly
+- click
+- tqdm
+- numpy
+- scipy
+- scikit-learn
+- nibabel
+- pytorch-ignite
+- [strix](https://github.com/Project-Strix/Strix)
+- [monai_ex](https://github.com/Project-Strix/MONAI_EX)
+- [utils_cw](https://gitlab.com/ChingRyu/py_utils_cw)

test.py

@@ -13,8 +13,8 @@
 from utils import DrawROCList, get_network, check_dir, get_items_from_file
 from metrics import AUC_Confidence_Interval, save_roc_curve_fn
 
-model_path = Path("/homes/yliu/Data/pn_cls_exp/lidc-paper_new-sphericity_prob/1215_2358-raw_hesam_agg-slice_1-lr_0.01-plateau-multi-BCE-sgd-sum-parallel-wooffset-2nd/Models/SBAA/BestModel@42with12.817.pt")
-config_path = model_path.parent.parent.parent/'param.list'
+model_path = Path("/homes/yliu/Data/pn_cls_exp/BestModel.pt")
+config_path = model_path.parent.parent.parent/'param.json'
 configures = get_items_from_file(config_path, format='json')
 
 IMAGE_KEY = 'image'
@@ -30,7 +30,7 @@
 device = torch.device("cuda")
 
 
-test_json = Path("/homes/yliu/Data/pn_cls_data/LIDC-IDRI/all_datalist_int_equals3.json")
+test_json = Path("/homes/yliu/Data/LIDC-IDRI/test_datalist.json")
 dataset_name = configures['dataset_name']
 dimensions = configures['dimensions']
 dataset_type = CLASSIFICATION_DATASETS[f'{dimensions}D'][dataset_name]['FN']