def random_svm_cope_searchlight(root_dir, subj):
"""SVM based searchlight analysis."""
# dir config
work_dir = os.path.join(root_dir, 'workshop', 'searchlight')
# read mask file
print 'Load mask data ...'
mask_file = os.path.join(work_dir, 'mask', 'func_mask.nii.gz')
mask_data = nib.load(mask_file).get_data()
mask_data = mask_data > 0
# load nii data list
print 'Load nii files ...'
cope_list = get_subj_cope_list(root_dir, subj)
# get trial sequence info
print 'Load trial sequence info ...'
tag_list = get_subj_cope_tag(root_dir, subj)
for i in range(100):
# svm results var
clf_results = np.zeros((91, 109, 91, 4))
# for loop for voxel-wise searchlight
mask_coord = niroi.get_roi_coord(mask_data)
ccount = 0
for c in mask_coord:
ccount += 1
print ccount
cube_roi = np.zeros((91, 109, 91))
cube_roi = niroi.cube_roi(cube_roi, c[0], c[1], c[2], 2, 1)
cube_coord = niroi.get_roi_coord(cube_roi)
[train_x, train_y, test_x,
test_y] = get_roi_cope_mvps(cope_list, tag_list, cube_coord)
clf = svm.SVC(kernel='sigmoid')
train_y = np.random.permutation(train_y)
clf.fit(train_x, train_y)
pred = clf.predict(test_x)
for e in range(4):
acc = np.sum(pred[test_y == (e + 1)] ==
(e + 1)) * 1.0 / np.sum(test_y == (e + 1))
print acc
clf_results[c[0], c[1], c[2], e] = acc
# save to nifti
fsl_dir = os.getenv('FSL_DIR')
template_file = os.path.join(fsl_dir, 'data', 'standard',
'MNI152_T1_2mm_brain.nii.gz')
aff = nib.load(template_file).affine
nibase.save2nifti(
clf_results, aff,
os.path.join(
work_dir,
'random_' + subj + '_svm_acc_cope_%s.nii.gz' % (i + 1)))
def get_mvp_group_roi(root_dir):
"""Get multivoxel activity pattern for each srimulus from each ROI."""
# directory config
nii_dir = os.path.join(root_dir, 'nii')
ppi_dir = os.path.join(root_dir, 'ppi')
# load rois
#mask_data = nib.load(os.path.join(ppi_dir, 'cube_rois.nii.gz')).get_data()
mask_data = nib.load(
os.path.join(root_dir, 'group-level', 'rois', 'neurosynth',
'cube_rois.nii.gz')).get_data()
roi_num = int(mask_data.max())
# get scan info from scanlist
scanlist_file = os.path.join(root_dir, 'doc', 'scanlist.csv')
[scan_info, subj_list] = pyunpack.readscanlist(scanlist_file)
for subj in subj_list:
# get run infor for emo task
sid = subj.sess_ID
subj_dir = os.path.join(nii_dir, sid, 'emo')
# get run index
if not 'emo' in subj.run_info:
continue
[run_idx, par_idx] = subj.getruninfo('emo')
# var for MVP
mvp_dict = {}
for r in range(roi_num):
mvp_dict['roi_%s' % (r + 1)] = []
for i in range(10):
if str(i + 1) in par_idx:
print 'Run %s' % (i + 1)
# load cope data
ipar = par_idx.index(str(i + 1))
run_dir = os.path.join(subj_dir, '00' + run_idx[ipar])
print run_dir
trn_file = os.path.join(run_dir, 'train_merged_cope.nii.gz')
test_file = os.path.join(run_dir, 'test_merged_cope.nii.gz')
trn_cope = nib.load(trn_file).get_data()
test_cope = nib.load(test_file).get_data()
run_cope = np.concatenate((trn_cope, test_cope), axis=3)
# XXX: remove mean cope from each trial
mean_cope = np.mean(run_cope, axis=3, keepdims=True)
run_cope = run_cope - mean_cope
# get MVP for each ROI
for r in range(roi_num):
roi_mask = mask_data.copy()
roi_mask[roi_mask != (r + 1)] = 0
roi_mask[roi_mask == (r + 1)] = 1
roi_coord = niroi.get_roi_coord(roi_mask)
for j in range(run_cope.shape[3]):
vtr = niroi.get_voxel_value(roi_coord, run_cope[...,
j])
mvp_dict['roi_%s' % (r + 1)].append(vtr.tolist())
for roi in mvp_dict:
mvp_dict[roi] = np.array(mvp_dict[roi])
outfile = r'%s_roi_mvp.mat' % (sid)
sio.savemat(outfile, mvp_dict)
def get_vxl_trial_rsp(root_dir):
"""Get multivoxel activity pattern for each srimulus
from whole brain mask.
