def generate(self):
while True:
rand_idxs = np.arange(self.nsamples)
if self.random:
self.rng.shuffle(rand_idxs)
for pos in xrange(0, len(rand_idxs), self.batch_size):
idxs_batch = rand_idxs[pos:pos + self.batch_size]
nb = len(idxs_batch)
# allocate batches
x_ch2_batch = np.zeros(
(nb, 30) + self.transformation_params['patch_size'],
dtype='float32')
x_ch4_batch = np.zeros(
(nb, 30) + self.transformation_params['patch_size'],
dtype='float32')
y0_batch = np.zeros((nb, 1), dtype='float32')
y1_batch = np.zeros((nb, 1), dtype='float32')
patients_ids = []
for i, idx in enumerate(idxs_batch):
pid = self.patient_ids[idx]
patients_ids.append(pid)
# do everything with sax
sax_slice_paths = self.pid2sax_slice_paths[pid]
sax_slicepath2metadata = {}
sax_slicepath2roi = {}
for s in sax_slice_paths:
sax_metadata = data.read_metadata(s)
sax_slicepath2metadata[s] = sax_metadata
sid = utils.get_slice_id(s)
roi = self.slice2roi[str(pid)][sid]
sax_slicepath2roi[s] = roi
# read ch2, ch4
if self.pid2ch2_path[pid]:
data_ch2 = data.read_slice(self.pid2ch2_path[pid])
metadata_ch2 = data.read_metadata(
self.pid2ch2_path[pid])
else:
data_ch2, metadata_ch2 = None, None
if self.pid2ch4_path[pid]:
data_ch4 = data.read_slice(self.pid2ch4_path[pid])
metadata_ch4 = data.read_metadata(
self.pid2ch4_path[pid])
else:
data_ch4, metadata_ch4 = None, None
if data_ch2 is None and data_ch4 is not None:
data_ch2 = data_ch4
if data_ch4 is None and data_ch2 is not None:
data_ch4 = data_ch2
# sample augmentation params per patient
random_params = data.sample_augmentation_parameters(
self.transformation_params)
x_ch2_batch[i], x_ch4_batch[
i], targets_zoom = data.transform_ch(
data_ch2,
metadata_ch2,
data_ch4,
metadata_ch4,
saxslice2metadata=sax_slicepath2metadata,
transformation=self.transformation_params,
sax2roi=sax_slicepath2roi,
random_augmentation_params=random_params)
if self.id2labels:
y0_batch[i] = self.id2labels[pid][0] * targets_zoom
y1_batch[i] = self.id2labels[pid][1] * targets_zoom
if self.full_batch:
if nb == self.batch_size:
yield [x_ch2_batch,
x_ch4_batch], [y0_batch, y1_batch], patients_ids
else:
yield [x_ch2_batch, x_ch4_batch], [y0_batch,
y1_batch], patients_ids
if not self.infinite:
break
def generate(self):
while True:
rand_idxs = np.arange(self.nsamples)
if self.random:
self.rng.shuffle(rand_idxs)
for pos in xrange(0, len(rand_idxs), self.batch_size):
idxs_batch = rand_idxs[pos:pos + self.batch_size]
nb = len(idxs_batch)
# allocate batches
x_ch2_batch = np.zeros((nb, 30) + self.transformation_params['patch_size'],
dtype='float32')
x_ch4_batch = np.zeros((nb, 30) + self.transformation_params['patch_size'],
dtype='float32')
y0_batch = np.zeros((nb, 1), dtype='float32')
y1_batch = np.zeros((nb, 1), dtype='float32')
patients_ids = []
for i, idx in enumerate(idxs_batch):
pid = self.patient_ids[idx]
patients_ids.append(pid)
# do everything with sax
sax_slice_paths = self.pid2sax_slice_paths[pid]
sax_slicepath2metadata = {}
sax_slicepath2roi = {}
for s in sax_slice_paths:
sax_metadata = data.read_metadata(s)
sax_slicepath2metadata[s] = sax_metadata
sid = utils.get_slice_id(s)
roi = self.slice2roi[str(pid)][sid]
sax_slicepath2roi[s] = roi
# read ch2, ch4
if self.pid2ch2_path[pid]:
data_ch2 = data.read_slice(self.pid2ch2_path[pid])
metadata_ch2 = data.read_metadata(self.pid2ch2_path[pid])
else:
data_ch2, metadata_ch2 = None, None
if self.pid2ch4_path[pid]:
data_ch4 = data.read_slice(self.pid2ch4_path[pid])
metadata_ch4 = data.read_metadata(self.pid2ch4_path[pid])
else:
data_ch4, metadata_ch4 = None, None
if data_ch2 is None and data_ch4 is not None:
data_ch2 = data_ch4
if data_ch4 is None and data_ch2 is not None:
data_ch4 = data_ch2
# sample augmentation params per patient
random_params = data.sample_augmentation_parameters(self.transformation_params)
x_ch2_batch[i], x_ch4_batch[i], targets_zoom = data.transform_ch(data_ch2, metadata_ch2,
data_ch4, metadata_ch4,
saxslice2metadata=sax_slicepath2metadata,
transformation=self.transformation_params,
sax2roi=sax_slicepath2roi,
random_augmentation_params=random_params)
if self.id2labels:
y0_batch[i] = self.id2labels[pid][0] * targets_zoom
y1_batch[i] = self.id2labels[pid][1] * targets_zoom
if self.full_batch:
if nb == self.batch_size:
yield [x_ch2_batch, x_ch4_batch], [y0_batch, y1_batch], patients_ids
else:
yield [x_ch2_batch, x_ch4_batch], [y0_batch, y1_batch], patients_ids
