3 dW/@sddZddlmZmZmZmZmZmZmZm Z m Z m Z m Z ddl Z GdddeZGdddeZGd d d eZGd d d eZGd ddeZGdddeZGdddeZGdddeZGdddeZGdddeZGdddeZGdddeZGdddeZGdd d eZGd!d"d"eZGd#d$d$eZGd%d&d&eZGd'd(d(eZdS))at The fix module provides classes for interfacing with the `FSL FIX ` command line tools. This was written to work with FSL version v5.0 The following example assumes that melodic has already been run, so the datagrabber is configured to start from there (a list of melodic output directories). If no hand_labels_noise.txt exists already, this will fail and comment on that. EXAMPLE: subject_list = ['1', '2', '3'] fix_pipeline = pe.Workflow(name='fix_pipeline') fix_pipeline.base_dir = os.path.abspath('./') info = dict(mel_ica=[['subject_id']]) datasource = pe.Node(interface=nio.DataGrabber(infields=['subject_id'], outfields=['mel_ica']), name='datasource') datasource.inputs.base_directory = os.path.abspath('') datasource.inputs.template = '%s/' datasource.inputs.template_args = info datasource.inputs.subject_id = subject_list datasource.inputs.sort_filelist = True datasource.iterables = ('subject_id', subject_list) # create training set by looking into which mel_icas have hand_labels_noise.txt files in them create_training_set = pe.JoinNode(interface=fix.TrainingSetCreator(), joinfield=['mel_icas_in'], joinsource='datasource', name='trainingset_creator') # train the classifier train_node = pe.Node(interface=fix.Training(trained_wts_filestem='foo'), name='train_node') # test accuracy. Probably not necessary, and also failing on my setup because of fix itself (no error msg) accuracy_tester = pe.Node(interface=fix.AccuracyTester(output_directory='more_foo'), name='accuracy_tester') # classify components classify_node = pe.Node(interface=fix.Classifier(), name='classify') # apply cleanup cleaner_node = pe.Node(interface=fix.Cleaner(), name='cleaner') fix_pipeline.connect(datasource, 'mel_ica', create_training_set, 'mel_icas_in') fix_pipeline.connect(create_training_set, 'mel_icas_out', train_node, 'mel_icas') fix_pipeline.connect(train_node, 'trained_wts_file', accuracy_tester, 'trained_wts_file') fix_pipeline.connect(datasource, 'mel_ica', accuracy_tester, 'mel_icas') fix_pipeline.connect(train_node, 'trained_wts_file', classify_node, 'trained_wts_file') fix_pipeline.connect(datasource, 'mel_ica', classify_node, 'mel_ica') fix_pipeline.connect(classify_node, 'artifacts_list_file', cleaner_node, 'artifacts_list_file') fix_pipeline.write_graph() outgraph = fix_pipeline.run() ) TraitedSpecCommandLineInputSpec CommandLineInputMultiPathOutputMultiPath BaseInterfaceBaseInterfaceInputSpectraits DirectoryFile isdefinedNc@s$eZdZeedddddd dZdS) TrainingSetCreatorInputSpecT)existsFzMelodic output directoriesz%s)copyfiledescargstrpositionN)__name__ __module__ __qualname__rr mel_icas_inrr;/tmp/pip-build-7vycvbft/nipype/nipype/interfaces/fsl/fix.pyrJs rc@s$eZdZeedddddd dZdS) TrainingSetCreatorOutputSpecT)rFzHand labels for noise vs signalz%sr)rrrrNr)rrrrr mel_icas_outrrrrrTs rc@s,eZdZdZeZeZdZddZ ddZ dS)TrainingSetCreatoraGoes through set of provided melodic output directories, to find all the ones that have a hand_labels_noise.txt file in them. This is outsourced as a separate class, so that the pipeline is rerun everytime a handlabeled file has been changed, or a new one created. TcCsZg}x0|jjD]$}tjjtjj|dr|j|qWt|dkrVtd|j j |f|S)Nzhand_labels_noise.txtr zP%s did not find any hand_labels_noise.txt files in the following directories: %s) inputsrospathrjoinappendlen Exception __class__r)selfZruntimemel_icasitemrrr_run_interfacels z!TrainingSetCreator._run_interfacecCsNg}x0|jjD]$}tjjtjj|dr|j|qW|jj}||d<|S)Nzhand_labels_noise.txtr) rrr r!rr"r#Z_outputsget)r'r(r)outputsrrr _list_outputszs z TrainingSetCreator._list_outputsN) rrr__doc__r input_specr output_specZ _always_runr*r-rrrrr^s rc@seZdZeddddddZdS) FeatureExtractorInputSpecTFz'Melodic output directory or directoriesz%sr)rrrrrNr)rrrr mel_icarrrrr1s r1c@seZdZeddddddZdS) FeatureExtractorOutputSpecTFz'Melodic