def segment_script(patch_size_x, patch_size_y, patch_size_z, hidden_layers_sizes, corruption_levels, prefix):
root = '../../varghese/10_1_brain_mean/'
#patch_size_x = 9
#patch_size_y = 9
#patch_size_z = 9
recon_flag = False
batch_size = 100
if recon_flag == True:
n_ins = patch_size_x * patch_size_y * patch_size_z * 5
else:
n_ins = patch_size_x * patch_size_y * patch_size_z * 4
n_outs = 5
noise_type = 1
noise_dict = {1:'Gaussian Noise',0:'Masking Noise'}
#hidden_layers_sizes = [1000,1000,1000]
#corruption_levels = [0.01,0.01,0.01]
test_path = root + 'testing'
#prefix = 'xxx'
print 'Extracting Balanaced training patches...'
B_Patch_Preprocess_recon_3D(patch_size_x,patch_size_y,patch_size_z,prefix,root+'training',
root+'BRATS_training_patches/',False)
print 'Training patches extracted!'
print 'Extracting Balanced validation patches...'
B_Patch_Preprocess_recon_3D(patch_size_x,patch_size_y,patch_size_z,prefix,root+'validation',
root+'BRATS_validation_patches/',False)
print 'Validation patches extracted!'
print 'Extracting UnBalanced training patches...'
U_Patch_Preprocess_recon_3D(patch_size_x,patch_size_y,patch_size_z,prefix,root+'training',
root+'BRATS_training_patches/',False)
print 'Training patches extracted!'
print 'Extracting Unbalanced validation patches...'
U_Patch_Preprocess_recon_3D(patch_size_x,patch_size_y,patch_size_z,prefix,root+'validation',
root+'BRATS_validation_patches/',False)
path = '../results/'
for subdir, dirs, files in os.walk(path):
test_num = len(dirs)+1
break
os.mkdir('../results/test_'+str(test_num)+'_'+prefix)
test_root = '../results/test_'+str(test_num)+'_'+prefix+'/'
print 'Calling test_SdA...'
finetune_lr = 0.1
pretraining_epochs = 50
pretrain_lr = 0.001
training_epochs = 100
f = open(test_root+prefix+'_params_info.txt', 'w')
f.write( "Current date & time " + time.strftime("%c"))
f.write('\nPrefix : '+prefix)
f.write('\n3D Patches. Patch_size : '+str(patch_size_x)+', '+str(patch_size_y)+', '+str(patch_size_z))
f.write('\nHidden Layer Sizes : ['+', '.join(map(str,hidden_layers_sizes))+' ]')
f.write('\nNoise Type : '+noise_dict[noise_type])
f.write('\nCorruption Levels : ['+', '.join(map(str,corruption_levels))+' ]')
f.write('\nNo. of pre-training epochs : '+str(pretraining_epochs))
f.write('\nNo. of Fine-tuning epochs : '+str(training_epochs))
f.write('\nPretraining Learning rate : '+str(pretrain_lr))
f.write('\nFine-tuning learning rate : '+str(finetune_lr))
f.close()
test_SdA(finetune_lr, pretraining_epochs,
pretrain_lr, training_epochs,
root+'BRATS_training_patches/b_trainpatch_3D_'+prefix+'_.npy',
root+'BRATS_training_patches/b_trainlabel_3D_'+prefix+'_.npy',
root+'BRATS_validation_patches/b_validpatch_3D_'+prefix+'_.npy',
root+'BRATS_validation_patches/b_validlabel_3D_'+prefix+'_.npy',
root+'BRATS_training_patches/u_trainpatch_3D_'+prefix+'_.npy',
root+'BRATS_training_patches/u_trainlabel_3D_'+prefix+'_.npy',
root+'BRATS_validation_patches/u_validpatch_3D_'+prefix+'_.npy',
root+'BRATS_validation_patches/u_validlabel_3D_'+prefix+'_.npy',batch_size, n_ins, n_outs, hidden_layers_sizes, test_root + prefix, corruption_levels)
print 'Network Trained and Saved!'
