Skip to content

Commit 6086483

Browse files
MMelQinMMelQin
committed
Address mem usage concerns by processing and capping each slice's ndarry to min size.
Signed-off-by: M Q <[email protected]>
1 parent c740e36 commit 6086483

File tree

1 file changed

+68
-57
lines changed

1 file changed

+68
-57
lines changed

‎monai/deploy/operators/dicom_series_to_volume_operator.py‎

Lines changed: 68 additions & 57 deletions
Original file line numberDiff line numberDiff line change
@@ -18,6 +18,7 @@
1818

1919
from monai.deploy.utils.importutil import optional_import
2020

21+
apply_presentation_lut, _ = optional_import("pydicom.pixels", name="apply_presentation_lut")
2122
apply_rescale, _ = optional_import("pydicom.pixels", name="apply_rescale")
2223

2324
from monai.deploy.core import ConditionType, Fragment, Operator, OperatorSpec
@@ -109,72 +110,82 @@ def generate_voxel_data(self, series):
109110
Returns:
110111
A 3D numpy tensor representing the volumetric data.
111112
"""
113+
114+
def _get_rescaled_pixel_array(sop_instance):
115+
# Use pydicom utility to apply a modality lookup table or rescale operator to the pixel array.
116+
# The pydicom Dataset is required which can be obtained from the first slice's native SOP instance.
117+
# If Modality LUT is present the return array is of np.uint8 or np.uint16, and if Rescale
118+
# Intercept and Rescale Slope are present, np.float64.
119+
# If the pixel array is already in the correct type, the return array is the same as the input array.
120+
121+
if not sop_instance:
122+
return np.array([])
123+
124+
native_sop = None
125+
try:
126+
native_sop = sop_instance.get_native_sop_instance()
127+
rescaled_pixel_data = apply_rescale(sop_instance.get_pixel_array(), native_sop)
128+
# In our use cases, pixel data will be interpreted as if MONOCHROME2, hence need to
129+
# apply the presentation lut.
130+
rescaled_pixel_data = apply_presentation_lut(rescaled_pixel_data, native_sop)
131+
except Exception as e:
132+
logging.error(f"Failed to apply rescale to DICOM volume: {e}")
133+
raise RuntimeError("Failed to apply rescale to DICOM volume.") from e
134+
135+
# The following tests are expecting the array is already of the Numpy type.
136+
if rescaled_pixel_data.dtype == np.uint8 or rescaled_pixel_data.dtype == np.uint16:
137+
logging.info("Rescaled pixel array is alreadydy of type uint8 or uint16.")
138+
# Check if casting to uint16 and back to float results in the same values.
139+
elif np.all(rescaled_pixel_data > 0) and np.array_equal(
140+
rescaled_pixel_data, rescaled_pixel_data.astype(np.uint16)
141+
):
142+
logging.info("Rescaled pixel array can be safely cast to uint16 with equivalence test.")
143+
rescaled_pixel_data = rescaled_pixel_data.astype(dtype=np.uint16)
144+
# Check if casting to int16 and back to float results in the same values.
145+
elif np.array_equal(rescaled_pixel_data, rescaled_pixel_data.astype(np.int16)):
146+
logging.info("Rescaled pixel array can be safely cast to int16 with equivalence test.")
147+
rescaled_pixel_data = rescaled_pixel_data.astype(dtype=np.int16)
148+
# Check casting to float32 with equivalence test
149+
elif np.array_equal(rescaled_pixel_data, rescaled_pixel_data.astype(np.float32)):
150+
logging.info("Rescaled pixel array can be cast to float32 with equivalence test.")
151+
rescaled_pixel_data = rescaled_pixel_data.astype(np.float32)
152+
else:
153+
logging.info("Rescaled pixel data remains as of type float64.")
154+
155+
return rescaled_pixel_data
156+
112157
slices = series.get_sop_instances()
113158
# The sop_instance get_pixel_array() returns a 2D NumPy array with index order
114159
# of `HW`. The pixel array of all instances will be stacked along the first axis,
115160
# so the final 3D NumPy array will have index order of [DHW]. This is consistent
116161
# with the NumPy array returned from the ITK GetArrayViewFromImage on the image
117162
# loaded from the same DICOM series.
118-
vol_data = np.stack([s.get_pixel_array() for s in slices], axis=0)
163+
# The below code loads all slice pixel data into a list of NumPy arrays in memory
164+
# before stacking them into a single 3D volume. This can be inefficient for series
165+
# with many slices.
166+
if not slices:
167+
return np.array([])
119168

120-
# Use pydicom utility to apply a modality lookup table or rescale operator to the pixel array.
121-
# The pydicom Dataset is required which can be obtained from the first slice's native SOP instance.
122-
# If Modality LUT is present the return array is of np.uint8 or np.uint16, and if Rescale
123-
# Intercept and Rescale Slope are present, np.float64.
124-
# If the pixel array is already in the correct type, the return array is the same as the input array.
169+
# Get shape and dtype from the first slice to pre-allocate numpy array.
125170
try:
126-
native_sop = slices[0].get_native_sop_instance()
127-
vol_data = apply_rescale(vol_data, native_sop)
171+
first_slice_pixel_array = _get_rescaled_pixel_array(slices[0])
172+
vol_shape = (len(slices),) + first_slice_pixel_array.shape
173+
dtype = first_slice_pixel_array.dtype
128174
except Exception as e:
129-
logging.error(f"Failed to apply rescale to DICOM volume: {e}")
130-
raise RuntimeError("Failed to apply rescale to DICOM volume.") from e
131-
132-
# For now we support monochrome image only, for which DICOM Photometric Interpretation
133-
# (0028,0004) has defined terms, MONOCHROME1 and MONOCHROME2, with the former being:
134-
# Pixel data represent a single monochrome image plane. The minimum sample value is
135-
# intended to be displayed as white after any VOI gray scale transformations have been
136-
# performed. See PS3.4. This value may be used only when Samples per Pixel (0028,0002)
137-
# has a value of 1. May be used for pixel data in a Native (uncompressed) or Encapsulated
138-
# (compressed) format; see Section 8.2 in PS3.5.
139-
# and for the latter "The minimum sample value is intended to be displayed as black"
140-
#
141-
# In this function, pixel data will be interpreted as if MONOCHROME2, hence inverting
142-
# MONOCHROME1 for the final voxel data.
143-
144-
photometric_interpretation = (
145-
slices[0].get_native_sop_instance().get("PhotometricInterpretation", "").strip().upper()
146-
)
147-
presentation_lut_shape = slices[0].get_native_sop_instance().get("PresentationLUTShape", "").strip().upper()
148-
149-
if not photometric_interpretation:
150-
logging.warning("Cannot get value of attribute Photometric Interpretation.")
151-
152-
if photometric_interpretation != "MONOCHROME2":
153-
if photometric_interpretation == "MONOCHROME1" or presentation_lut_shape == "INVERSE":
154-
logging.debug("Applying INVERSE transformation as required for MONOCHROME1 image.")
155-
vol_data = np.amax(vol_data) - vol_data
156-
else:
157-
raise ValueError(
158-
f"Cannot process pixel data with Photometric Interpretation of {photometric_interpretation}."
159-
)
160-
161-
# The following tests are expecting the array is already of the Numpy type.
162-
if vol_data.dtype == np.uint8 or vol_data.dtype == np.uint16:
163-
logging.info("Rescaled pixel array is alrady of type uint8 or uint16.")
164-
# Check if casting to uint16 and back to float results in the same values.
165-
elif np.all(vol_data > 0) and np.array_equal(vol_data, vol_data.astype(np.uint16)):
166-
logging.info("Rescaled pixel array can be safely cast to uint16 with equivalence test.")
167-
vol_data = vol_data.astype(dtype=np.uint16)
168-
# Check if casting to int16 and back to float results in the same values.
169-
elif np.array_equal(vol_data, vol_data.astype(np.int16)):
170-
logging.info("Rescaled pixel array can be safely cast to int16 with equivalence test.")
171-
vol_data = vol_data.astype(dtype=np.int16)
172-
# Check casting to float32 with equivalence test
173-
elif np.array_equal(vol_data, vol_data.astype(np.float32)):
174-
logging.info("Rescaled pixel array can be cast to float32 with equivalence test.")
175-
vol_data = vol_data.astype(np.float32)
176-
else:
177-
logging.info("Rescaled pixel data remains as of type float64.")
175+
logging.error(f"Failed to get pixel array from the first slice: {e}")
176+
raise
177+
178+
# Pre-allocate the volume data array.
179+
vol_data = np.empty(vol_shape, dtype=dtype)
180+
vol_data[0] = first_slice_pixel_array
181+
182+
# Read subsequent slices directly into the pre-allocated array.
183+
for i, s in enumerate(slices[1:], 1):
184+
try:
185+
vol_data[i] = _get_rescaled_pixel_array(s)
186+
except Exception as e:
187+
logging.error(f"Failed to get pixel array from slice {i}: {e}")
188+
raise
178189

179190
return vol_data
180191

0 commit comments

Comments
 (0)