# ==============================================================================
# Copyright (C) 2023 Haresh Rengaraj Rajamohan, Tianyu Wang, Kevin Leung,
# Gregory Chang, Kyunghyun Cho, Richard Kijowski & Cem M. Deniz
#
# This file is part of OAI-MRI-TKR
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as published
# by the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <https://www.gnu.org/licenses/>.
# ==============================================================================
import numpy as np
import pandas as pd
import h5py
import nibabel as nib
import keras
from mpl_toolkits.mplot3d import Axes3D
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.cm
import matplotlib.colorbar
import matplotlib.colors
import pandas as pd
import numpy as np
from sklearn import metrics
import os
from keras.models import load_model
from Augmentation import RandomCrop, CenterCrop, RandomFlip
from sklearn.metrics import roc_auc_score,auc,roc_curve,average_precision_score
import tensorflow as tf
from DataGenerator import DataGenerator as DG
class DataGenerator(keras.utils.Sequence):
'Generates data for Keras'
def __init__(self, directory, file_folder,batch_size=8, dim=(384,384,160), n_channels=1,
n_classes=10, shuffle=True,normalize = True, randomCrop = True, randomFlip = True,
flipProbability = -1, cropDim = (384,384,160)):
'Initialization'
self.dim = dim
self.batch_size = batch_size
self.dataset = pd.read_csv(directory)
#self.list_IDs = list_IDs
self.list_IDs = pd.read_csv(directory)['h5Name']
self.n_channels = n_channels
self.n_classes = n_classes
self.shuffle = shuffle
self.on_epoch_end()
self.file_folder = file_folder+"00m/"
self.normalize = normalize
self.randomCrop = randomCrop
self.randomFlip = randomFlip
self.flipProbability = flipProbability
self.cropDim = cropDim
def __len__(self):
'Denotes the number of batches per epoch'
return int(np.floor(len(self.list_IDs) / self.batch_size))
def __getitem__(self, index):
'Generate one batch of data'
# Generate indexes of the batch
indexes = self.indexes[index*self.batch_size:(index+1)*self.batch_size]
# Find list of IDs
list_IDs_temp = [self.list_IDs[k] for k in indexes]
# Generate data
X, y = self.__data_generation(list_IDs_temp)
return X, y
def on_epoch_end(self):
'Updates indexes after each epoch'
self.indexes = np.arange(len(self.list_IDs))
if self.shuffle == True:
np.random.shuffle(self.indexes)
def __data_generation(self, list_IDs_temp):
'Generates data containing batch_size samples' # X : (n_samples, *dim, n_channels)
# Initialization
X = np.empty((self.batch_size, *self.dim, self.n_channels))
y = np.empty((self.batch_size), dtype=int)
# Generate data
for i, ID in enumerate(list_IDs_temp):
# Store sample
pre_image = h5py.File(self.file_folder + ID, "r")['data/'].value.astype('float64')
#pre_image = padding_image(data = image,shape = [448,448,48])
#pre_image = np.zeros(image.shape)
#pre_image = image
if pre_image.shape[2]<144:
pre_image = padding_image2(data = pre_image)
# normalize
if self.normalize:
pre_image = normalize_MRIs(pre_image)
# Augmentation
if self.randomFlip:
pre_image = RandomFlip(image=pre_image,p=0.5).horizontal_flip(p=self.flipProbability)
if self.randomCrop:
pre_image = RandomCrop(pre_image).crop_along_hieght_width_depth(self.cropDim)
else:
pre_image = CenterCrop(image=pre_image).crop(size = self.cropDim)
#print(ID,pre_image.shape)
X[i,:,:,:,0] = pre_image
# Store class
y[i] = self.dataset[self.dataset.h5Name == ID].Label
return X, y
def getXvalue(self,index):
return self.__getitem__(index)
def padding_image2(data):
l,w,h = data.shape
images = np.zeros((l,w,144))