"""
# directory config
nii_dir = os.path.join(root_dir, 'prepro')
rsp_dir = os.path.join(root_dir, 'workshop', 'trial_rsp', 'whole_brain')
# load rois
mask_data = nib.load(
os.path.join(root_dir, 'group-level', 'rois', 'neurosynth',
'cube_rois_r2.nii.gz')).get_data()
mask_data = mask_data > 0
# get scan info from scanlist
scanlist_file = os.path.join(root_dir, 'doc', 'scanlist.csv')
[scan_info, subj_list] = pyunpack.readscanlist(scanlist_file)
for subj in subj_list:
# get run infor for emo task
sid = subj.sess_ID
print sid
subj_dir = os.path.join(nii_dir, sid)
# get run index
if not 'emo' in subj.run_info:
continue
[run_idx, par_idx] = subj.getruninfo('emo')
# var for MVP
for i in range(10):
if not str(i + 1) in par_idx:
continue
print 'Run %s' % (i + 1)
mvp_data = []
# load cope data
ipar = par_idx.index(str(i + 1))
run_dir = os.path.join(subj_dir, '00' + run_idx[ipar])
print run_dir
rsp_file = os.path.join(run_dir, 'mni_sfunc_data_mcf_hp.nii.gz')
rsp = nib.load(rsp_file).get_data()
# derive trial-wise response
trsp = np.zeros((91, 109, 91, 88))
for t in range(88):
trsp[..., t] = (rsp[..., 4 * t + 5] + rsp[..., 4 * t + 6]) / 2
# get MVP of mask
vxl_coord = niroi.get_roi_coord(mask_data)
for j in range(trsp.shape[3]):
vtr = niroi.get_voxel_value(vxl_coord, trsp[..., j])
mvp_data.append(vtr.tolist())
outfile = os.path.join(rsp_dir,
'%s_r%s_mvp.npy' % (sid[:2], i + 1))
np.save(outfile, np.array(mvp_data))
def get_mvp_group_roi(scanlist_file, mask_file):
"""Get multivoxel activity pattern for each srimulus from each ROI."""
# get scan info from scanlist
[scan_info, subj_list] = pyunpack.readscanlist(scanlist_file)
# directory config
nii_dir = scan_info['sessdir']
# load rois
mask_data = nib.load(mask_file).get_data()
roi_dict = {'rOFA': 1, 'lOFA': 2, 'rFFA': 3, 'lFFA': 4}
#output_file = os.path.join(roi_dir, 'neo_group_roi_mvpa.csv')
#f = open(output_file, 'wb')
#f.write('SID,rOFA,lOFA,rFFA,lFFA,rpcSTS,lpcSTS\n')
for subj in subj_list:
# get run infor for emo task
sid = subj.sess_ID
subj_dir = os.path.join(nii_dir, sid, 'emo')
# get par index for each emo run
if not 'emo' in subj.run_info:
continue
[run_idx, par_idx] = subj.getruninfo('emo')
# var for MVP
mvp_dict = {}
for roi in roi_dict:
mvp_dict[roi] = []
for i in range(len(run_idx)):
run_dir = os.path.join(subj_dir, '00'+run_idx[i])
trn_file = os.path.join(run_dir, 'train_merged_cope.nii.gz')
test_file = os.path.join(run_dir, 'test_merged_cope.nii.gz')
trn_cope = nib.load(trn_file).get_data()
test_cope = nib.load(test_file).get_data()
run_cope = np.concatenate((trn_cope, test_cope), axis=3)
# XXX: remove mean cope from each trial
#mean_cope = np.mean(run_cope, axis=3, keepdims=True)
#run_cope = run_cope - mean_cope
# get MVP for each ROI
for roi in roi_dict:
roi_mask = mask_data.copy()
roi_mask[roi_mask!=roi_dict[roi]] = 0
roi_mask[roi_mask==roi_dict[roi]] = 1
roi_coord = niroi.get_roi_coord(roi_mask)
for j in range(run_cope.shape[3]):
trl_vtr = niroi.get_voxel_value(roi_coord, run_cope[..., j])
f.write(','.join(temp)+'\n')
print 'cost %s s'%(time.time() - start_time)
def get_roi_bold_ts(root_dir, subj, roi):
"""Get one ROI's BOLD time course."""
# load nii data list
print 'load nii files ...'
nii_list = get_subj_nii_list(root_dir, subj)
# output data shape: #time x #voxel
bold_ts = np.zeros((3550, roi.sum()))
c = 0
roi_coord = niroi.get_roi_coord(roi)
for i in range(10):
nii = nii_list[i]
for j in range(nii.shape[3]):
vtr = niroi.get_voxel_value(roi_coord, nii[..., j])
bold_ts[c] = vtr
c += 1
return bold_ts
def get_roi_mvp_data(root_dir, subj, roi_file):
"""Get MVP data from specific ROI."""
# dir config
work_dir = os.path.join(root_dir, 'workshop', 'searchlight')
# read ROI file
print 'Load ROI data ...'
roi_data = nib.load(roi_file).get_data()
# load cope data list
print 'Load cope files ...'
cope_list = get_subj_cope_list(root_dir, subj)
# get trial sequence info
print 'Load trial sequence info ...'
tag_list = get_subj_cope_tag(root_dir, subj)
# get ROI mvp data
roi_coord = niroi.get_roi_coord(roi_data)
[train_x, train_y, test_x, test_y] = get_roi_cope_mvps(cope_list,
tag_list,
roi_coord)
out_file = os.path.join(work_dir, 'cingulate_mvp_data.npz')
np.savez(out_file, train_x=train_x, train_y=train_y,
test_x=test_x, test_y=test_y)
