def savenpy(id, filelist, prep_folder, data_path, use_existing=True):
resolution = np.array([1, 1, 1])
name = filelist[id]
if use_existing:
if os.path.exists(os.path.join(
prep_folder, name + '_label.npy')) and os.path.exists(
os.path.join(prep_folder, name + '_clean.npy')):
print(name + ' had been done')
return
try:
im, m1, m2, spacing = step1_python(os.path.join(data_path, name))
Mask = m1 + m2
newshape = np.round(np.array(Mask.shape) * spacing / resolution)
xx, yy, zz = np.where(Mask)
box = np.array([[np.min(xx), np.max(xx)], [np.min(yy),
np.max(yy)],
[np.min(zz), np.max(zz)]])
box = box * np.expand_dims(spacing, 1) / np.expand_dims(resolution, 1)
box = np.floor(box).astype('int')
margin = 5
extendbox = np.vstack([
np.max([[0, 0, 0], box[:, 0] - margin], 0),
np.min([newshape, box[:, 1] + 2 * margin], axis=0).T
]).T
extendbox = extendbox.astype('int')
convex_mask = m1
dm1 = process_mask(m1)
dm2 = process_mask(m2)
dilatedMask = dm1 + dm2
Mask = m1 + m2
extramask = dilatedMask ^ Mask
bone_thresh = 210
pad_value = 170
im[np.isnan(im)] = -2000
sliceim = lumTrans(im)
sliceim = sliceim * dilatedMask + pad_value * (
1 - dilatedMask).astype('uint8')
bones = sliceim * extramask > bone_thresh
sliceim[bones] = pad_value
sliceim1, _ = resample(sliceim, spacing, resolution, order=1)
sliceim2 = sliceim1[extendbox[0, 0]:extendbox[0, 1],
extendbox[1, 0]:extendbox[1, 1],
extendbox[2, 0]:extendbox[2, 1]]
sliceim = sliceim2[np.newaxis, ...]
np.save(os.path.join(prep_folder, name + '_clean'), sliceim)
np.save(os.path.join(prep_folder, name + '_label'),
np.array([[0, 0, 0, 0]]))
img = sliceim[0]
print sliceim.shape
plt.imshow(img, cmap=plt.cm.gray)
plt.savefig('/home/ly/data/dsb2017/sample_processing/666666.jpg')
except:
print('bug in ' + name)
raise
print(name + ' done')
def savenpy(id,annos,filelist,data_path,prep_folder):
resolution = np.array([1,1,1])
name = filelist[id]
label = annos[annos[:,0]==name]
label = label[:,[3,1,2,4]].astype('float')
im, m1, m2, spacing = step1_python(os.path.join(data_path,name))
Mask = m1+m2
newshape = np.round(np.array(Mask.shape)*spacing/resolution)
xx,yy,zz= np.where(Mask)
box = np.array([[np.min(xx),np.max(xx)],[np.min(yy),np.max(yy)],[np.min(zz),np.max(zz)]])
box = box*np.expand_dims(spacing,1)/np.expand_dims(resolution,1)
box = np.floor(box).astype('int')
margin = 5
extendbox = np.vstack([np.max([[0,0,0],box[:,0]-margin],0),np.min([newshape,box[:,1]+2*margin],axis=0).T]).T
extendbox = extendbox.astype('int')
convex_mask = m1
dm1 = process_mask(m1)
dm2 = process_mask(m2)
dilatedMask = dm1+dm2
Mask = m1+m2
extramask = dilatedMask - Mask
bone_thresh = 210
pad_value = 170
im[np.isnan(im)]=-2000
sliceim = lumTrans(im)
sliceim = sliceim*dilatedMask+pad_value*(1-dilatedMask).astype('uint8')
bones = sliceim*extramask>bone_thresh
sliceim[bones] = pad_value
sliceim1,_ = resample(sliceim,spacing,resolution,order=1)
sliceim2 = sliceim1[extendbox[0,0]:extendbox[0,1],
extendbox[1,0]:extendbox[1,1],
extendbox[2,0]:extendbox[2,1]]
sliceim = sliceim2[np.newaxis,...]
