import numpy as np
import SimpleITK as sitk
import random
def generate_negative_patches(studies):
art_arr = sitk.GetArrayFromImage(studies[0])
ven_arr = sitk.GetArrayFromImage(studies[1])
noncon_arr = sitk.GetArrayFromImage(studies[2])
mask_arr = sitk.GetArrayFromImage(studies[3])
tot_num_neg = 1
neg_patch = 0
#### Choose where to randomly select a negative patch using
#### Gaussian dist. with peak and std according to
#### blocks of 32 voxels. For example, 16 blocks in 512 voxels.
rand_xy = np.around(np.random.normal(4,2,size=(2,1000)))
rand_z = np.around(np.random.normal(10,4,1000))
rand_xyz = np.vstack((rand_xy, rand_z))
i = 0
chosen_xyz = list()
output_patches = list()
##Create Random Negative Patches
while neg_patch<tot_num_neg:
bad_patch_flag=False
i = i+1
print (i)
## If no patch is found after 80 attempts, abort
if i == 80:
return
## Randomly selected blocks of 32 converted to voxels
x_start = int(rand_xyz[0][i]*32)
y_start = int(rand_xyz[1][i]*32)
z_start = int(rand_xyz[2][i]*32)
# Choose 256^3 patch
x_end = x_start+256
y_end = y_start+256
z_end = z_start+256
print ("Random patch: ")
print (x_start,x_end,y_start,y_end,z_start,z_end)
patch_noncon = noncon_arr[z_start:z_end,y_start:y_end,x_start:x_end]
patch_ven = ven_arr[z_start:z_end,y_start:y_end,x_start:x_end]
patch_art = art_arr[z_start:z_end,y_start:y_end,x_start:x_end]
patch_mask = mask_arr[z_start:z_end,y_start:y_end,x_start:x_end]
if [rand_xyz[0][i],rand_xyz[1][i],rand_xyz[2][i]] in chosen_xyz or rand_xyz[0][i]<0 \
or rand_xyz[0][i]>11 or rand_xyz[1][i]>11 or rand_xyz[1][i]<0 or rand_xyz[2][i]>15 or rand_xyz[2][i]<0:
print ("Exclude Random Patch --> Outside FOV")
continue
##Determine if patch has a low mean value, std or excessive 0 values
for arr_patch in [patch_noncon,patch_ven,patch_art]:
if len(np.where(arr_patch==0)[0]) > 0.3*(256**3) or arr_patch.mean()<-450:
print ("Try new patch, likely off target given stats")
print (len(np.where(arr_patch==0)[0])/(256**3))
print (arr_patch.mean())
print (arr_patch.std())
bad_patch_flag=True
if bad_patch_flag == True or patch_mask.mean()>0:
print ("Rejected 'cause of flag or patch mask mean greater than 0")
print (patch_mask.mean()>0)
continue
print ([x_start,x_end,y_start,y_end,z_start,z_end])
output_patches.append([x_start,x_end,y_start,y_end,z_start,z_end])
chosen_xyz.append([rand_xyz[0][i],rand_xyz[1][i],rand_xyz[2][i]])
print ("Identified a neg_patch number : "+str(neg_patch))
neg_patch = neg_patch + 1
#sitk.WriteImage(sitk.GetImageFromArray(arr_patch),os.path.join(OUT_DIR,str(i)+".nii.gz"))
return output_patches
def generate_positive_patches(studies,patch_size,random_center_displacement):
q = random_center_displacement
pos_patch = list()
mask_arr = sitk.GetArrayFromImage(studies[3])
where = np.where(mask_arr==1)
z_mid = int(where[0].mean())
y_mid = int(where[1].mean())
x_mid = int(where[2].mean())
print (z_mid,y_mid,x_mid)
z_start = random.randint(-(q),q)+z_mid-(patch_size[0]/2)
y_start = random.randint(-(q),q)+y_mid-(patch_size[1]/2)
x_start = random.randint(-(q),q)+x_mid-(patch_size[2]/2)
z_end = z_start+patch_size[0]
y_end = y_start+patch_size[1]
x_end = x_start+patch_size[2]
numbers = [ int(x) for x in [x_start,x_end,y_start,y_end,z_start,z_end] ]
pos_patch.append(numbers)
return pos_patch
def test_onbody_patch(load_path):
try:
arr = sitk.GetArrayFromImage(sitk.ReadImage(load_path))
print (load_path)
print (np.mean(arr))
# print (np.max(arr))
# print (np.min(arr))
except Exception as e:
print ("Could not convert because "+str(e))
Datasets
Models
