Merge pull request #15 from RomainUSA/master

Universal Labeling Merging and Separating algorithm deployment

Docker/Dockerfile_ULMS

@@ -0,0 +1,49 @@
+FROM python:3.7
+
+RUN apt-get update && apt-get install -y \
+    git \
+    libx11-6 \
+    libgl1 \
+    libopengl0 \
+    libegl1 \
+    wget
+
+
+RUN mkdir /app
+WORKDIR /app
+
+RUN wget https://github.com/RomainUSA/fly-by-cnn/releases/download/2.1.0/requirements.txt
+RUN pip3 install -r /app/requirements.txt
+
+RUN wget https://github.com/RomainUSA/fly-by-cnn/archive/refs/tags/2.1.0.zip
+RUN unzip 2.1.0.zip
+RUN rm -rf 2.1.0.zip 
+
+
+RUN mkdir /app/models
+WORKDIR /app/models
+
+RUN wget https://github.com/RomainUSA/fly-by-cnn/releases/download/2.1.0/nnLU_model_5.hdf5
+RUN wget https://github.com/RomainUSA/fly-by-cnn/releases/download/2.1.0/model_features.zip
+RUN unzip model_features.zip
+RUN rm -rf model_features.zip
+
+
+WORKDIR /app
+RUN wget https://github.com/RomainUSA/fly-by-cnn/releases/download/2.1.0/groundtruth.zip
+RUN unzip groundtruth.zip
+RUN rm -rf groundtruth.zip
+
+RUN mkdir /app/data
+RUN mkdir /app/data/input
+RUN mkdir /app/data/uid
+RUN mkdir /app/data/out_tmp
+RUN mkdir /app/data/merged
+RUN mkdir /app/data/out
+
+
+
+
+
+
+

Docker/README.md

@@ -19,3 +19,53 @@ This example will extract the normals and the depth map as a separate component
 ```
 python /app/fly-by-cnn/src/py/fly_by_features.py --surf my_surf.(vtk,.stl,.obj) --subdivision 2 --resolution 512 --out out.nrrd --radius 2 --use_z 1 --split_z 1"
 ```
+
+
+## Running Universal Labeling, Merging and Separating
+
+To get the docker:
+
+```
+docker pull dcbia/ulms:latest
+```
+
+To run the docker:
+
+**For the Universal Labeling:**
+
+Input: 
+- Dental crown model that contains the teeth of a patient (vtk file)
+
+Output: 
+- Dental crown model with the universal labels on each tooth (vtk file)
+
+```
+docker run --rm -v */my/input/file*:/app/data/input/P1_teeth.vtk -v */my/output/folder*:/app/data/out python3 fly-by-cnn-2.1.0/src/py/universal_labeling.py --surf /app/data/input/P1_teeth.vtk --label_GT_dir /app/groundtruth --model_ft /app/models/model_features --model_LU /app/models/nnLU_model_5.hdf5 --out_feature /app/data/feature.nrrd --out /app/data/out
+```
+
+**For the Merging and Separating:**
+
+Inputs: 
+- Dental crown model that contains the teeth of a patient with the universal labels on each tooth (vtk file)
+- The root canal segmentation file (.nii)
+
+Output: 
+- A merged model with the teeth and roots labeled based on the universal labels and individual teeth and roots (vtk files)
+
+```
+docker run --rm -v */my/input/file*:/app/data/input/P1_teeth_uid.vtk -v '*/my/input/file*:/app/data/input/lower_P1_scan_lower_RCSeg.nii.gz -v */my/output/folder*:/app/data/out ulms:latest bash fly-by-cnn-2.1.0/src/sh/compute_MergingSeparating.sh --src_code fly-by-cnn-2.1.0/src  --input_file_uid /app/data/input/P1_teeth_uid.vtk --input_file_root /app/data/input/lower_P1_scan_lower_RCSeg.nii.gz --out_tmp /app/data/out_tmp --out_merge /app/data/merged --out_separate /app/data/out
+```
+
+**For the ULMS:**
+
+Inputs: 
+- Dental crown model that contains the teeth of a patient (vtk file)
+- The root canal segmentation file (.nii)
+
+Output: 
+- A merged model with the teeth and roots labeled based on the universal labels and individual teeth and roots (vtk files)
+
+```
+docker run --rm -v */my/input/file*:/app/data/input/P1_teeth.vtk -v '*/my/input/file*:/app/data/input/lower_P1_scan_lower_RCSeg.nii.gz -v */my/output/folder*:/app/data/out ulms:latest bash fly-by-cnn-2.1.0/src/sh/compute_ULMS.sh --src_code fly-by-cnn-2.1.0/src  --input_file_surf /app/data/input/P1_teeth.vtk 
+--label_GT_dir /app/groundtruth --model_ft /app/models/model_features --model_LU /app/models/nnLU_model_5.hdf5 --out_feature /app/data/feature.nrrd --output_dir_uid /app/data/uid --input_file_root /app/data/input/lower_P1_scan_lower_RCSeg.nii.gz --out_tmp /app/data/out_tmp --out_merge /app/data/merged --out_separate /app/data/out
+```