np.save(os.path.join(prep_folder,name+'_clean.npy'),sliceim)
if len(label)==0:
label2 = np.array([[0,0,0,0]])
elif len(label[0])==0:
label2 = np.array([[0,0,0,0]])
elif label[0][0]==0:
label2 = np.array([[0,0,0,0]])
else:
haslabel = 1
label2 = np.copy(label).T
label2[:3] = label2[:3][[0,2,1]]
label2[:3] = label2[:3]*np.expand_dims(spacing,1)/np.expand_dims(resolution,1)
label2[3] = label2[3]*spacing[1]/resolution[1]
label2[:3] = label2[:3]-np.expand_dims(extendbox[:,0],1)
label2 = label2[:4].T
np.save(os.path.join(prep_folder,name+'_label.npy'),label2)
print(name)
def savenpy(id,annos,filelist,data_path,prep_folder):
resolution = np.array([1,1,1])
name = filelist[id]
label = annos[annos[:,0]==name]
label = label[:,[3,1,2,4]].astype('float')
im, m1, m2, spacing = step1_python(os.path.join(data_path,name))
Mask = m1+m2
newshape = np.round(np.array(Mask.shape)*spacing/resolution)
xx,yy,zz= np.where(Mask)
box = np.array([[np.min(xx),np.max(xx)],[np.min(yy),np.max(yy)],[np.min(zz),np.max(zz)]])
box = box*np.expand_dims(spacing,1)/np.expand_dims(resolution,1)
box = np.floor(box).astype('int')
margin = 5
extendbox = np.vstack([np.max([[0,0,0],box[:,0]-margin],0),np.min([newshape,box[:,1]+2*margin],axis=0).T]).T
extendbox = extendbox.astype('int')
convex_mask = m1
dm1 = process_mask(m1)
dm2 = process_mask(m2)
dilatedMask = dm1+dm2
Mask = m1+m2
extramask = dilatedMask - Mask
bone_thresh = 210
pad_value = 170
im[np.isnan(im)]=-2000
sliceim = lumTrans(im)
sliceim = sliceim*dilatedMask+pad_value*(1-dilatedMask).astype('uint8')
bones = sliceim*extramask>bone_thresh
sliceim[bones] = pad_value
sliceim1,_ = resample(sliceim,spacing,resolution,order=1)
sliceim2 = sliceim1[extendbox[0,0]:extendbox[0,1],
extendbox[1,0]:extendbox[1,1],
extendbox[2,0]:extendbox[2,1]]
sliceim = sliceim2[np.newaxis,...]
np.save(os.path.join(prep_folder,name+'_clean.npy'),sliceim)
if len(label)==0:
label2 = np.array([[0,0,0,0]])
elif len(label[0])==0:
label2 = np.array([[0,0,0,0]])
elif label[0][0]==0:
label2 = np.array([[0,0,0,0]])
else:
haslabel = 1
label2 = np.copy(label).T
label2[:3] = label2[:3][[0,2,1]]
label2[:3] = label2[:3]*np.expand_dims(spacing,1)/np.expand_dims(resolution,1)
label2[3] = label2[3]*spacing[1]/resolution[1]
label2[:3] = label2[:3]-np.expand_dims(extendbox[:,0],1)
label2 = label2[:4].T
np.save(os.path.join(prep_folder,name+'_label.npy'),label2)
print(name)
def savenpy(id,filelist,prep_folder,data_path,use_existing=True):
resolution = np.array([1,1,1])
name = filelist[id]
if use_existing:
if os.path.exists(os.path.join(prep_folder,name+'_label.npy')) and os.path.exists(os.path.join(prep_folder,name+'_clean.npy')):
print(name+' had been done')
return
try:
im, m1, m2, spacing = step1_python(os.path.join(data_path,name))
Mask = m1+m2
newshape = np.round(np.array(Mask.shape)*spacing/resolution)
xx,yy,zz= np.where(Mask)
box = np.array([[np.min(xx),np.max(xx)],[np.min(yy),np.max(yy)],[np.min(zz),np.max(zz)]])
box = box*np.expand_dims(spacing,1)/np.expand_dims(resolution,1)
box = np.floor(box).astype('int')
margin = 5
extendbox = np.vstack([np.max([[0,0,0],box[:,0]-margin],0),np.min([newshape,box[:,1]+2*margin],axis=0).T]).T
extendbox = extendbox.astype('int')
convex_mask = m1
dm1 = process_mask(m1)
dm2 = process_mask(m2)
dilatedMask = dm1+dm2
Mask = m1+m2
extramask = dilatedMask ^ Mask
bone_thresh = 210
pad_value = 170
im[np.isnan(im)]=-2000
sliceim = lumTrans(im)
sliceim = sliceim*dilatedMask+pad_value*(1-dilatedMask).astype('uint8')
bones = sliceim*extramask>bone_thresh
sliceim[bones] = pad_value
sliceim1,_ = resample(sliceim,spacing,resolution,order=1)
sliceim2 = sliceim1[extendbox[0,0]:extendbox[0,1],
extendbox[1,0]:extendbox[1,1],
extendbox[2,0]:extendbox[2,1]]
sliceim = sliceim2[np.newaxis,...]
