diff --git a/clabtoolkit/dicomtools.py b/clabtoolkit/dicomtools.py index 888cfde..bff11c9 100755 --- a/clabtoolkit/dicomtools.py +++ b/clabtoolkit/dicomtools.py @@ -337,12 +337,16 @@ def _create_session_series_names(dataset): return ses_id, ser_id -def _uncompress_dicom_session(dic_dir: str, subj_ids=None): +def _uncompress_dicom_session(dic_dir: str, + boolrmtar: bool = False, + subj_ids=None): """ Uncompress session folders @params: dic_dir - Required : Directory containing the subjects. It assumes an organization in: //(Str) + boolrmtar - Optional : Boolean variable to remove the tar files after uncompressing the session. Default is False. + subj_ids - Optional : List of subject IDs to be considered. If not provided, it will consider all the subjects in the directory. """ if subj_ids is None: @@ -362,48 +366,43 @@ def _uncompress_dicom_session(dic_dir: str, subj_ids=None): elif not isinstance(subj_ids, list): raise ValueError("The subj_ids parameter must be a list or a string") + n_subj = len(subj_ids) # Failed sessions fail_sess = [] - - # Loop around all the subjects - nsubj = len(subj_ids) - for i, subj_id in enumerate(subj_ids): # Loop along the IDs - subj_dir = os.path.join(dic_dir, subj_id) - - cltmisc._printprogressbar( - i + 1, - nsubj, - "Processing subject " - + subj_id - + ": " - + "(" - + str(i + 1) - + "/" - + str(nsubj) - + ")", - ) - - # Loop along all the sessions inside the subject directory - for ses_tar in glob( - subj_dir + os.path.sep + "*.tar.gz" - ): # Loop along the session - # print('SubjectId: ' + subjId + ' ======> Session: ' + sesId) - # Compress only if it is a folder - if os.path.isfile(ses_tar): - try: - # Compressing the folder - subprocess.run( - ["tar", "xzf", ses_tar, "-C", subj_dir], - stdout=subprocess.PIPE, - universal_newlines=True, - ) - - # Removing the uncompressed dicom folder - # subprocess.run( - # ['rm', '-r', ses_tar], stdout=subprocess.PIPE, universal_newlines=True) - - except: - fail_sess.append(ses_tar) + with Progress() as pb: + t1 = pb.add_task('[green]Compressing subjects...', total=n_subj) + + + # Loop around all the subjects + n_subj = len(subj_ids) + for i, subj_id in enumerate(subj_ids): # Loop along the IDs + subj_dir = os.path.join(dic_dir, subj_id) + pb.update(task_id=t1, description= f'[green]Uncompressing sessions for {subj_id} ({i+1}/{n_subj})', completed=i+1) + + # Loop along all the sessions inside the subject directory + for ses_tar in glob( + subj_dir + os.path.sep + "*.tar.gz" + ): # Loop along the session + # print('SubjectId: ' + subjId + ' ======> Session: ' + sesId) + # Compress only if it is a folder + if os.path.isfile(ses_tar): + try: + # Compressing the folder + subprocess.run( + ["tar", "xzf", ses_tar, "-C", subj_dir], + stdout=subprocess.PIPE, + universal_newlines=True, + ) + + # Removing the uncompressed dicom folder + if boolrmtar: + subprocess.run( + ['rm', '-r', ses_tar], stdout=subprocess.PIPE, universal_newlines=True) + except: + fail_sess.append(ses_tar) + + pb.update(task_id=t1, description= f'[green]Compressing sessions for {subj_id} ({n_subj}/{n_subj})', completed=n_subj) + if fail_sess: print("THE PROCESS FAILED TO UNCOMPRESS THE FOLLOWING TAR FILES:") for i in fail_sess: @@ -437,51 +436,53 @@ def _compress_dicom_session(dic_dir: str, subj_ids=None): elif not isinstance(subj_ids, list): raise ValueError("The subj_ids parameter must be a list or a string") + n_subj = len(subj_ids) # Failed sessions fail_sess = [] - - # Loop around all the subjects - nsubj = len(subj_ids) - for i, subj_id in enumerate(subj_ids): # Loop along the IDs - subj_dir = os.path.join(dic_dir, subj_id) - - cltmisc._printprogressbar( - i + 1, - nsubj, - "Processing subject " - + subj_id - + ": " - + "(" - + str(i + 1) - + "/" - + str(nsubj) - + ")", - ) - - # Loop along all the sessions inside the subject directory - for ses_id in os.listdir(subj_dir): # Loop along the session - ses_dir = os.path.join(subj_dir, ses_id) - # print('SubjectId: ' + subjId + ' ======> Session: ' + sesId) - # Compress only if it is a folder - if os.path.isdir(ses_dir): - tar_filename = ses_dir + ".tar.gz" - try: - # Compressing the folder - subprocess.run( - ["tar", "-C", subj_dir, "-czvf", tar_filename, ses_id], - stdout=subprocess.PIPE, - universal_newlines=True, - ) - - # Removing the uncompressed dicom folder - subprocess.run( - ["rm", "-r", ses_dir], - stdout=subprocess.PIPE, - universal_newlines=True, - ) - - except: - fail_sess.append(ses_dir) + with Progress() as pb: + t1 = pb.add_task('[green]Compressing subjects...', total=n_subj) + + # Loop around all the subjects + nsubj = len(subj_ids) + for i, subj_id in enumerate(subj_ids): # Loop along the IDs + subj_dir = os.path.join(dic_dir, subj_id) + pb.update(task_id=t1, description= f'[green]Compressing sessions for {subj_id} ({i+1}/{n_subj})', completed=i+1) + + # Loop along all the sessions inside the subject directory + ses_dirs = os.listdir(subj_dir) + + # Detect which of the folders are sessions + ses_dirs = [x for x in ses_dirs if os.path.isdir(os.path.join(subj_dir, x))] + + # Detect which of the folders start with 'ses-' + ses_dirs = [x for x in ses_dirs if x.startswith('ses-')] + n_sessions = len(ses_dirs) + + for n_ses, ses_id in enumerate(ses_dirs): # Loop along the session + ses_dir = os.path.join(subj_dir, ses_id) + # print('SubjectId: ' + subjId + ' ======> Session: ' + sesId) + # Compress only if it is a folder + if os.path.isdir(ses_dir): + tar_filename = ses_dir + ".tar.gz" + try: + # Compressing the folder + subprocess.run( + ["tar", "-C", subj_dir, "-czvf", tar_filename, ses_id], + stdout=subprocess.PIPE, + universal_newlines=True, + ) + + # Removing the uncompressed dicom folder + subprocess.run( + ["rm", "-r", ses_dir], + stdout=subprocess.PIPE, + universal_newlines=True, + ) + except: + fail_sess.append(ses_dir) + + pb.update(task_id=t1, description= f'[green]Compressing sessions for {subj_id} ({n_subj}/{n_subj})', completed=n_subj) + if fail_sess: print("THE PROCESS FAILED TO COMPRESS THE FOLLOWING SESSIONS:") for i in fail_sess: