Merge pull request #111 from SyneRBI/GE_DMI3_Torso

GE_DMI3_Torso dataset

SIRF_data_preparation/GE_DMI3_Torso/README.md

@@ -0,0 +1,25 @@
+# Torso phantom scanned on a GE Discovery MI (3 ring)
+
+This data is from a scan performed by GE of the
+Data Spectrum [Antropomorphic Torso Phantom](https://www.spect.com/pdf/anthropomorphic-torso-phantom.pdf)
+with additional lesion inserts of diameter 10mm.
+
+Activity ratio between "liver" and "torso" is approximately 2. Lungs
+are empty (air). Contrast of the lesions is unfortunately unknown.
+,v 
+"Corrections" were obtained with GE Duetto and converted to STIR Interfile
+using internal scripts.
+
+Due to data size issues and count levels, the supplied data is **reduced** from the original scan
+by keeping only the "direct" sinograms (segment 0, corresponding to min/max ring differences -1/+1), e.g. on the command line
+```
+SSRB prompts.hs prompts_f1b1.hs 1 1 0 0 1
+```
+
+In addition, as current PETRIC examples do not cope with non-TOF multfactors for TOF data, the
+multfactors were expanded to TOF (by duplication).
+
+## VOIs
+
+VOIs were determined by Kris Thielemans on reconstructed images and approximately placed
+where some of the lesions are visible. They are not necessarily at the correct location.

SIRF_data_preparation/GE_DMI3_Torso/VOI_prep.py

@@ -0,0 +1,85 @@
+#%% file to prepare VOIs for the Torso phantom, using coordinates derived manually
+
+import matplotlib.pyplot as plt
+import numpy as np
+from pathlib import Path
+import os
+from scipy.ndimage import binary_fill_holes
+import sirf.STIR as STIR
+import SIRF_data_preparation.data_QC as data_QC
+from SIRF_data_preparation.data_utilities import the_data_path, the_orgdata_path
+from SIRF_data_preparation.dataset_settings import get_settings
+#%%
+scanID = 'GE_DMI3_Torso'
+data_path = Path(the_data_path(scanID))
+output_path = Path(the_data_path(scanID, 'PETRIC'))
+os.makedirs(output_path, exist_ok=True)
+#%%
+image_filename = str(data_path / 'OSEM_image.hv')
+image = STIR.ImageData(image_filename)
+# %% find whole phantom VOI by thresholding/filling
+threshold = image.max() * .05
+im_arr = image.as_array() >= threshold
+# we want to use binary_fill_holes to fill the spine and lung. However,
+# that function does not fill holes connected to the boundary, so let's
+# fill the boundary planes first
+im_arr[0, :, :] = True
+im_arr[-1, :, :] = True
+im_arr = binary_fill_holes(im_arr)
+# now exclude boundary
+im_arr[0, :, :] = False
+im_arr[-1, :, :] = False
+VOI = image.allocate(0)
+VOI.fill(im_arr.astype(np.float32))
+VOI.show()
+VOI.write(str(output_path / 'VOI_whole_object.hv'))
+#%% find centre image as KT got the coordinates from Amide
+# note that geo.get_index_to_physical_point_matrix() is still unstable, so we won't use it.
+# Warning: the following will likely fail for future versions of SIRF
+zcentre = (image.shape[0] - 1)/2 * image.voxel_sizes()[0]
+# %% background VOI (in approx middle of torso)
+shape = STIR.EllipticCylinder()
+shape.set_length(23)
+shape.set_radii((120, 120))
+shape.set_origin((zcentre + 0, 0, 0))
+VOI.fill(0)
+VOI.add_shape(shape, scale = 1)
+VOI.write(str(output_path / 'VOI_background.hv'))
+# %% spine (approximately top half of it)
+shape = STIR.EllipticCylinder()
+shape.set_length(100)
+shape.set_radii((22, 22))
+shape.set_origin((zcentre + 42.6, 86.9, 3))
+VOI.fill(0)
+VOI.add_shape(shape, scale = 1)
+VOI.write(str(output_path / 'VOI_spine.hv'))
+VOI.show()
+# %% lung lesion
+shape = STIR.Ellipsoid()
+shape.set_radius_z(5)
+shape.set_radius_y(5)
+shape.set_radius_x(5)
+shape.set_origin((zcentre + 25, -42.2, -105.3))
+VOI.fill(0)
+VOI.add_shape(shape, scale = 1)
+VOI.write(str(output_path / 'VOI_lung_lesion.hv'))
+# %%
+# %% liver lesion
+shape = STIR.Ellipsoid()
+shape.set_radius_z(5)
+shape.set_radius_y(5)
+shape.set_radius_x(5)
+shape.set_origin((zcentre - 38.2, -38.9, -49.7))
+VOI.fill(0)
+VOI.add_shape(shape, scale = 1)
+VOI.write(str(output_path / 'VOI_liver_lesion.hv'))
+# %% lung lesion
+shape = STIR.Ellipsoid()
+shape.set_radius_z(5)
+shape.set_radius_y(5)
+shape.set_radius_x(5)
+shape.set_origin((zcentre + 47.6, 12.5, 151.5))
+VOI.fill(0)
+VOI.add_shape(shape, scale = 1)
+VOI.write(str(output_path / 'VOI_chest_lesion.hv'))
+

SIRF_data_preparation/dataset_settings.py

@@ -5,7 +5,7 @@
 
 DATA_SUBSETS = {
     'Siemens_mMR_NEMA_IQ': 7, 'Siemens_mMR_NEMA_IQ_lowcounts': 7, 'Siemens_mMR_ACR': 7, 'NeuroLF_Hoffman_Dataset': 16,
-    'Mediso_NEMA_IQ': 12, 'Siemens_Vision600_thorax': 5}
+    'Mediso_NEMA_IQ': 12, 'Siemens_Vision600_thorax': 5, 'GE_DMI3_Torso': 8}
 
 
 @dataclass

petric.py

@@ -252,7 +252,8 @@ def get_image(fname):
     'Siemens_mMR_ACR': {'transverse_slice': 99},
     'NeuroLF_Hoffman_Dataset': {'transverse_slice': 72},
     'Mediso_NEMA_IQ': {'transverse_slice': 22, 'coronal_slice': 89, 'sagittal_slice': 66},
-    'Siemens_Vision600_thorax': {}}
+    'Siemens_Vision600_thorax': {},
+    'GE_DMI3_Torso': {}}
 
 if SRCDIR.is_dir():
     # create list of existing data