np.save(os.path.join(prep_folder,name+'_clean'),sliceim)
np.save(os.path.join(prep_folder,name+'_label'),np.array([[0,0,0,0]]))
except:
print('bug in '+name)
raise
print(name+' done')
def savenpy(id, filelist, prep_folder, data_path, use_existing=True):
resolution = np.array([1, 1, 1])
name = filelist[id]
print 'lungSeg %s...' % name
#if use_existing:
if os.path.exists(
os.path.join(prep_folder,
name.split('.')[0] + '_clean.npy')):
print(name + ' had been done')
return
try:
#if(1):
im, m1, m2, spacing = step1_python(os.path.join(data_path, name))
# print m1.shape,m2.shape,spacing
if m1.shape[0] == 0:
print('save time: ' + name + ',using tutorial lungSeg...')
tutorialLungSeg.lungSeg(im, prep_folder, name.split('.')[0])
return
Mask = m1 + m2
newshape = np.round(np.array(Mask.shape) * spacing / resolution)
xx, yy, zz = np.where(Mask)
box = np.array([[np.min(xx), np.max(xx)], [np.min(yy),
np.max(yy)],
[np.min(zz), np.max(zz)]])
box = box * np.expand_dims(spacing, 1) / np.expand_dims(resolution, 1)
box = np.floor(box).astype('int')
margin = 5
extendbox = np.vstack([
np.max([[0, 0, 0], box[:, 0] - margin], 0),
np.min([newshape, box[:, 1] + 2 * margin], axis=0).T
]).T
extendbox = extendbox.astype('int')
convex_mask = m1
dm1 = process_mask(m1)
dm2 = process_mask(m2)
dilatedMask = dm1 + dm2
Mask = m1 + m2
extramask = dilatedMask ^ Mask
bone_thresh = 210
pad_value = 170
im[np.isnan(im)] = -2000
sliceim = lumTrans(im)
sliceim = sliceim * dilatedMask + pad_value * (
1 - dilatedMask).astype('uint8')
bones = sliceim * extramask > bone_thresh
sliceim[bones] = pad_value
#sliceim1,_ = resample(sliceim,spacing,resolution,order=1)
#sliceim2 = sliceim1[extendbox[0,0]:extendbox[0,1],
# extendbox[1,0]:extendbox[1,1],
# extendbox[2,0]:extendbox[2,1]]
#sliceim = sliceim2[np.newaxis,...]
x, y, z = sliceim.shape
np.save(os.path.join(prep_folder,
name.split('.')[0] + '_clean'),
sliceim.reshape(1, x, y, z))
np.save(os.path.join(prep_folder,
name.split('.')[0] + '_label'),
np.array([[0, 0, 0, 0]]))
except:
print('bug in ' + name + ',using tutorial lungSeg...')
itk_img = sitk.ReadImage(os.path.join(data_path, name))
im = sitk.GetArrayFromImage(itk_img)
tutorialLungSeg.lungSeg(im, prep_folder, name.split('.')[0])
#raise
print(name + ' done')
def savenpy(dirname, prep_folder, data_path, use_existing=True):
print('saving %s...' % dirname)
resolution = np.array([1, 1, 1])
if use_existing:
label_path = p.join(prep_folder, dirname + '_label.npy')
clean_path = p.join(prep_folder, dirname + '_clean.npy')
exists = p.exists(label_path) and p.exists(clean_path)
else:
exists = False
if exists:
print(dirname + ' already processed')
processed = 0
else:
print(dirname + ' not yet processed')
case_path = p.join(data_path, dirname)
im, m1, m2, spacing = step1_python(case_path)
Mask = m1 + m2
newshape = np.round(np.array(Mask.shape) * spacing / resolution)
xx, yy, zz = np.where(Mask)
box = np.array(
[
[np.min(xx), np.max(xx)],
[np.min(yy), np.max(yy)],
[np.min(zz), np.max(zz)]])
box = box * np.expand_dims(spacing,1) / np.expand_dims(resolution, 1)
box = np.floor(box).astype('int')
margin = 5
extendbox = np.vstack(
[
np.max([[0, 0, 0], box[:,0] - margin], 0),
np.min([newshape, box[:,1] + 2 * margin], axis=0).T]).T
extendbox = extendbox.astype('int')
convex_mask = m1
dm1 = process_mask(m1)
dm2 = process_mask(m2)
dilatedMask = dm1 + dm2
Mask = m1 + m2
extramask = dilatedMask ^ Mask
bone_thresh = 210
pad_value = 170
im[np.isnan(im)] =- 2000
sliceim = lumTrans(im)
sliceim = sliceim * dilatedMask + pad_value * (1 - dilatedMask).astype('uint8')
bones = sliceim * extramask > bone_thresh
sliceim[bones] = pad_value
sliceim1 = resample(sliceim, spacing, resolution, order=1)[0]
sliceim2 = sliceim1[
extendbox[0, 0]:extendbox[0, 1],
extendbox[1, 0]:extendbox[1, 1],
extendbox[2, 0]:extendbox[2, 1]]
sliceim = sliceim2[np.newaxis, ...]