test_network(test_root , prefix, test_path, patch_size_x, patch_size_y, patch_size_z, recon_flag)
convert_mha(root+'testing', prefix)
print '####################################################################'
print
print 'Completed one network with prefix : ', prefix
print
print '####################################################################'
f.write('\nHidden Layer Sizes : ['+', '.join(map(str,hidden_layers_sizes))+' ]')
f.write('\nNoise Type : '+noise_dict[noise_type])
f.write('\nCorruption Levels : ['+', '.join(map(str,corruption_levels))+' ]')
f.write('\nNo. of pre-training epochs : '+str(pretraining_epochs))
f.write('\nNo. of Fine-tuning epochs : '+str(training_epochs))
f.write('\nPretraining Learning rate : '+str(pretrain_lr))
f.write('\nFine-tuning learning rate : '+str(finetune_lr))
f.close()
test_SdA(finetune_lr, pretraining_epochs,
pretrain_lr, training_epochs,
root+'BRATS_training_patches/b_trainpatch_3D_'+prefix+'_.npy',
root+'BRATS_training_patches/b_trainlabel_3D_'+prefix+'_.npy',
root+'BRATS_validation_patches/b_validpatch_3D_'+prefix+'_.npy',
root+'BRATS_validation_patches/b_validlabel_3D_'+prefix+'_.npy',
root+'BRATS_training_patches/u_trainpatch_3D_'+prefix+'_.npy',
root+'BRATS_training_patches/u_trainlabel_3D_'+prefix+'_.npy',
root+'BRATS_validation_patches/u_validpatch_3D_'+prefix+'_.npy',
root+'BRATS_validation_patches/u_validlabel_3D_'+prefix+'_.npy',batch_size, n_ins, n_outs, hidden_layers_sizes, test_root + prefix, corruption_levels)
print 'Network Trained and Saved!'
test_network(test_root , prefix, test_path, patch_size_x, patch_size_y, patch_size_z, recon_flag)
# convert_mha(root+'testing', prefix)
f.write("\nNo. of Fine-tuning epochs : " + str(training_epochs))
f.write("\nPretraining Learning rate : " + str(pretrain_lr))
f.write("\nFine-tuning learning rate : " + str(finetune_lr))
f.close()
test_SdA(
finetune_lr,
pretraining_epochs,
pretrain_lr,
training_epochs,
root + "BRATS_training_patches/b_10_trainpatch_2D_" + prefix + "_.npy",
root + "BRATS_training_patches/b_10_trainlabel_2D_" + prefix + "_.npy",
root + "BRATS_validation_patches/b_10_validpatch_2D_" + prefix + "_.npy",
root + "BRATS_validation_patches/b_10_validlabel_2D_" + prefix + "_.npy",
root + "BRATS_training_patches/u_10_trainpatch_2D_" + prefix + "_.npy",
root + "BRATS_training_patches/u_10_trainlabel_2D_" + prefix + "_.npy",
root + "BRATS_validation_patches/u_10_validpatch_2D_" + prefix + "_.npy",
root + "BRATS_validation_patches/u_10_validlabel_2D_" + prefix + "_.npy",
batch_size,
n_ins,
n_outs,
hidden_layers_sizes,
test_root + prefix,
corruption_levels,
)
print "Network Trained and Saved!"
test_network(test_root, prefix, test_path, patch_size, patch_size, patch_size, recon_flag, 2)
# convert_mha(root+'testing', prefix, 2)
# for i in xrange(len(dropout_rates)):
# if i == 1:
# break
# runMLP2(finetune_lr,
# 150,
# batch_size,
# layer_sizes,
# dropout_rates[i],
# test_root + prefix + 'pre_training.pkl',
# root+'BRATS_training_patches/u_trainpatch_3D_'+prefix+'_.npy',
# root+'BRATS_training_patches/u_trainlabel_3D_'+prefix+'_.npy',
# root+'BRATS_validation_patches/u_validpatch_3D_'+prefix+'_.npy',
# root+'BRATS_validation_patches/u_validlabel_3D_'+prefix+'_.npy',
# test_root + prefix + '_' + str(i))
#########################################################################################
test_network(test_root,new_prefix,test_path,11,11,3,False,3)
# ### START OF POST-PROCESSING PIPELINE ###
LesionMasker(test_path + '/' , new_prefix)
print 'binarizing...'
