def __init__(self, series_uid):
mhd_path = glob.glob(
"data-unversioned/part2/luna/subset*/{}.mhd".format(series_uid))[0]
ct_mhd = sitk.ReadImage(mhd_path)
self.hu_a = np.array(sitk.GetArrayFromImage(ct_mhd), dtype=np.float32)
# CTs are natively expressed in https://en.wikipedia.org/wiki/Hounsfield_scale
# HU are scaled oddly, with 0 g/cc (air, approximately) being -1000 and 1 g/cc (water) being 0.
self.series_uid = series_uid
self.origin_xyz = XyzTuple(*ct_mhd.GetOrigin())
self.vxSize_xyz = XyzTuple(*ct_mhd.GetSpacing())
self.direction_a = np.array(ct_mhd.GetDirection()).reshape(3, 3)
candidateInfo_list = getCandidateInfoDict()[self.series_uid]
self.positiveInfo_list = [
candidate_tup for candidate_tup in candidateInfo_list
if candidate_tup.isNodule_bool
]
self.positive_mask = self.buildAnnotationMask(self.positiveInfo_list)
self.positive_indexes = (self.positive_mask.sum(
axis=(1, 2)).nonzero()[0].tolist())
def __init__(self, series_uid):
mhd_path = glob.glob("data-unversioned/part2/luna/subset*/{}.mhd"\
.format(series_uid))[0]
ct_mhd = sitk.ReadImage(mhd_path)
ct_a = np.array(sitk.GetArrayFromImage(ct_mhd), dtype=np.float32)
# CTs are natively expressed in
# https://en.wikipedia.org/wiki/Hounsfield_scale HU are scaled
# oddly, with 0 g/cc (air, approximately) being -1000 and 1 g/cc
# (water) being 0.
# This gets rid of negative density stuff used to indicate
# out-of-FOV
ct_a[ct_a < -1000] = -1000
# This nukes any weird hotspots and clamps bone down
ct_a[ct_a > 1000] = 1000
self.series_uid = series_uid
self.hu_a = ct_a
self.origin_xyz = XyzTuple(*ct_mhd.GetOrigin())
self.vxSize_xyz = XyzTuple(*ct_mhd.GetSpacing())
self.direction_tup = tuple(int(round(x)) for x in
ct_mhd.GetDirection())
def __init__(self, series_uid):
mhd_path = glob.glob('data/luna/subset*/{}.mhd'.format(series_uid))[0]
ct_mhd = sitk.ReadImage(mhd_path)
ct_ary = np.array(sitk.GetArrayFromImage(ct_mhd), dtype=np.float32)
# CTs are natively expressed in https://en.wikipedia.org/wiki/Hounsfield_scale
# HU are scaled oddly, with 0 g/cc (air, approximately) being -1000 and 1 g/cc (water) being 0.
# This converts HU to g/cc.
ct_ary += 1000
ct_ary /= 1000
# This gets rid of negative density stuff used to indicate out-of-FOV
ct_ary[ct_ary < 0] = 0
# This nukes any weird hotspots and clamps bone down
ct_ary[ct_ary > 2] = 2
self.series_uid = series_uid
self.ary = ct_ary
self.origin_xyz = XyzTuple(*ct_mhd.GetOrigin())
self.vxSize_xyz = XyzTuple(*ct_mhd.GetSpacing())
self.direction_tup = tuple(
int(round(x)) for x in ct_mhd.GetDirection())
def __init__(self, series_uid, buildMasks_bool=True):
mhd_path = glob.glob('data-unversioned/part2/luna/subset*/{}.mhd'.format(series_uid))[0]
ct_mhd = sitk.ReadImage(mhd_path)
ct_a = np.array(sitk.GetArrayFromImage(ct_mhd), dtype=np.float32)
# CTs are natively expressed in https://en.wikipedia.org/wiki/Hounsfield_scale
# HU are scaled oddly, with 0 g/cc (air, approximately) being -1000 and 1 g/cc (water) being 0.
# This gets rid of negative density stuff used to indicate out-of-FOV
ct_a[ct_a < -1000] = -1000
# This nukes any weird hotspots and clamps bone down
ct_a[ct_a > 1000] = 1000
self.series_uid = series_uid
self.hu_a = ct_a
self.origin_xyz = XyzTuple(*ct_mhd.GetOrigin())
self.vxSize_xyz = XyzTuple(*ct_mhd.GetSpacing())
self.direction_tup = tuple(int(round(x)) for x in ct_mhd.GetDirection())
noduleInfo_list = getNoduleInfoList()
self.benignInfo_list = [ni_tup
for ni_tup in noduleInfo_list
if not ni_tup.isMalignant_bool
and ni_tup.series_uid == self.series_uid]
self.benign_mask = self.buildAnnotationMask(self.benignInfo_list)[0]
self.benign_indexes = sorted(set(self.benign_mask.nonzero()[0]))
self.malignantInfo_list = [ni_tup
for ni_tup in noduleInfo_list
if ni_tup.isMalignant_bool
and ni_tup.series_uid == self.series_uid]
self.malignant_mask = self.buildAnnotationMask(self.malignantInfo_list)[0]
self.malignant_indexes = sorted(set(self.malignant_mask.nonzero()[0]))
def __init__(self, series_uid, buildMasks_bool=True):
mhd_path = glob.glob(
"data-unversioned/part2/luna/subset*/{}.mhd".format(series_uid))[0]
ct_mhd = sitk.ReadImage(mhd_path)
ct_a = np.array(sitk.GetArrayFromImage(ct_mhd), dtype=np.float32)
# CTs are natively expressed in
# https://en.wikipedia.org/wiki/Hounsfield_scale HU are scaled
# oddly, with 0 g/cc (air, approximately) being -1000 and 1 g/cc
# (water) being 0.
# This gets rid of negative density stuff used to indicate
# out-of-FOV
ct_a[ct_a < -1000] = -1000
# This nukes any weird hotspots and clamps bone down
ct_a[ct_a > 1000] = 1000
self.series_uid = series_uid
self.hu_a = ct_a
# Origin: coordinates of the pixel/voxel with index (0,0,0) in
# physical units (i.e. mm). Default is zero i.e. origin of
# physical space.
# Spacing: distance between adjacent pixels/voxels in each
# dimension given in physical units. Default is one i.e.
# (1 mm, 1 mm, 1 mm).
# Direction Matrix: mapping/rotation between direction of the
# pixel/voxel axes and physical directions. Default is identity
# matrix. The matrix is passed as a 1D array in row-major form.
self.origin_xyz = XyzTuple(*ct_mhd.GetOrigin())
self.vxSize_xyz = XyzTuple(*ct_mhd.GetSpacing())
self.direction_tup = tuple(
int(round(x)) for x in ct_mhd.GetDirection())
noduleInfo_list = getNoduleInfoList()
# List of benign nodules for a specific CT scan
self.benignInfo_list = [
ni_tup for ni_tup in noduleInfo_list if not ni_tup.isMalignant_bool
and ni_tup.series_uid == self.series_uid
]
self.benign_mask = self.buildAnnotationMask(self.benignInfo_list)[0]
self.benign_indexes = sorted(set(self.benign_mask.nonzero()[0]))
# List of malignant nodules for a specific CT scan
self.malignantInfo_list = [
ni_tup for ni_tup in noduleInfo_list
if ni_tup.isMalignant_bool and ni_tup.series_uid == self.series_uid
]
self.malignant_mask = self.buildAnnotationMask(
self.malignantInfo_list)[0]
self.malignant_indexes = sorted(set(self.malignant_mask.nonzero()[0]))
def __init__(self, series_uid):
mhd_path = glob.glob('/content/drive/MyDrive/LUNA 16/Dataset/{}.mhd'.format(series_uid))[0]
#sitk.ReadImage implicitly consumes the .raw file in addition to the passed in .mhd file.
ct_mhd = sitk.ReadImage(mhd_path)
#Creates a numpy array
ct_array = np.array(sitk.GetArrayFromImage(ct_mhd), dtype=np.float32)
#pixel clipping
ct_array.clip(-1000, 1000, ct_array)
self.series_uid = series_uid
self.hu_array = ct_array
self.origin_xyz = XyzTuple(*ct_mhd.GetOrigin())
self.vxSize_xyz = XyzTuple(*ct_mhd.GetSpacing())
self.direction_a = np.array(ct_mhd.GetDirection()).reshape(3, 3)
def __init__(self, series_uid):
mhd_path = glob.glob(DATA_DIR + 'subset*/{}.mhd'.format(series_uid))[0]
ct_mhd = sitk.ReadImage(mhd_path)
ct_a = np.array(sitk.GetArrayFromImage(ct_mhd), dtype=np.float32)
# CTs are natively expressed in https://en.wikipedia.org/wiki/Hounsfield_scale
# HU are scaled oddly, with 0 g/cc (air, approximately) being -1000 and 1 g/cc (water) being 0.
# The lower bound gets rid of negative density stuff used to indicate out-of-FOV
# The upper bound nukes any weird hotspots and clamps bone down
ct_a.clip(-1000, 1000, ct_a)
self.series_uid = series_uid
self.hu_a = ct_a
self.origin_xyz = XyzTuple(*ct_mhd.GetOrigin())
self.vxSize_xyz = XyzTuple(*ct_mhd.GetSpacing())
self.direction_a = np.array(ct_mhd.GetDirection()).reshape(3, 3)
def __init__(self, series_uid):
mhd_path = glob.glob(
'data-unversioned/part2/luna/subset*/{}.mhd'.format(series_uid)
)[0]
#加载ct图像，转化为numpy数组
ct_mhd = sitk.ReadImage(mhd_path)#除了传入的.mhd文件之外，sitk.ReadImage 还会隐式使用.raw文件
ct_a = np.array(sitk.GetArrayFromImage(ct_mhd), dtype=np.float32)#将读取出来的图像信息用像素值表示出来
# CTs are natively expressed in https://en.wikipedia.org/wiki/Hounsfield_scale
# HU are scaled oddly, with 0 g/cc (air, approximately) being -1000 and 1 g/cc (water) being 0.
# The lower bound gets rid of negative density stuff used to indicate out-of-FOV
# The upper bound nukes any weird hotspots and clamps bone down
ct_a.clip(-1000, 1000, ct_a)#将ct_a数组中的元素值限定在(-1000, 1000)
self.series_uid = series_uid
self.hu_a = ct_a
#对下面三个参数的详细解释 https://blog.csdn.net/qq_39071739/article/details/106487597?utm_medium=distribute.wap_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-2.wap_blog_relevant_pic2&depth_1-utm_source=distribute.wap_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-2.wap_blog_relevant_pic2
self.origin_xyz = XyzTuple(*ct_mhd.GetOrigin())#读取图像的原点信息
self.vxSize_xyz = XyzTuple(*ct_mhd.GetSpacing())#图像各维度上像素之间的物理距离（单位一般为mm)
self.direction_a = np.array(ct_mhd.GetDirection()).reshape(3, 3)#采用方向余弦矩阵，这里是指图像本身坐标系相对于世界坐标系（固定不动的）的角度余弦