np.save(p.join(prep_folder, dirname + '_clean'), sliceim)
np.save(p.join(prep_folder, dirname + '_label'), np.array([[0,0,0,0]]))
print(dirname + ' done')
processed = 1
return processed
def savenpy(id, annos, filelist, data_path, prep_folder):
resolution = np.array([1, 1, 1])
name = filelist[id]
shortname = name[:-7]
label = annos[annos[:, 0] == shortname]
label = label[:, [3, 1, 2, 4]].astype('float')
im, m1, m2, spacing, origin = step1_python(os.path.join(data_path, name))
Mask = m1 + m2
newshape = np.round(np.array(Mask.shape) * spacing / resolution)
xx, yy, zz = np.where(Mask)
box = np.array([[np.min(xx), np.max(xx)], [np.min(yy),
np.max(yy)],
[np.min(zz), np.max(zz)]])
box = box * np.expand_dims(spacing, 1) / np.expand_dims(resolution, 1)
box = np.floor(box).astype('int')
margin = 5
extendbox = np.vstack([
np.max([[0, 0, 0], box[:, 0] - margin], 0),
np.min([newshape, box[:, 1] + 2 * margin], axis=0).T
]).T
extendbox = extendbox.astype('int')
convex_mask = m1
dm1 = process_mask(m1)
dm2 = process_mask(m2)
dilatedMask = dm1 + dm2
Mask = m1 + m2
extramask = dilatedMask.astype(np.int) - Mask.astype(np.int)
bone_thresh = 210
pad_value = 170
im[np.isnan(im)] = -2000
sliceim = lumTrans(im)
sliceim = sliceim * dilatedMask + pad_value * (1 -
dilatedMask).astype('uint8')
bones = sliceim * extramask > bone_thresh
sliceim[bones] = pad_value
sliceim1, _ = resample(sliceim, spacing, resolution, order=1)
sliceim2 = sliceim1[extendbox[0, 0]:extendbox[0, 1],
extendbox[1, 0]:extendbox[1, 1],
extendbox[2, 0]:extendbox[2, 1]]
sliceim = sliceim2[np.newaxis, ...]
# save image into nii.gz to check
# import nibabel as nib
# new_image = nib.Nifti1Image(np.squeeze(sliceim), affine=np.eye(4))
# nib.save(new_image, os.path.join('./','test4d.nii.gz'))
np.save(os.path.join(prep_folder, shortname + '_clean.npy'), sliceim)
np.save(os.path.join(prep_folder, shortname + '_spacing.npy'), spacing)
np.save(os.path.join(prep_folder, shortname + '_extendbox.npy'), extendbox)
np.save(os.path.join(prep_folder, shortname + '_origin.npy'), origin)
np.save(os.path.join(prep_folder, shortname + '_mask.npy'), Mask)
if len(label) == 0:
label2 = np.array([[0, 0, 0, 0]])
else:
label2 = np.copy(label).T
label2[:3] = label2[:3][[0, 2, 1]]
label2[:3] = label2[:3] * np.expand_dims(spacing, 1) / np.expand_dims(
resolution, 1)
label2[3] = label2[3] * spacing[1] / resolution[1]
label2[:3] = label2[:3] - np.expand_dims(extendbox[:, 0], 1)
label2 = label2[:4].T
np.save(os.path.join(prep_folder, shortname + '_label.npy'), label2)
print(name)
def savenpy(id, filelist, prep_folder, data_path, use_existing=True):
resolution = np.array([1, 1, 1])
name = filelist[id]
if use_existing:
if os.path.exists(os.path.join(
prep_folder, name + '_label.npy')) and os.path.exists(
os.path.join(prep_folder, name + '_clean.npy')):
print(name + ' had been done')
return
try:
# im, m1, m2, spacing = step1_python(os.path.join(data_path,name))
im, m1, m2, spacing = step1_python(name)
Mask = m1 + m2
newshape = np.round(np.array(Mask.shape) * spacing / resolution)
xx, yy, zz = np.where(Mask)
box = np.array([[np.min(xx), np.max(xx)], [np.min(yy),
np.max(yy)],
[np.min(zz), np.max(zz)]])
box = box * np.expand_dims(spacing, 1) / np.expand_dims(resolution, 1)
box = np.floor(box).astype('int')
margin = 5
extendbox = np.vstack([
np.max([[0, 0, 0], box[:, 0] - margin], 0),
np.min([newshape, box[:, 1] + 2 * margin], axis=0).T
]).T
extendbox = extendbox.astype('int')
convex_mask = m1
dm1 = process_mask(m1)
dm2 = process_mask(m2)
dilatedMask = dm1 + dm2
Mask = m1 + m2
extramask = dilatedMask ^ Mask
bone_thresh = 210
pad_value = 170
im[np.isnan(im)] = -2000
sliceim = lumTrans(im)
sliceim = sliceim * dilatedMask + pad_value * (
1 - dilatedMask).astype('uint8')
bones = sliceim * extramask > bone_thresh
sliceim[bones] = pad_value
# sliceim1,_ = resample(sliceim,spacing,resolution,order=1)
# sliceim2 = sliceim1[extendbox[0,0]:extendbox[0,1],
# extendbox[1,0]:extendbox[1,1],
# extendbox[2,0]:extendbox[2,1]]
# sliceim = sliceim2[np.newaxis,...]
name, ext = os.path.splitext(os.path.split(name)[1])
# np.save(os.path.join(prep_folder, name + '.npy'), sliceim)
# np.save(os.path.join(prep_folder,name+'_clean'),sliceim)
# np.save(os.path.join(prep_folder,name+'_label'),np.array([[0,0,0,0]]))
# for idx, slice in enumerate(sliceim):
# slicename = "../output/test_masks/lung_masks/%d.jpg" % idx
# cv2.imwrite(slicename, slice)
#sliceim = sliceim.reshape(sliceim.shape[1], sliceim.shape[2], sliceim.shape[0])
#write_mhd_file(os.path.join(prep_folder, name + '.mhd'), sliceim, sliceim.shape)
np.save(os.path.join(prep_folder, name + '.npy'), sliceim)
except:
print('bug in ' + name)
raise
print(name + ' done')
def savenpy(id, filelist, prep_folder, data_path, use_existing=True):
resolution = np.array([1, 1, 1])
name = filelist[id]
if use_existing:
if os.path.exists(os.path.join(
prep_folder, name + '_label.npy')) and os.path.exists(
os.path.join(prep_folder, name + '_clean.npy')):
print(name + ' had been done')
return
try:
im, m1, m2, spacing = step1_python(os.path.join(data_path, name))
Mask = m1 + m2
newshape = np.round(np.array(Mask.shape) * spacing / resolution)
xx, yy, zz = np.where(Mask)
box = np.array([[np.min(xx), np.max(xx)], [np.min(yy),
np.max(yy)],
[np.min(zz), np.max(zz)]])
box = box * np.expand_dims(spacing, 1) / np.expand_dims(resolution, 1)
box = np.floor(box).astype('int')
margin = 5
extendbox = np.vstack([
np.max([[0, 0, 0], box[:, 0] - margin], 0),
np.min([newshape, box[:, 1] + 2 * margin], axis=0).T
]).T
extendbox = extendbox.astype('int')
convex_mask = m1
dm1 = process_mask(m1)
dm2 = process_mask(m2)
dilatedMask = dm1 + dm2
Mask = m1 + m2
extramask = dilatedMask ^ Mask
bone_thresh = 210
pad_value = 170
im[np.isnan(im)] = -2000
sliceim = lumTrans(im)
sliceim = sliceim * dilatedMask + pad_value * (
1 - dilatedMask).astype('uint8')
bones = sliceim * extramask > bone_thresh
sliceim[bones] = pad_value
sliceim1, _ = resample(sliceim, spacing, resolution, order=1)
sliceim2 = sliceim1[extendbox[0, 0]:extendbox[0, 1],
extendbox[1, 0]:extendbox[1, 1],
extendbox[2, 0]:extendbox[2, 1]]
sliceim = sliceim2[np.newaxis, ...]
#save as npy
np.save(os.path.join(prep_folder, name + '_clean'), sliceim)
np.save(os.path.join(prep_folder, name + '_label'),
np.array([[0, 0, 0, 0]]))
#Save as nifti
sliceim = sliceim.reshape(sliceim.shape[-3], sliceim.shape[-2],
sliceim.shape[-1])
matr = np.array([[0, 0, -1, 0], [0, -1, 0, 0], [1, 0, 0, 0],
[0, 0, 0, 1]])
ni_img = nib.Nifti1Image(sliceim, matr)
#activate the following line to keep conformity with old data
#ni_img = nib.Nifti1Image(sliceim, np.eye(4))
nib.save(ni_img, os.path.join(prep_folder, name + '_clean.nii.gz'))
except Exception as e:
print('bug in ' + name)
print(e)
#raise
print(name + ' done')
def savenpy(id,
filelist,
prep_folder,
data_path,
use_existing=True,
annos=None):
"""
stage数据集处理并保存为numpy的格式（单个filelist对应一个npy文件，包括label.npy和clean.npy）
:param id: os.listdir的list的index
:param filelist: os.listdir的结果，list格式
:param prep_folder: stage-2数据集处理过后的保存地址
:param data_path: stage-2数据集的下载地址
:param use_existing:
:return:
"""
resolution = np.array([1, 1, 1])
name = filelist[id]
print(id)
if use_existing:
if os.path.exists(os.path.join(
prep_folder, name + '_label.npy')) and os.path.exists(
os.path.join(prep_folder, name + '_clean.npy')):
print(name + ' had been done')
return
try:
# step1.py中的3D图像数据预处理函数step1_python，用来返回：
##im：像素点的HU值
##m1：mask以内的像素
##m2：mask以外的像素
##spacing：像素距离
im, m1, m2, spacing, origin, isflip = step1_python(
os.path.join(data_path, name))
## 真实的mask边界
Mask = m1 + m2
## mask世界CT坐标系到真实坐标系的转换
newshape = np.round(np.array(Mask.shape) * spacing / resolution)
## mask真实坐标系下的位置坐标
xx, yy, zz = np.where(Mask)
## 结节标注的边界，用一个刚好包裹下的cube-box代替，并padding部分像素
box = np.array([[np.min(xx), np.max(xx)], [np.min(yy),
np.max(yy)],
[np.min(zz), np.max(zz)]])
box = box * np.expand_dims(spacing, 1) / np.expand_dims(resolution, 1)
box = np.floor(box).astype('int')
margin = 5
extendbox = np.vstack([
np.max([[0, 0, 0], box[:, 0] - margin], 0),
np.min([newshape, box[:, 1] + 2 * margin], axis=0).T
]).T
extendbox = extendbox.astype('int')
# 凸包+扩张情况下的mask提取，并把mask以外的像素灰度值均设为170，为避免肺结节分类干扰，扩张区域内的像素灰度值高于210（骨组织）也设为170
convex_mask = m1
dm1 = process_mask(m1)
dm2 = process_mask(m2)
dilatedMask = dm1 + dm2
Mask = m1 + m2
extramask = dilatedMask ^ Mask
bone_thresh = 210
pad_value = 170
im[np.isnan(im)] = -2000
sliceim = lumTrans(im)
sliceim = sliceim * dilatedMask + pad_value * (
1 - dilatedMask).astype('uint8')
bones = sliceim * extramask > bone_thresh
sliceim[bones] = pad_value
sliceim1, _ = resample(sliceim, spacing, resolution, order=1)
sliceim2 = sliceim1[extendbox[0, 0]:extendbox[0, 1],
extendbox[1, 0]:extendbox[1, 1],
extendbox[2, 0]:extendbox[2, 1]]
sliceim = sliceim2[np.newaxis, ...]