binarize(root, new_prefix + 'Masked_RawOutput.nii')
print new_prefix + 'Masked_RawOutput.nii'
print 'binarized'
callString = 'python analysis.py > ' + new_prefix + '.txt'
print callString
subprocess.call(callString, shell = True)
calculateDice(new_prefix + '.txt', testingImages)
@author: bmi
"""
from test_network import *
#prefix = ['11x11_1000-500-250-100_G1','11x11_2000-1000-1000-500_G1','13x13_1000-500-250_G1','13x13_2000-1000-500-100_G1','9x9_2000-1000-500-100_G1',
# '9x9_1000-500-250-100_G1','9x9_1000-1000-1000_G1','9x9_1000-500-100_G1','7x7_1000-500-250_G1']
prefix = ['Tumor_NonTumor_9x9x9_5000-200-500G1','Tumor_NonTumor_11x11x11_5000-2000-500G15']
root = '../../varghese/10_1_brain_mean/'
#test_root = ['/home/bmi/BRATS/results/test2D9_11x11_1000-500-250-100_G1/',
# '/home/bmi/BRATS/results/test2D10_11x11_2000-1000-1000-500_G1/',
# '/home/bmi/BRATS/results/test2D11_13x13_1000-500-250_G1/',
# '/home/bmi/BRATS/results/test2D12_13x13_2000-1000-500-100_G1/',
# '/home/bmi/BRATS/results/test2D13_9x9_2000-1000-500-100_G1/',
# '/home/bmi/BRATS/results/test2D14_9x9_1000-500-250-100_G1/',
# '/home/bmi/BRATS/results/test2D15_9x9_1000-1000-1000_G1/',
# '/home/bmi/BRATS/results/test2D16_9x9_1000-500-100_G1/',
# '/home/bmi/BRATS/results/test2D17_7x7_1000-500-250_G1/']
test_root = ['/home/bmi/BRATS/results/test17_Tumor_NonTumor_9x9x9_5000-200-500G1/',
'/home/bmi/BRATS/results/test18_Tumor_NonTumor_9x9x9_5000-2000-500G15/']
patch_z = [11,11,13,13,9,9,9,9,7]
test_path = root+'testing'
for i in xrange(len(prefix)):
test_network(test_root[i],prefix[i],test_path,9,9,9,False,3)
print i
def segment_2D_script(patch_size, hidden_layers_sizes, corruption_levels, prefix):
root = '../../varghese/10_1_brain_mean/'
# patch_size = 11
recon_flag = False
batch_size = 100
if recon_flag == True:
n_ins = patch_size * patch_size * 5
else:
n_ins = patch_size * patch_size * 4
n_outs = 5
# hidden_layers_sizes = [1000,1000,1000]
# corruption_levels = [0.01,0.01,0.01]
# noise_type = 1 #1- Gaussian, 0 - masking
test_path = root + 'testing'
# prefix = 'yyy'
print 'Extracting training patches...'
Patch_Preprocess_recon_2D(patch_size,patch_size, prefix,root+'training',root+'BRATS_training_patches/',False)
print 'Training patches extracted!'
print 'Extracting validation patches...'
Patch_Preprocess_recon_2D(patch_size,patch_size,prefix,root+'validation',
root+'BRATS_validation_patches/',False)
print 'Validation patches extracted!'
path = '../results/'
for subdir, dirs, files in os.walk(path):
test_num = len(dirs)+1
break
# print test_num
##########---------SET PREFIX--------##########
os.mkdir('../results/test2D'+str(test_num)+'_'+prefix)
#
test_root = '../results/test2D'+str(test_num)+'_'+prefix+'/'
print 'Calling test_SdA...'
finetune_lr = 0.1
pretraining_epochs = 1
pretrain_lr = 0.001
training_epochs = 1
test_SdA(finetune_lr, pretraining_epochs,
pretrain_lr, training_epochs,
root+'BRATS_training_patches/trainpatch_2D_'+prefix+'_.npy',
root+'BRATS_training_patches/trainlabel_2D_'+prefix+'_.npy',
root+'BRATS_validation_patches/validpatch_2D_'+prefix+'_.npy',
root+'BRATS_validation_patches/validlabel_2D_'+prefix+'_.npy',batch_size, n_ins, n_outs, hidden_layers_sizes, test_root + prefix, corruption_levels)
print 'Network Trained and Saved!'
test_network(test_root , prefix, test_path, patch_size, patch_size, patch_size, recon_flag, 2)
convert_mha(root+'testing', prefix, 2)
#
#
from test_network import * from lesion_masker import * prefix = '2013new_9x9x9_5000-2000-500_M244' root = '/media/varkey/mega_test/HGG_NORM/' test_root = '/home/bmi/BRATS/results/test_147_2013new_9x9x9_5000-2000-500_M244/' test_path = root test_network(test_root,prefix,test_path,9,9,9,False,3) #LesionMasker(prefix, test_path)
