def make_age_maps(data_file, info_file, out_name):
"""
对每个顶点，计算跨同一年龄被试的平均和sem，分别保存在
out_name-mean.dscalar.nii, out_name-sem.dscalar.nii中
Args:
data_file (str): end with .dscalar.nii
shape=(n_subj, LR_count_32k)
info_file (str): subject info file
out_name (str): filename to save
"""
# prepare
data_maps = nib.load(data_file).get_fdata()
info_df = pd.read_csv(info_file)
ages = np.array(info_df['age in years'])
ages_uniq = np.unique(ages)
n_age = len(ages_uniq)
# calculate
mean_maps = np.ones((n_age, LR_count_32k)) * np.nan
sem_maps = np.ones((n_age, LR_count_32k)) * np.nan
for age_idx, age in enumerate(ages_uniq):
data = data_maps[ages == age]
mean_maps[age_idx] = np.mean(data, 0)
sem_maps[age_idx] = sem(data, 0)
# save
map_names = [str(i) for i in ages_uniq]
reader = CiftiReader(mmp_map_file)
save2cifti(f'{out_name}-mean.dscalar.nii', mean_maps,
reader.brain_models(), map_names)
save2cifti(f'{out_name}-sem.dscalar.nii', sem_maps, reader.brain_models(),
map_names)
def vtx_corr_col(data_file1, atlas_name, roi_name, data_file2, column, idx_col,
out_file):
"""
计算data_file1中指定atlas ROI内的所有顶点序列和data_file2中指定column序列的相关
Args:
data_file1 (str): end with .dscalar.nii
shape=(N, LR_count_32k)
atlas_name (str):
roi_name (str):
data_file2 (str): end with .csv
shape=(N, n_col)
column (str):
idx_col (int): specify index column of csv file
If None, means no index column.
"""
# prepare
reader = CiftiReader(data_file1)
maps = reader.get_data()
atlas = Atlas(atlas_name)
assert atlas.maps.shape == (1, LR_count_32k)
roi_idx_map = atlas.maps[0] == atlas.roi2label[roi_name]
maps = maps[:, roi_idx_map]
df = pd.read_csv(data_file2, index_col=idx_col)
col_vec = np.array(df[column])
col_vec = np.expand_dims(col_vec, 0)
# calculate
data = np.ones((1, LR_count_32k)) * np.nan
data[0, roi_idx_map] = 1 - cdist(col_vec, maps.T, 'correlation')[0]
# save
save2cifti(out_file, data, reader.brain_models())
def merge_smoothed_data(dataset_name, meas_name, sigma):
"""
合并我平滑过后的cifti文件
Args:
dataset_name (str): HCPD | HCPA
meas_name (str): thickness | myelin
sigma (float): the size of the gaussian surface smoothing kernel in mm
"""
# outputs
out_file = pjoin(work_dir,
f'{dataset_name}_{meas_name}_{sigma}mm.dscalar.nii')
# prepare
src_dir = pjoin(work_dir, f'{dataset_name}_{meas_name}_{sigma}mm')
src_file = pjoin(src_dir, '{sid}.dscalar.nii')
df = pd.read_csv(dataset_name2info[dataset_name])
n_subj = df.shape[0]
data = np.zeros((n_subj, LR_count_32k), np.float64)
# calculate
for subj_idx, subj_id in enumerate(df['subID']):
time1 = time.time()
meas_file = src_file.format(sid=subj_id)
data[subj_idx] = nib.load(meas_file).get_fdata()[0]
print(f'Finished: {subj_idx+1}/{n_subj},'
f'cost: {time.time() - time1} seconds.')