np.save(os.path.join(prep_folder, name + '_clean'), sliceim)
# np.save(os.path.join(prep_folder,name+'_label'),np.array([[0,0,0,0]]))
this_annos = np.copy(
annos[annos[:, 0] == str(name)]) # 一行代表一个结节，一个病例可能对应多行标签
label = []
if len(this_annos) > 0:
for c in this_annos:
pos = worldToVoxelCoord(c[1:4][::-1],
origin=origin,
spacing=spacing) # 将世界坐标转换为真实的体素坐标
if isflip:
pos[1:] = Mask.shape[1:3] - pos[1:]
label.append(np.concatenate([pos, [c[4] / spacing[1]]
])) # 这1段dataframe的操作没看懂？？？
label = np.array(label)
if len(label) == 0:
label2 = np.array([[0, 0, 0,
0]]) # #若没有结节则设为全0，第1次读取label_numpy就是全设为0的问题
else:
label2 = np.copy(label).T
label2[:3] = label2[:3] * np.expand_dims(
spacing, 1) / np.expand_dims(resolution, 1) # 对标签应用新的分辨率
label2[3] = label2[3] * spacing[1] / resolution[1] # 对直径应用新的分辨率
label2[:3] = label2[:3] - np.expand_dims(
extendbox[:, 0], 1) # 将box外的长度砍掉，也就是相对于box的坐标
label2 = label2[:4].T
np.save(os.path.join(prep_folder, name + '_label.npy'), label2)
print(name, '\n')
except:
print('bug in ' + name)
raise
print(name + ' done')
def savenpy(id, annos, filelist, data_path, prep_folder):
print('id', id)
resolution = np.array([1, 1, 1])
name = filelist[id]
label = annos[annos[:, 0] == name]
label = label[:, [3, 1, 2, 4]].astype('float') #zxyd
im, m1, m2, spacing = step1_python(os.path.join(data_path,
name)) #preprocess
Mask = m1 + m2
# print('Mask.shape',Mask.shape)#(104, 512, 512)
newshape = np.round(np.array(Mask.shape) * spacing / resolution)
# print('newshape',newshape)#[260. 350. 350.]
xx, yy, zz = np.where(Mask)
box = np.array([[np.min(xx), np.max(xx)], [np.min(yy),
np.max(yy)],
[np.min(zz), np.max(zz)]])
# print('box0',box, box.shape)#[[ 30. 235.][85.44924855 287.79306912][42.38282728 307.61729479]] (3, 2)
box = box * np.expand_dims(spacing, 1) / np.expand_dims(
resolution, 1) #lung box np.expand_dims:(3,)to (3, 1)
# print('spacing',spacing, np.expand_dims(spacing,1))#[2.5 0.664062 0.664062] [2.5 ][0.664062][0.664062]]
# print('box1',box, box.shape)#[[45. 287.5][82.265625 303.046875][27.421875 336.796875]] (3, 2)
#
box = np.floor(box).astype('int')
margin = 5
extendbox = np.vstack([
np.max([[0, 0, 0], box[:, 0] - margin], 0),
np.min([newshape, box[:, 1] + 2 * margin], axis=0).T
]).T
extendbox = extendbox.astype('int')
convex_mask = m1
dm1 = process_mask(m1) #(凸包&扩张)
dm2 = process_mask(m2)
dilatedMask = dm1 + dm2
Mask = m1 + m2
# print('---0',im.shape,Mask.shape)
extramask = dilatedMask ^ Mask
bone_thresh = 210
pad_value = 170
im[np.isnan(im)] = -2000
sliceim = lumTrans(im) #归一化到0-255
sliceim = sliceim * dilatedMask + pad_value * (1 -
dilatedMask).astype('uint8')
bones = sliceim * extramask > bone_thresh
sliceim[bones] = pad_value
# print('---1',im.shape,sliceim.shape,Mask.shape)
sliceim1, _ = resample(sliceim, spacing, resolution, order=1)
Mask1, _ = resample(Mask, spacing, resolution, order=1)
sliceim2 = sliceim1[extendbox[0, 0]:extendbox[0, 1],
extendbox[1, 0]:extendbox[1, 1],
extendbox[2, 0]:extendbox[2, 1]]
Mask2 = Mask1[extendbox[0, 0]:extendbox[0, 1],
extendbox[1, 0]:extendbox[1, 1],
extendbox[2, 0]:extendbox[2, 1]]
sliceim = sliceim2[np.newaxis, ...]
Mask = Mask2[np.newaxis, ...]