output directory or directoriesz%sr)rrrrrNr)rrrr r2rrrrr3s r3c@s$eZdZdZeZeZdZddZ dS)FeatureExtractorzB Extract features (for later training and/or classifying) zfix -fcCs|jj}|jj|d<|S)Nr2)r0r+rr2)r'r,rrrr-s  zFeatureExtractor._list_outputsN) rrrr.r1r/r3r0cmdr-rrrrr4s r4c@sDeZdZeedddddddZejdddd Zej d d d d Z dS)TrainingInputSpecT)rFzMelodic output directoriesz%sr)rrrrzJtrained-weights filestem, used for trained_wts_file and output directories)rrrz-lz0full leave-one-out test with classifier trainingr)rrrNr) rrrrr r(r Strtrained_wts_filestemBoolZloorrrrr6sr6c@seZdZedddZdS)TrainingOutputSpecTzTrained-weights file)rrN)rrrr trained_wts_filerrrrr:sr:c@s$eZdZdZeZeZdZddZ dS)TrainingzO Train the classifier based on your own FEAT/MELODIC output directory. zfix -tcCsF|jj}t|jjr2tjj|jjd|d<ntjjd|d<|S)Nz.RDatar;ztrained_wts_file.RData)r0r+r rr8r r!abspath)r'r,rrrr-s   zTraining._list_outputsN) rrrr.r6r/r:r0r5r-rrrrr<s r<c@sFeZdZeeddddddddZeddd dd Zed dd dd Zd S)AccuracyTesterInputSpecT)rFzMelodic output directoriesz%s)rrrr mandatoryztrained-weights filer)rrrr@zBPath to folder in which to store the results of the accuracy test.rN) rrrrr r(r r;output_directoryrrrrr>sr>c@seZdZeddddZdS)AccuracyTesterOutputSpeczBPath to folder in which to store the results of the accuracy test.z%sr)rrrN)rrrr rArrrrrBsrBc@s$eZdZdZeZeZdZddZ dS)AccuracyTesterz Test the accuracy of an existing training dataset on a set of hand-labelled subjects. Note: This may or may not be working. Couldn't presently not confirm because fix fails on this (even outside of nipype) without leaving an error msg. zfix -CcCsB|jj}t|jjr.td|jjd|d<ntddd|d<|S)NF)rvaluerAZ accuracy_test)r0r+r rrAr )r'r,rrrr-s   zAccuracyTester._list_outputsN) rrrr.r>r/rBr0r5r-rrrrrCs rCc@sNeZdZeddddddZeddddddd Zejd d ddd Z ed dZ dS)ClassifierInputSpecTFz'Melodic output directory or directoriesz%sr)rrrrrztrained-weights filer)rrrrr@rz%dzThreshold for cleanup.)rrrr@zkText file listing which ICs are artifacts; can be the output from classification or can be created manually)rNr) rrrr r2r r;r ZIntthreshartifacts_list_filerrrrrEs"rEc@seZdZeddZdS)ClassifierOutputSpeczkText file listing which ICs are artifacts; can be the output from classification or can be created manually)rN)rrrr rGrrrrrH!srHc@s,eZdZdZeZeZdZddZ ddZ dS) Classifierzu Classify ICA components using a specific training dataset ( is in the range 0-100, typically 5-20). zfix -ccCsLtjj|jj\}}|jdd}d|d}tjj||t|d}|S)N.r Z fix4melview_Z_thrz.txt)r r!splitrr;r"str)r'r2rF_r;r8Zfilestemfnamerrr_gen_artifacts_list_file0s  z#Classifier._gen_artifacts_list_filecCs(|jj}|j|jj|jj|d<|S)NrG)r0r+rOrr2rF)r'r,rrrr-9s zClassifier._list_outputsN) rrrr.rEr/rHr0r5rOr-rrrrrI's  rIc@s|eZdZeddddddZejdddd Zejd d dd dd Z ejdddd Z edddd Z edddd Z edddd Z dS)CleanerInputSpecTz%srzkText file listing which ICs are artifacts; can be the output from classification or can be created manually)rrrr@rz-mzJcleanup motion confounds, looks for design.fsf for highpass filter cut-offr)rrrdz-m -h %fzcleanup motion confounds)rZ usedefaultrrz-AzkApply aggressive (full variance) cleanup, instead of the default less-aggressive (unique variance) cleanup.r?z-x %sz!Include additional confound file.N)rrrr rGr r9Zcleanup_motionZFloatZhighpassZ aggressiveZ confound_fileZconfound_file_1Zconfound_file_2rrrrrPBs4  rPc@seZdZedddZdS)CleanerOutputSpecTzCleaned session data)rrN)rrrr cleaned_functional_filerrrrrUlsrUc@s,eZdZdZeZeZdZddZ ddZ dS)CleanerzB Extract features (for later training and/or classifying) zfix -acCs>t|d}|jjd\}}tjj|\}}tjj||dS)zEextract the proper filename from the first line of the artifacts filerrJz _clean.nii.gz)openreadlinerKr r!r")r'Zartifacts_list_filenamerGZfunctional_filename extensionZartifacts_list_file_pathrrr _get_cleaned_functional_filenameys   z(Cleaner._get_cleaned_functional_filenamecCs"|jj}|j|jj|d<|S)NrV)r0r+r\rrG)r'r,rrrr-s zCleaner._list_outputsN) rrrr.rPr/rUr0r5r\r-rrrrrWps  rW) r.baserrrrrrrr r r r r rrrr1r3r4r6r:r<r>rBrCrErHrIrPrUrWrrrr8s(4   &  *