zstart = int(np.ceil((144-data.shape[2])/2))
images[:,:,zstart:zstart + h] = data
return images
def padding_image(data, shape):
images = np.zeros(shape)
candi = data
candi_shape = data.shape
xstart = int(np.ceil((448-candi_shape[0])/2))
ystart = int(np.ceil((448-candi_shape[1])/2))
zstart = int(np.ceil((48-candi_shape[2])/2))
images[xstart:xstart+candi_shape[0],ystart:ystart + candi_shape[1],zstart:zstart+candi_shape[2]] = candi
return images
def normalize_MRIs(image):
mean = np.mean(image)
std = np.std(image)
image -= mean
#image -= 95.09
image /= std
#image /= 86.38
return image
tf.app.flags.DEFINE_string('model_path', '/gpfs/data/denizlab/Users/hrr288/Radiology_test/SAG3D_lr24_18_stride221_kernel777773/', 'Folder with the models')
tf.app.flags.DEFINE_string('val_csv_path', '/gpfs/data/denizlab/Users/hrr288/Tianyu_dat/TestSets/', 'Folder with the fold splits')
tf.app.flags.DEFINE_string('test_csv_path', '/gpfs/data/denizlab/Users/hrr288/data/OAI_SAG_DESS_test.csv', 'Folder with the test csv')
tf.app.flags.DEFINE_string('result_path', './', 'Folder to save output csv with preds')
tf.app.flags.DEFINE_bool('vote', False, 'Choice to generate binary predictions for each model to compute final sensitivity/specificity')
tf.app.flags.DEFINE_string('file_folder','/gpfs/data/denizlab/Datasets/OAI/SAG_3D_DESS/', 'Path to DESS HDF5 radiographs of test set')
tf.app.flags.DEFINE_string('train_file_folder','/gpfs/data/denizlab/Datasets/OAI/SAG_3D_DESS/', 'Path to DESS HDF5 radiographs of OAI train/val set')
FLAGS = tf.app.flags.FLAGS
def main(argv=None):
val_params = {'dim': (352,352,144),
'batch_size': 1,
'n_classes': 2,
'n_channels': 1,
'shuffle': False,
'normalize' : True,
'randomCrop' : False,
'randomFlip' : False,
'flipProbability' : -1,
'cropDim' : (352,352,144)}
validation_generator = DataGenerator(directory = FLAGS.test_csv_path,file_folder=FLAGS.file_folder, **val_params)
df = pd.read_csv(FLAGS.test_csv_path,index_col=0)
base_path = FLAGS.model_path
models= {'fold_1':[],'fold_2':[],'fold_3':[],'fold_4':[],'fold_5':[],'fold_6':[],'fold_7':[]}
for fold in np.arange(1,8):
tmp_mod_list = []
for cv in np.arange(1,7):
dir_1 = 'Fold_'+str(fold)+'/CV_'+str(cv)+'/'
files_avai = os.listdir(base_path+dir_1)
cands = []
cands_score = []
for fs in files_avai:
if 'weights' not in fs:
continue
else:
cands_score.append(float(fs.split('-')[2]))
cands.append(dir_1+fs)
ind_c = int(np.argmin(cands_score))
tmp_mod_list.append(cands[ind_c])
models['fold_'+str(fold)]=tmp_mod_list
AUCS = []
preds = []
dfs = []
pred_arr = np.zeros(df.shape[0])
for i in np.arange(1,8):
for j in np.arange(1,7):
model = load_model(base_path+'/'+models['fold_'+str(i)][j-1])
if FLAGS.vote:
test_df = pd.read_csv(FLAGS.val_csv_path+'Fold_'+str(i)+'/CV_'+str(j)+'_val.csv')
test_generator = DG(directory = FLAGS.val_csv_path+'Fold_'+str(i)+'/CV_'+str(j)+'_val.csv',file_folder=FLAGS.train_file_folder, **val_params)
test_pred = model.predict_generator(test_generator)
test_df["Pred"] = test_pred
fpr, tpr, thresholds = metrics.roc_curve(test_df["Label"], test_df["Pred"])
opt_ind = np.argmax(tpr-fpr)
opt_thresh = thresholds[int(opt_ind)]
s = model.predict_generator(validation_generator)
pred_arr += (np.squeeze(s)>=opt_thresh)
else:
s = model.predict_generator(validation_generator)
pred_arr += np.squeeze(s)
#AUCS.append(roc_auc_score(df['Label'],pred_arr))
#preds.extend(list(pred_arr))
pred_arr = pred_arr/42
# In[ ]:
df["Preds"] = pred_arr
if FLAGS.vote:
df.to_csv(FLAGS.result_path+"OAI_results_vote.csv")
else:
df.to_csv(FLAGS.result_path+"OAI_results.csv")
if __name__ == "__main__":
tf.app.run()
Datasets
Models
