WIP: SR from existing outputs (Issue 35)

.gitignore

@@ -5,7 +5,6 @@ __pycache__/
 
 .DS_Store
 outputs3*
-*outputs*
 models/*
 *data/*
 *.nii.gz*

deploy/do_sr_on_existing_outputs.py

@@ -0,0 +1,83 @@
+import os
+threads = "8"
+os.environ["TF_NUM_INTEROP_THREADS"] = threads
+os.environ["TF_NUM_INTRAOP_THREADS"] = threads
+os.environ["ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS"] = threads
+
+import ants
+import antspynet
+import antspyt1w
+import superiq
+import pandas as pd
+import tensorflow as tf
+
+# example image
+img = ants.image_read( "PPMI-41664-20191028-MRI_T1-I1350813-antspyt1w-V0-brain_n4_dnz.nii.gz" )
+# this is the default template for antspyt1w
+template = ants.image_read( antspyt1w.get_data( "T_template0", target_extension="nii.gz") )
+################################################################################
+# DESCRIBE how to derive SR from antspyt1w.hierarchical outputs
+################################################################################
+# basically, two ways: label then SR or SR then label
+# here, we define the approach for doing labeling then SR.
+# first - map to a different template via hemi_reg (this example: CIT168)
+templatebxt = template * antspynet.brain_extraction( template )
+templateLR = ants.image_read( antspyt1w.get_data( "T_template0_LR", target_extension="nii.gz") )
+cit168lab = ants.image_read( antspyt1w.get_data( "det_atlas_25_pad_LR_adni", target_extension="nii.gz") )
+cit168 = ants.image_read( antspyt1w.get_data( "CIT168_T1w_700um_pad_adni", target_extension="nii.gz") )
+cit168desc = pd.read_csv( antspyt1w.get_data( "CIT168_Reinf_Learn_v1_label_descriptions_pad", target_extension="csv") )
+cithemi = antspyt1w.label_hemispheres( cit168, templatebxt, templateLR, reg_iterations=[200, 50, 2, 0])
+# new registration
+imgseg = ants.image_read( "PPMI-41664-20191028-MRI_T1-I1350813-antspyt1w-V0-dkt_parc_tissue_segmentation.nii.gz" )
+imghemi = ants.image_read( "PPMI-41664-20191028-MRI_T1-I1350813-antspyt1w-V0-dkt_parc_hemisphere_labels.nii.gz")
+istest = False
+citreg = antspyt1w.hemi_reg( img, imgseg, imghemi, cit168, cithemi,
+    output_prefix="PPMI-41664-20191028-MRI_T1-I1350813-CITREG", padding=10,
+    labels_to_register=[2, 3, 4, 5], is_test=istest )
+# apply transformations to this new template's label set (CIT168 SNc etc)
+cit2ppmiL = ants.apply_transforms( img, cit168lab, citreg['synL']['invtransforms'],
+          interpolator='genericLabel' ) * ants.threshold_image( imghemi, 1, 1  )
+cit2ppmiR = ants.apply_transforms( img, cit168lab, citreg['synR']['invtransforms'],
+          interpolator='genericLabel' ) * ants.threshold_image( imghemi, 2, 2  )
+citdfL = antspyt1w.map_segmentation_to_dataframe( "CIT168_Reinf_Learn_v1_label_descriptions_pad", cit2ppmiL )
+citdfR = antspyt1w.map_segmentation_to_dataframe( "CIT168_Reinf_Learn_v1_label_descriptions_pad", cit2ppmiR )
+# now do SR+seg
+mysupmdl = tf.keras.models.load_model( "SEGSR_32_ANINN222_3.h5" )
+sup = superiq.super_resolution_segmentation_per_label( img, cit2ppmiL,
+    upFactor=[2,2,2],
+    sr_model=mysupmdl,
+    segmentation_numbers=[7,8,9,10], # should do all on left side
+    dilation_amount=6,
+    probability_images=None,
+    probability_labels=None, max_lab_plus_one=True, verbose=True)
+ants.image_write( sup['super_resolution'], '/tmp/tempi.nii.gz' )
+ants.image_write( sup['super_resolution_segmentation'], '/tmp/temps.nii.gz' )
+# output style should be the same as usual : map to dataframe / pivot, etc.
+################################################################################
+################################################################################
+# next, we define the approach for SR then labeling
+################################################################################
+# do super resolution on whole brain
+brainmask = ants.threshold_image( ants.iMath(img,"Normalize"), 1.e-9, 1.0 )
+hemisr = superiq.super_resolution_segmentation_per_label( img, imghemi * brainmask,
+    upFactor=[2,2,2],
+    sr_model=mysupmdl,
+    segmentation_numbers=[1,2] )
+imgsr = hemisr[ 'super_resolution' ]
+imghemisr = hemisr['super_resolution_segmentation']
+imgsegsr = ants.resample_image_to_target( imgseg, imgsr, interp_type='nearestNeighbor')
+# do high-res registration and then proceed just as above ...
+citregsr = antspyt1w.hemi_reg( imgsr, imgsegsr, imghemisr,
+    cit168, cithemi,
+    output_prefix="PPMI-41664-20191028-MRI_T1-I1350813-CITREGSR",
+    padding=10,
+    labels_to_register=[2, 3, 4, 5], is_test=istest )
+# apply transformations to this new template's label set (CIT168 SNc etc)
+cit2ppmiLSR = ants.apply_transforms( imgsr, cit168lab, citregsr['synL']['invtransforms'],
+          interpolator='genericLabel' ) * ants.threshold_image( imghemisr, 1, 1  )
+cit2ppmiRSR = ants.apply_transforms( imgsr, cit168lab, citregsr['synR']['invtransforms'],
+          interpolator='genericLabel' ) * ants.threshold_image( imghemisr, 2, 2  )
+citdfLSR = antspyt1w.map_segmentation_to_dataframe( "CIT168_Reinf_Learn_v1_label_descriptions_pad", cit2ppmiLSR )
+citdfRSR = antspyt1w.map_segmentation_to_dataframe( "CIT168_Reinf_Learn_v1_label_descriptions_pad", cit2ppmiRSR )
+
+