readme.md

@@ -42,3 +42,104 @@ This example will extract the normals and the depth map as a separate component
 python /app/fly-by-cnn/src/py/fly_by_features.py --surf my_surf.(vtk,.stl,.obj) --subdivision 2 --resolution 512 --out out.nrrd --radius 2 --use_z 1 --split_z 1"
 ```
 
+## Running Universal Labeling, Merging and Separating algorithm
+
+The easiest way to use the docker container.
+[Docker/README.md](https://github.com/RomainUSA/fly-by-cnn/tree/master/Docker)
+
+Here are the command lines to run the scripts:
+
+**Universal Labeling:**
+
+Input: 
+- Dental crown model that contains the teeth of a patient (vtk file)
+
+Output: 
+- Dental crown model with the universal labels on each tooth (vtk file)
+
+```
+python3 fly-by-cnn/src/py/universal_labeling.py --help
+```
+
+```
+usage: universal_labeling.py [-h] --surf SURF --label_groundtruth
+                             LABEL_GROUNDTRUTH --model_feature MODEL_FEATURE
+                             --model_LU MODEL_LU --out_feature OUT_FEATURE
+                             [--out OUT]
+
+Label the teeth from 2 to 16 or with the universal IDs used by clinicians
+
+optional arguments:
+  -h, --help            show this help message and exit
+  --surf SURF           Input surface mesh to label (default: None)
+  --out OUT             Output model with labels (default: out.vtk)
+
+Label parameters:
+  --label_groundtruth LABEL_GROUNDTRUTH
+                        directory of the template labels (default: None)
+
+Prediction parameters:
+  --model_feature MODEL_FEATURE
+                        path of the VGG19 model (default: None)
+  --model_LU MODEL_LU   path of the LowerUpper model (default: None)
+  --out_feature OUT_FEATURE
+                        out of the feature (default: None)
+```
+
+**Merging and Separating:**
+
+Inputs: 
+- Dental crown model that contains the teeth of a patient with the universal labels on each tooth (vtk file)
+- The root canal segmentation file (.nii)
+
+Output: 
+- A merged model with the teeth and roots labeled based on the universal labels and individual teeth and roots (vtk files)
+
+```
+bash fly-by-cnn/src/sh/compute_MergingSeparating.sh --help
+```
+
+```
+Program to run the Merging and Separating algorithm
+
+Syntax: compute_MergingSeparating.sh [--options]
+options:
+--src_code                    Path of the source code 
+--input_file_uid              Output directory of the teeth with the universal labels.
+--input_file_root             Input directory with the root canal segmentation files.
+--out_tmp                     Temporary output folder.
+--out_merge                   Output directory of the merged surfaces.
+--out_separate                Output directory of the separated surfaces.
+```
+
+**ULMS algortihm:**
+
+Inputs: 
+- Dental crown model that contains the teeth of a patient (vtk file)
+- The root canal segmentation file (.nii)
+
+Output: 
+- A merged model with the teeth and roots labeled based on the universal labels and individual teeth and roots (vtk files)
+
+```
+bash fly-by-cnn/src/sh/compute_ULMS.sh --help
+```
+
+```
+Program to run the Universal Labeling Merging and Separated algorithm
+
+Syntax: compute_ULMS.sh [--options]
+options:
+--src_code                    Path of the source code 
+--input_file_surf             Input file surface with only the teeth.
+--label_GT_dir                Folder containing the template for the Upper/Lower classification.
+--model_ft                    Path to the feature model .
+--model_LU                    Path to the LowerUpper classification model.
+--out_ft                      Output of the fearure.
+--output_dir_uid              Output directory of the teeth with the universal labels.