if not self.infinite:
break
def generate(self):
while True:
rand_idxs = np.arange(self.nsamples)
if self.random:
self.rng.shuffle(rand_idxs)
for pos in xrange(0, len(rand_idxs), self.batch_size):
idxs_batch = rand_idxs[pos:pos + self.batch_size]
nb = len(idxs_batch)
# allocate batches
x_batch = np.zeros((nb, self.nslices, 30) +
self.transformation_params['patch_size'],
dtype='float32')
sex_age_batch = np.zeros((nb, 2), dtype='float32')
slice_location_batch = np.zeros((nb, self.nslices, 1),
dtype='float32')
slice_mask_batch = np.zeros((nb, self.nslices),
dtype='float32')
y0_batch = np.zeros((nb, 1), dtype='float32')
y1_batch = np.zeros((nb, 1), dtype='float32')
patients_ids = []
for i, idx in enumerate(idxs_batch):
pid = self.patient_ids[idx]
patients_ids.append(pid)
slice_paths = self.pid2slice_paths[pid]
# fill metadata dict for linefinder code and sort slices
slicepath2metadata = {}
for sp in slice_paths:
slicepath2metadata[sp] = data.read_metadata(sp)
slicepath2location = data.slice_location_finder(
slicepath2metadata)
slice_paths = sorted(slicepath2location.keys(),
key=slicepath2location.get)
# sample augmentation params per patient
random_params = data.sample_augmentation_parameters(
self.transformation_params)
for j, sp in enumerate(slice_paths):
slice_roi = self.slice2roi[str(pid)][
utils.get_slice_id(sp)] if self.slice2roi else None
slice_data = data.read_slice(sp)
x_batch[i, j], targets_zoom = self.data_prep_fun(
slice_data,
slicepath2metadata[sp],
self.transformation_params,
roi=slice_roi,
random_augmentation_params=random_params)
slice_location_batch[i, j] = slicepath2location[sp]
slice_mask_batch[i, j] = 1.
sex_age_batch[i, 0] = slicepath2metadata[
slice_paths[0]]['PatientSex']
sex_age_batch[i, 1] = slicepath2metadata[
slice_paths[0]]['PatientAge']
if self.id2labels:
y0_batch[i] = self.id2labels[pid][0] * targets_zoom
y1_batch[i] = self.id2labels[pid][1] * targets_zoom
if self.full_batch:
if nb == self.batch_size:
yield [
x_batch, slice_mask_batch, slice_location_batch,
sex_age_batch
], [y0_batch, y1_batch], patients_ids
else:
yield [
x_batch, slice_mask_batch, slice_location_batch,
sex_age_batch
], [y0_batch, y1_batch], patients_ids
if not self.infinite:
break
def generate(self):
while True:
rand_idxs = np.arange(self.nsamples)
if self.random:
self.rng.shuffle(rand_idxs)
for pos in xrange(0, len(rand_idxs), self.batch_size):
idxs_batch = rand_idxs[pos:pos + self.batch_size]
nb = len(idxs_batch)
# allocate batches
x_batch = np.zeros((nb, self.nslices, 30) + self.transformation_params['patch_size'],
dtype='float32')
sex_age_batch = np.zeros((nb, 2), dtype='float32')
slice_location_batch = np.zeros((nb, self.nslices, 1), dtype='float32')
slice_mask_batch = np.zeros((nb, self.nslices), dtype='float32')
y0_batch = np.zeros((nb, 1), dtype='float32')
y1_batch = np.zeros((nb, 1), dtype='float32')
patients_ids = []
for i, idx in enumerate(idxs_batch):
pid = self.patient_ids[idx]
patients_ids.append(pid)
slice_paths = self.pid2slice_paths[pid]
# fill metadata dict for linefinder code and sort slices
slicepath2metadata = {}
for sp in slice_paths:
slicepath2metadata[sp] = data.read_metadata(sp)
slicepath2location = data.slice_location_finder(slicepath2metadata)
slice_paths = sorted(slicepath2location.keys(), key=slicepath2location.get)
# sample augmentation params per patient
random_params = data.sample_augmentation_parameters(self.transformation_params)
for j, sp in enumerate(slice_paths):
slice_roi = self.slice2roi[str(pid)][
utils.get_slice_id(sp)] if self.slice2roi else None
slice_data = data.read_slice(sp)
x_batch[i, j], targets_zoom = self.data_prep_fun(slice_data, slicepath2metadata[sp],
self.transformation_params,
roi=slice_roi,
random_augmentation_params=random_params)
slice_location_batch[i, j] = slicepath2location[sp]
slice_mask_batch[i, j] = 1.
sex_age_batch[i, 0] = slicepath2metadata[slice_paths[0]]['PatientSex']
sex_age_batch[i, 1] = slicepath2metadata[slice_paths[0]]['PatientAge']
if self.id2labels:
y0_batch[i] = self.id2labels[pid][0] * targets_zoom
y1_batch[i] = self.id2labels[pid][1] * targets_zoom
if self.full_batch:
if nb == self.batch_size:
yield [x_batch, slice_mask_batch, slice_location_batch, sex_age_batch], [y0_batch,
y1_batch], patients_ids
else:
yield [x_batch, slice_mask_batch, slice_location_batch, sex_age_batch], [y0_batch,
y1_batch], patients_ids
if not self.infinite:
break