Skip to content

Commit

Permalink
Enabling the use of all 5 GBM postop segmentation models, based on th…
Browse files Browse the repository at this point in the history 
…e provided inputs.
  • Loading branch information
@dbouget
dbouget committed Mar 5, 2024
1 parent fed761a commit 93ae942
Showing 1 changed file with 50 additions and 212 deletions.
262 changes: 50 additions & 212 deletions utils/logic/PipelineCreationHandler.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -64,14 +64,16 @@ def create_pipeline(model_name: str, patient_parameters, task: str) -> dict:
elif task == 'preop_segmentation':
return __create_segmentation_pipeline(model_name, patient_parameters)
elif task == 'postop_segmentation':
download_model(model_name='MRI_GBM_Postop_FV_4p')
model_name = select_appropriate_postop_model(patient_parameters)
download_model(model_name=model_name)
return __create_postop_segmentation_pipeline(model_name, patient_parameters)
elif task == 'other_segmentation':
return __create_other_segmentation_pipeline(model_name, patient_parameters)
elif task == 'preop_reporting':
return __create_preop_reporting_pipeline(model_name, patient_parameters)
elif task == 'postop_reporting':
download_model(model_name='MRI_GBM_Postop_FV_4p')
model_name = select_appropriate_postop_model(patient_parameters)
download_model(model_name=model_name)
return __create_postop_reporting_pipeline(model_name, patient_parameters)
else:
return __create_custom_pipeline(task, model_name, patient_parameters)
Expand Down Expand Up @@ -179,13 +181,8 @@ def __create_postop_segmentation_pipeline(model_name, patient_parameters):
"""
The default postop segmentation model is the one with four inputs, but based on the loaded images another fitting
model could be used.
@TODO. Ideally, in the future, the disambiguation of the best model to use should be done in the backend. In that
case, how to properly retrieve the corresponding pipeline.json?
"""
postop_model_name = "MRI_GBM_Postop_FV_4p"
if UserPreferencesStructure.getInstance().use_manual_sequences:
postop_model_name = select_appropriate_postop_model(patient_parameters)
infile = open(os.path.join(SoftwareConfigResources.getInstance().models_path, postop_model_name, 'pipeline.json'), 'rb')
infile = open(os.path.join(SoftwareConfigResources.getInstance().models_path, model_name, 'pipeline.json'), 'rb')
raw_pip = json.load(infile)

pip = {}
Expand Down Expand Up @@ -317,6 +314,9 @@ def __create_postop_reporting_pipeline(model_name, patient_parameters):
"""
"""
infile = open(os.path.join(SoftwareConfigResources.getInstance().models_path, model_name, 'pipeline.json'), 'rb')
raw_pip = json.load(infile)

pip = {}
pip_num_int = 0
if not UserPreferencesStructure.getInstance().use_manual_sequences:
Expand All @@ -329,206 +329,14 @@ def __create_postop_reporting_pipeline(model_name, patient_parameters):
pip[pip_num]["description"] = "Classification of the MRI sequence type for all input scans"
download_model(model_name='MRI_Sequence_Classifier')

pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
pip[pip_num] = {}
pip[pip_num]["task"] = 'Segmentation'
pip[pip_num]["inputs"] = {}
pip[pip_num]["inputs"]["0"] = {}
pip[pip_num]["inputs"]["0"]["timestamp"] = 0
pip[pip_num]["inputs"]["0"]["sequence"] = "T1-CE"
pip[pip_num]["inputs"]["0"]["labels"] = None
pip[pip_num]["inputs"]["0"]["space"] = {}
pip[pip_num]["inputs"]["0"]["space"]["timestamp"] = 0
pip[pip_num]["inputs"]["0"]["space"]["sequence"] = "T1-CE"
pip[pip_num]["target"] = ["Brain"]
pip[pip_num]["model"] = "MRI_Brain"
pip[pip_num]["description"] = "Brain segmentation in T1CE (T0)"
download_model(model_name='MRI_Brain')

pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
pip[pip_num] = {}
pip[pip_num]["task"] = 'Segmentation'
pip[pip_num]["inputs"] = {}
pip[pip_num]["inputs"]["0"] = {}
pip[pip_num]["inputs"]["0"]["timestamp"] = 0
pip[pip_num]["inputs"]["0"]["sequence"] = "T1-CE"
pip[pip_num]["inputs"]["0"]["labels"] = None
pip[pip_num]["inputs"]["0"]["space"] = {}
pip[pip_num]["inputs"]["0"]["space"]["timestamp"] = 0
pip[pip_num]["inputs"]["0"]["space"]["sequence"] = "T1-CE"
pip[pip_num]["target"] = ["Tumor"]
pip[pip_num]["model"] = model_name
pip[pip_num]["description"] = "Tumor segmentation in T1CE (T0)"
download_model(model_name=model_name)

pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
pip[pip_num] = {}
pip[pip_num]["task"] = 'Segmentation'
pip[pip_num]["inputs"] = {}
pip[pip_num]["inputs"]["0"] = {}
pip[pip_num]["inputs"]["0"]["timestamp"] = 1
pip[pip_num]["inputs"]["0"]["sequence"] = "T1-CE"
pip[pip_num]["inputs"]["0"]["labels"] = None
pip[pip_num]["inputs"]["0"]["space"] = {}
pip[pip_num]["inputs"]["0"]["space"]["timestamp"] = 1
pip[pip_num]["inputs"]["0"]["space"]["sequence"] = "T1-CE"
pip[pip_num]["target"] = ["Brain"]
pip[pip_num]["model"] = "MRI_Brain"
pip[pip_num]["description"] = "Brain segmentation in T1CE (T1)"

pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
pip[pip_num] = {}
pip[pip_num]["task"] = 'Registration'
pip[pip_num]["moving"] = {}
pip[pip_num]["moving"]["timestamp"] = 0
pip[pip_num]["moving"]["sequence"] = "T1-CE"
pip[pip_num]["fixed"] = {}
pip[pip_num]["fixed"]["timestamp"] = 1
pip[pip_num]["fixed"]["sequence"] = "T1-CE"
pip[pip_num]["description"] = "Registration from T1CE (T0) to T1CE (T1)"

pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
pip[pip_num] = {}
pip[pip_num]["task"] = 'Apply registration'
pip[pip_num]["moving"] = {}
pip[pip_num]["moving"]["timestamp"] = 0
pip[pip_num]["moving"]["sequence"] = "T1-CE"
pip[pip_num]["fixed"] = {}
pip[pip_num]["fixed"]["timestamp"] = 1
pip[pip_num]["fixed"]["sequence"] = "T1-CE"
pip[pip_num]["direction"] = "forward"
pip[pip_num]["description"] = "Apply registration from T1CE (T0) to T1CE (T1)"

pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
pip[pip_num] = {}
pip[pip_num]["task"] = 'Segmentation'
pip[pip_num]["inputs"] = {}
pip[pip_num]["inputs"]["0"] = {}
pip[pip_num]["inputs"]["0"]["timestamp"] = 1
pip[pip_num]["inputs"]["0"]["sequence"] = "FLAIR"
pip[pip_num]["inputs"]["0"]["labels"] = None
pip[pip_num]["inputs"]["0"]["space"] = {}
pip[pip_num]["inputs"]["0"]["space"]["timestamp"] = 1
pip[pip_num]["inputs"]["0"]["space"]["sequence"] = "FLAIR"
pip[pip_num]["target"] = ["Brain"]
pip[pip_num]["model"] = "MRI_Brain"
pip[pip_num]["description"] = "Brain segmentation in FLAIR (T1)"

pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
pip[pip_num] = {}
pip[pip_num]["task"] = 'Registration'
pip[pip_num]["moving"] = {}
pip[pip_num]["moving"]["timestamp"] = 1
pip[pip_num]["moving"]["sequence"] = "FLAIR"
pip[pip_num]["fixed"] = {}
pip[pip_num]["fixed"]["timestamp"] = 1
pip[pip_num]["fixed"]["sequence"] = "T1-CE"
pip[pip_num]["description"] = "Registration from FLAIR (T1) to T1CE (T1)"

pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
pip[pip_num] = {}
pip[pip_num]["task"] = 'Apply registration'
pip[pip_num]["moving"] = {}
pip[pip_num]["moving"]["timestamp"] = 1
pip[pip_num]["moving"]["sequence"] = "FLAIR"
pip[pip_num]["fixed"] = {}
pip[pip_num]["fixed"]["timestamp"] = 1
pip[pip_num]["fixed"]["sequence"] = "T1-CE"
pip[pip_num]["direction"] = "forward"
pip[pip_num]["description"] = "Apply registration from FLAIR (T1) to T1CE (T1)"

pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
pip[pip_num] = {}
pip[pip_num]["task"] = 'Segmentation'
pip[pip_num]["inputs"] = {}
pip[pip_num]["inputs"]["0"] = {}
pip[pip_num]["inputs"]["0"]["timestamp"] = 1
pip[pip_num]["inputs"]["0"]["sequence"] = "T1-w"
pip[pip_num]["inputs"]["0"]["labels"] = None
pip[pip_num]["inputs"]["0"]["space"] = {}
pip[pip_num]["inputs"]["0"]["space"]["timestamp"] = 1
pip[pip_num]["inputs"]["0"]["space"]["sequence"] = "T1-w"
pip[pip_num]["target"] = ["Brain"]
pip[pip_num]["model"] = "MRI_Brain"
pip[pip_num]["description"] = "Brain segmentation in T1w (T1)"

pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
pip[pip_num] = {}
pip[pip_num]["task"] = 'Registration'
pip[pip_num]["moving"] = {}
pip[pip_num]["moving"]["timestamp"] = 1
pip[pip_num]["moving"]["sequence"] = "T1-w"
pip[pip_num]["fixed"] = {}
pip[pip_num]["fixed"]["timestamp"] = 1
pip[pip_num]["fixed"]["sequence"] = "T1-CE"
pip[pip_num]["description"] = "Registration from T1w (T1) to T1CE (T1)"

pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
pip[pip_num] = {}
pip[pip_num]["task"] = 'Apply registration'
pip[pip_num]["moving"] = {}
pip[pip_num]["moving"]["timestamp"] = 1
pip[pip_num]["moving"]["sequence"] = "T1-w"
pip[pip_num]["fixed"] = {}
pip[pip_num]["fixed"]["timestamp"] = 1
pip[pip_num]["fixed"]["sequence"] = "T1-CE"
pip[pip_num]["direction"] = "forward"
pip[pip_num]["description"] = "Apply registration from T1w (T1) to T1CE (T1)"

pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
pip[pip_num] = {}
pip[pip_num]["task"] = 'Segmentation'
pip[pip_num]["inputs"] = {}
pip[pip_num]["inputs"]["0"] = {}
pip[pip_num]["inputs"]["0"]["timestamp"] = 1
pip[pip_num]["inputs"]["0"]["sequence"] = "T1-CE"
pip[pip_num]["inputs"]["0"]["labels"] = None
pip[pip_num]["inputs"]["0"]["space"] = {}
pip[pip_num]["inputs"]["0"]["space"]["timestamp"] = 1
pip[pip_num]["inputs"]["0"]["space"]["sequence"] = "T1-CE"
pip[pip_num]["inputs"]["1"] = {}
pip[pip_num]["inputs"]["1"]["timestamp"] = 1
pip[pip_num]["inputs"]["1"]["sequence"] = "T1-w"
pip[pip_num]["inputs"]["1"]["labels"] = None
pip[pip_num]["inputs"]["1"]["space"] = {}
pip[pip_num]["inputs"]["1"]["space"]["timestamp"] = 1
pip[pip_num]["inputs"]["1"]["space"]["sequence"] = "T1-CE"
pip[pip_num]["inputs"]["2"] = {}
pip[pip_num]["inputs"]["2"]["timestamp"] = 1
pip[pip_num]["inputs"]["2"]["sequence"] = "FLAIR"
pip[pip_num]["inputs"]["2"]["labels"] = None
pip[pip_num]["inputs"]["2"]["space"] = {}
pip[pip_num]["inputs"]["2"]["space"]["timestamp"] = 1
pip[pip_num]["inputs"]["2"]["space"]["sequence"] = "T1-CE"
pip[pip_num]["inputs"]["3"] = {}
pip[pip_num]["inputs"]["3"]["timestamp"] = 0
pip[pip_num]["inputs"]["3"]["sequence"] = "T1-CE"
pip[pip_num]["inputs"]["3"]["labels"] = None
pip[pip_num]["inputs"]["3"]["space"] = {}
pip[pip_num]["inputs"]["3"]["space"]["timestamp"] = 1
pip[pip_num]["inputs"]["3"]["space"]["sequence"] = "T1-CE"
pip[pip_num]["inputs"]["4"] = {}
pip[pip_num]["inputs"]["4"]["timestamp"] = 0
pip[pip_num]["inputs"]["4"]["sequence"] = "T1-CE"
pip[pip_num]["inputs"]["4"]["labels"] = "Tumor"
pip[pip_num]["inputs"]["4"]["space"] = {}
pip[pip_num]["inputs"]["4"]["space"]["timestamp"] = 1
pip[pip_num]["inputs"]["4"]["space"]["sequence"] = "T1-CE"
pip[pip_num]["target"] = ["Tumor"]
pip[pip_num]["model"] = "MRI_GBM_Postop_FV_4p"
pip[pip_num]["description"] = "Tumor segmentation in T1CE (T1)"
for steps in list(raw_pip.keys()):
# Excluding brain segmentation step if the inputs are already skull-stripped
if (UserPreferencesStructure.getInstance().use_stripped_inputs and
(raw_pip[steps]["task"] == "Segmentation" and raw_pip[steps]["model"] == "MRI_Brain")):
continue
pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
pip[pip_num] = raw_pip[steps]

pip_num_int = pip_num_int + 1
pip_num = str(pip_num_int)
Expand Down Expand Up @@ -635,9 +443,39 @@ def __create_custom_pipeline(task, tumor_type, patient_parameters):

def select_appropriate_postop_model(patient_parameters) -> str:
"""
Temporary method, which will be deported in the backend, for selecting the best postoperative glioblastoma
segmentation model based on available inputs.
Method for selecting the best postoperative glioblastoma segmentation model based on available inputs.
Should it be deported in the RADS backend?
"""
model_name = "MRI_GBM_Postop_FV_4p"

model_name = "MRI_GBM_Postop_FV_1p"
if not UserPreferencesStructure.getInstance().use_manual_sequences:
# Case where the model selection should then be deported to the backend, or the MRI sequence identification
# should happen before calling a segmentation/reporting pipeline?
return "MRI_GBM_Postop_FV_4p"

exist_preop_t1 = False
exist_postop_t1ce = False
exist_postop_t1w = False
exist_postop_flair = False

for v in list(patient_parameters.mri_volumes.keys()):
volume_object = patient_parameters.mri_volumes[v]
if volume_object.timestamp_uid == "T0":
if volume_object.get_sequence_type_str() == "T1-CE":
exist_preop_t1 = True
elif volume_object.timestamp_uid == "T1":
if volume_object.get_sequence_type_str() == "T1-CE":
exist_postop_t1ce = True
elif volume_object.get_sequence_type_str() == "T1-w":
exist_postop_t1w = True
elif volume_object.get_sequence_type_str() == "FLAIR":
exist_postop_flair = True

if exist_postop_t1ce and exist_postop_t1w:
model_name = "MRI_GBM_Postop_FV_2p"
if exist_postop_t1ce and exist_postop_t1w and exist_postop_flair:
model_name = "MRI_GBM_Postop_FV_3p"
if exist_postop_t1ce and exist_postop_t1w and exist_preop_t1:
model_name = "MRI_GBM_Postop_FV_4p"
if exist_postop_t1ce and exist_postop_t1w and exist_postop_flair and exist_preop_t1:
model_name = "MRI_GBM_Postop_FV_5p"
return model_name

0 comments on commit 93ae942

Please sign in to comment.