+--input_file_root             Root canal segmentation file.
+--out_tmp                     Temporary output folder.
+--out_merge                   Output directory of the merged surfaces.
+--out_separate                Output directory of the separated surfaces.
+```
+

src/py/Classification_LU/model.py

@@ -0,0 +1,30 @@
+import os
+
+from keras.models import Sequential
+from keras.layers import LSTM, Dense, Input, Bidirectional
+from tensorflow.keras.optimizers import *
+
+
+def LSTM_model():
+    model = Sequential()
+
+    # Two bidirectional LSTM layers 
+    model.add(Bidirectional(LSTM(64, return_sequences=True), input_shape=(64,512)))
+    model.add(Bidirectional(LSTM(64)))
+
+    # Sequence voting layer
+    model.add(Dense(64))
+
+    # Fully connected layer
+    model.add(Dense(1, activation='sigmoid'))
+
+    # Compile the model
+    model.compile(optimizer=Adam(lr = 1e-4), loss='binary_crossentropy', metrics=['accuracy'])
+    model.summary()
+
+    return model
+
+
+
+
+

src/py/Classification_LU/testing.py

@@ -0,0 +1,88 @@
+from model import *
+from tensorflow.keras.models import load_model
+from sklearn.metrics import classification_report
+
+import argparse
+import os
+import itk
+
+import numpy as np
+import tensorflow as tf
+
+
+parser = argparse.ArgumentParser(description='Predict an input with a trained neural network', formatter_class=argparse.ArgumentDefaultsHelpFormatter)
+
+in_group = parser.add_mutually_exclusive_group(required=True)
+in_group.add_argument('--features', type=str, help='Input features to be predict as an Upper or Lower teeth')
+in_group.add_argument('--dir_features', type=str, help='Input dir features to be predict as an Upper or Lower teeth')
+
+parser.add_argument('--load_model', type=str, help='Saved model', required=True)
+parser.add_argument('--display', type=bool, help='display the prediction', default=False)
+
+args = parser.parse_args()
+
+inputFeatures = args.features
+inputdirFeatures = args.dir_features
+loadModelPath = args.load_model
+
+
+
+def ReadImage(fName, image_dimension=2, pixel_dimension=-1):
+	if(image_dimension == 1):
+		if(pixel_dimension != -1):
+			ImageType = itk.Image[itk.Vector[itk.F, pixel_dimension], 2]
+		else:
+			ImageType = itk.VectorImage[itk.F, 2]
+	else:
+		if(pixel_dimension != -1):
+			ImageType = itk.Image[itk.Vector[itk.F, pixel_dimension], image_dimension]
+		else:
+			ImageType = itk.VectorImage[itk.F, image_dimension]
+
+	img_read = itk.ImageFileReader[ImageType].New(FileName=fName)
+	img_read.Update()
+	img = img_read.GetOutput()
+	return img
+
+def process_scan(path):
+    img = ReadImage(path)
+    img_np = itk.GetArrayViewFromImage(img)
+    img_np = np.reshape(img_np, [s for s in img_np.shape if s != 1])
+    return img_np
+
+
+
+if (args.dir_features):
+    features_paths = [os.path.join(inputdirFeatures, x) for x in os.listdir(inputdirFeatures) if not x.startswith(".")]
+    features_name = [y for y in os.listdir(inputdirFeatures) if not y.startswith(".")]
+    features = np.array([process_scan(path) for path in features_paths])
+
+if (args.features):
+    features_name = inputFeatures.split("/")[-1]
+    features = np.array([process_scan(inputFeatures)])
+
+
+model = load_model(loadModelPath)
+predictions = model.predict(features)
+
+
+y_true = []
+for i in range(len(features_name)):
+    if features_name[i][:1]=="L":
+        y_true.append(0)
+
+    if features_name[i][:1]=="U":
+        y_true.append(1)
+
+predictions[predictions>0.5]=1
+predictions[predictions<=0.5]=0
+
+
+
+if (args.display):
+    target_names = ['Lower', 'Upper']
+    print(classification_report(y_true, predictions, target_names=target_names))
+
+
+
+