# print('---1',im.shape,sliceim.shape,Mask.shape)
print('save mask clean label to', prep_folder)
np.save(os.path.join(prep_folder, name + '_mask.npy'), Mask)
np.save(os.path.join(prep_folder, name + '_clean.npy'), sliceim)
if len(label) == 0:
label2 = np.array([[0, 0, 0, 0]])
elif len(label[0]) == 0:
label2 = np.array([[0, 0, 0, 0]])
elif label[0][0] == 0:
label2 = np.array([[0, 0, 0, 0]])
else:
haslabel = 1
label2 = np.copy(label).T
label2[:3] = label2[:3][[0, 2, 1]]
label2[:3] = label2[:3] * np.expand_dims(spacing, 1) / np.expand_dims(
resolution, 1)
label2[3] = label2[3] * spacing[1] / resolution[1]
label2[:3] = label2[:3] - np.expand_dims(extendbox[:, 0], 1)
label2 = label2[:4].T
np.save(os.path.join(prep_folder, name + '_label.npy'), label2)
print(name)
def savenpy(id, annos, filelist, data_path, prep_folder):
"""
图像预处理重点-提取肺部区域ROI，以及提取的数据保存为numpy
:param id:
:param annos:
:param filelist:
:param data_path:
:param prep_folder:
:return:
"""
resolution = np.array([1, 1, 1])
name = filelist[id]
label = annos[annos[:, 0] == name]
# z轴坐标放在第一位
label = label[:, [3, 1, 2, 4]].astype('float')
# 切片后，im是像素点HU值，m1是只有目标标注的mask像素值，m2是只有非目标区域标的的mask像素值，spacing是体素距离
im, m1, m2, spacing = step1_python(os.path.join(data_path, name))
Mask = m1 + m2
# mask世界坐标系到真实坐标系的转换
newshape = np.round(np.array(Mask.shape) * spacing / resolution)
xx, yy, zz = np.where(Mask)
# box立方体的坐标表示形式
box = np.array([[np.min(xx), np.max(xx)], [np.min(yy),
np.max(yy)],
[np.min(zz), np.max(zz)]])
# np.expand_dims：用于扩展数组的形状，spacing.shape=（3，2），resolution.shape=（3，2）
box = box * np.expand_dims(spacing, 1) / np.expand_dims(resolution, 1)
# np.floor：返回不大于输入参数的最大整数
box = np.floor(box).astype('int')
# 每个轴左右方向上的扩张像素值
margin = 5
# box向外扩张10个像素
extendbox = np.vstack([
np.max([[0, 0, 0], box[:, 0] - margin], 0),
np.min([newshape, box[:, 1] + 2 * margin], axis=0).T
]).T
extendbox = extendbox.astype('int')
convex_mask = m1
dm1 = process_mask(m1)
dm2 = process_mask(m2)
# 膨胀后的mask
dilatedMask = dm1 + dm2
# 原始的mask
Mask = m1 + m2
# 膨胀新增区域
extramask = dilatedMask - Mask
# 像素灰度值阈值（骨头区域）
bone_thresh = 210
# mask以外区域的像素值
pad_value = 170
# HU值初始化为（-1000，1000）？
im[np.isnan(im)] = -2000
# HU值（[-1200, 600]）线性变换至0~255内的灰度值
sliceim = lumTrans(im)
# 切片img中mask以外的像素灰度值均设为170
sliceim = sliceim * dilatedMask + pad_value * (1 -
dilatedMask).astype('uint8')
# 切片img中大于像素灰度值阈值的像素点（骨头区域）
bones = sliceim * extramask > bone_thresh
# 切片img中像素灰度值高于210的区域灰度值也设为170
sliceim[bones] = pad_value
# 像素距离标准化（体素值进行映射与归一化）
sliceim1, _ = resample(sliceim, spacing, resolution, order=1)
# 左右肺的凸包扩张处理
sliceim2 = sliceim1[extendbox[0, 0]:extendbox[0, 1],
extendbox[1, 0]:extendbox[1, 1],
extendbox[2, 0]:extendbox[2, 1]]
sliceim = sliceim2[np.newaxis, ...]
# name+'_clean.npy'：就是去除肺部周边噪声后的，3D切片图数组
np.save(os.path.join(prep_folder, name + '_clean.npy'), sliceim)
# name+'_label.npy'：2D切片图相应的label数组
if len(label) == 0:
label2 = np.array([[0, 0, 0, 0]])
elif len(label[0]) == 0:
label2 = np.array([[0, 0, 0, 0]])
elif label[0][0] == 0:
label2 = np.array([[0, 0, 0, 0]])
else:
haslabel = 1
label2 = np.copy(label).T
label2[:3] = label2[:3][[0, 2, 1]]
label2[:3] = label2[:3] * np.expand_dims(spacing, 1) / np.expand_dims(
resolution, 1)
label2[3] = label2[3] * spacing[1] / resolution[1]
label2[:3] = label2[:3] - np.expand_dims(extendbox[:, 0], 1)
label2 = label2[:4].T
np.save(os.path.join(prep_folder, name + '_label.npy'), label2)
print(name)