# save
mmp_reader = CiftiReader(mmp_map_file)
save2cifti(out_file, data, mmp_reader.brain_models(), df['subID'])
def make_non_outlier_map(fpath,
thr,
roi_name,
out_file_mask=None,
out_file_prob=None):
"""
将同时在thr%以上的被试中被认定为outlier的顶点判定为跨被试的outlier
If out_file_mask is not None, save mask map. (.dlabel.nii or .npy)
If out_file_prob is not None, save prob map. (.dscalar.nii)
"""
# prepare
data = np.load(fpath)
n_subj, n_vtx = data.shape
atlas1 = Atlas('Cole_visual_LR')
atlas2 = Atlas('Cole_visual_ROI')
assert atlas1.maps.shape == (1, LR_count_32k)
assert atlas2.maps.shape == (1, LR_count_32k)
roi_idx_map = atlas1.maps[0] == atlas1.roi2label[roi_name]
if roi_name == 'R_cole_visual':
prefix = 'R_'
elif roi_name == 'L_cole_visual':
prefix = 'L_'
else:
raise ValueError("error roi_name:", roi_name)
mmp_reader = CiftiReader(mmp_map_file)
mmp_lbl_tab = mmp_reader.label_tables()[0]
# calculate
if out_file_mask is not None:
data_tmp = np.sum(data, axis=0)
outlier_vec = data_tmp > thr / 100 * n_subj
print(f'#outliers/total: {np.sum(outlier_vec)}/{n_vtx}')
mask_npy = np.zeros(LR_count_32k, bool)
mask_npy[roi_idx_map] = ~outlier_vec
if out_file_mask.endswith('.npy'):
np.save(out_file_mask, mask_npy)
elif out_file_mask.endswith('.dlabel.nii'):
mask_cii = atlas2.maps.copy()
mask_cii[0, ~mask_npy] = np.nan
lbl_tab = nib.cifti2.cifti2.Cifti2LabelTable()
for roi, lbl in atlas2.roi2label.items():
if roi.startswith(prefix):
lbl_tab[lbl] = mmp_lbl_tab[lbl]
save2cifti(out_file_mask,
mask_cii,
mmp_reader.brain_models(),
label_tables=[lbl_tab])
else:
raise ValueError("Not supported file name:", out_file_mask)
if out_file_prob is not None:
data_tmp = np.mean(data, axis=0)
prob_map = np.ones((1, LR_count_32k), dtype=np.float64) * np.nan
prob_map[0, roi_idx_map] = data_tmp
assert out_file_prob.endswith('.dscalar.nii')
save2cifti(out_file_prob, prob_map, mmp_reader.brain_models())
def pca(data_file, atlas_name, roi_name, n_component, axis, out_name):
"""
对n_subj x n_vtx形状的矩阵进行PCA降维
Args:
data_file (str): end with .dscalar.nii
shape=(n_subj, LR_count_32k)
atlas_name (str): include ROIs' labels and mask map
roi_name (str): 决定选用哪个区域内的顶点来参与PCA
n_component (int): the number of components
axis (str): vertex | subject
vertex: 对顶点数量进行降维，得到几个主成分时间序列，
观察某个主成分在各顶点上的权重，刻画其空间分布。
subject: 对被试数量进行降维，得到几个主成分map，
观察某个主成分在各被试上的权重，按年龄排序即可得到时间序列。
out_name (str): output name
If axis=vertex, output
1. n_subj x n_component out_name.csv
2. n_component x LR_count_32k out_name.dscalar.nii
3. out_name.pkl with fitted PCA model
"""
# prepare
component_names = [f'C{i}' for i in range(1, n_component + 1)]
meas_maps = nib.load(data_file).get_fdata()
atlas = Atlas(atlas_name)
assert atlas.maps.shape == (1, LR_count_32k)
roi_idx_map = atlas.maps[0] == atlas.roi2label[roi_name]
meas_maps = meas_maps[:, roi_idx_map]
bms = CiftiReader(mmp_map_file).brain_models()
# calculate
pca = PCA(n_components=n_component)
data = np.ones((n_component, LR_count_32k), np.float64) * np.nan
if axis == 'vertex':
X = meas_maps
pca.fit(X)
Y = pca.transform(X)
df = pd.DataFrame(data=Y, columns=component_names)
data[:, roi_idx_map] = pca.components_
elif axis == 'subject':
X = meas_maps.T
pca.fit(X)
Y = pca.transform(X)
df = pd.DataFrame(data=pca.components_.T, columns=component_names)
data[:, roi_idx_map] = Y.T
else:
raise ValueError('Invalid axis:', axis)
# save
df.to_csv(f'{out_name}.csv', index=False)
save2cifti(f'{out_name}.dscalar.nii', data, bms, component_names)
pkl.dump(pca, open(f'{out_name}.pkl', 'wb'))
def separate_networks():
"""
把ColeNetwork的12个网络分到单独的map里。
每个map的MMP parcel的label保留原来的样子。
需要注意的是multimodal_glasser的MMP_mpmLR32k.dlabel.nii中，
ROI label的编号1~180是右脑，181~360是左脑。0对应的是???
而cortex_parcel_network_assignments.mat中0~359的index是左脑在先
"""
import numpy as np
import nibabel as nib
from scipy.io import loadmat
from cxy_visual_dev.lib.ColeNet import get_name_label_of_ColeNetwork
from magicbox.io.io import CiftiReader, save2cifti
# inputs
mmp_file = '/nfs/p1/atlases/multimodal_glasser/surface/'\
'MMP_mpmLR32k.dlabel.nii'
roi2net_file = pjoin(cole_dir, 'cortex_parcel_network_assignments.mat')
# outputs
out_file = pjoin(work_dir, 'networks.dlabel.nii')
# load
mmp_reader = CiftiReader(mmp_file)
mmp_map = mmp_reader.get_data()[0]
lbl_tab_raw = mmp_reader.label_tables()[0]
roi2net = loadmat(roi2net_file)['netassignments'][:, 0]
roi2net = np.r_[roi2net[180:], roi2net[:180]]
net_labels = np.unique(roi2net)
# prepare
data = np.zeros((len(net_labels), len(mmp_map)), dtype=np.uint16)
map_names = []
label_tables = []
net_lbl2name = {}
for name, lbl in zip(*get_name_label_of_ColeNetwork()):
net_lbl2name[lbl] = name
# calculate
for net_idx, net_lbl in enumerate(net_labels):
roi_labels = np.where(roi2net == net_lbl)[0] + 1
lbl_tab = nib.cifti2.cifti2.Cifti2LabelTable()
lbl_tab[0] = lbl_tab_raw[0]
for roi_lbl in roi_labels:
data[net_idx, mmp_map == roi_lbl] = roi_lbl
lbl_tab[roi_lbl] = lbl_tab_raw[roi_lbl]
map_names.append(net_lbl2name[net_lbl])
label_tables.append(lbl_tab)
# save
save2cifti(out_file,
data,
mmp_reader.brain_models(),
map_names,
label_tables=label_tables)
def zscore_data(data_file, out_file):
"""
对每个被试做全脑zscore
Args:
data_file (str): .dscalar.nii
out_file (str): .dscalar.nii
"""
reader = CiftiReader(data_file)
data = reader.get_data()
data = zscore(data, 1)
save2cifti(out_file, data, reader.brain_models(), reader.map_names())
def kendall2cifti(data_file, rois, atlas_name, out_file):
"""
把指定ROI的Tau值和p值存成cifti格式 方便可视化在大脑上
"""
# prepare
df = pd.read_csv(data_file, index_col=0)
atlas = Atlas(atlas_name)
assert atlas.maps.shape == (1, LR_count_32k)
out_data = np.ones((2, LR_count_32k)) * np.nan
# calculate
for roi in rois:
mask = atlas.maps[0] == atlas.roi2label[roi]
out_data[0, mask] = df.loc['tau', roi]
out_data[1, mask] = -np.log10(df.loc['p', roi])
# save
save2cifti(out_file, out_data,
CiftiReader(mmp_map_file).brain_models(), ['tau', '-lg(p)'])
def polyfit(data_file, info_file, deg, out_file):
"""
对时间序列进行多项式拟合
Args:
data_file (str): .csv | .dscalar.nii
If is .csv, fit each column with ages.
If is .dscalar.nii, fit each vertex with ages.
info_file (str): .csv
deg (int): degree of polynomial
out_file (str): file to output
same postfix with data_file
"""
# prepare
if data_file.endswith('.csv'):
assert out_file.endswith('.csv')
reader = pd.read_csv(data_file)
data = np.array(reader)
elif data_file.endswith('.dscalar.nii'):
assert out_file.endswith('.dscalar.nii')
reader = CiftiReader(data_file)
data = reader.get_data()
else:
raise ValueError(1)
# calculate
ages = np.array(pd.read_csv(info_file)['age in years'])
coefs = np.polyfit(ages, data, deg)
n_row = coefs.shape[0]
if deg == 1:
assert n_row == 2
row_names = ['coef', 'intercept']
else:
row_names = [''] * n_row
raise Warning("Can't provide row names for degree:", deg)
# save
if out_file.endswith('.csv'):
df = pd.DataFrame(coefs, row_names, reader.columns)
df.to_csv(out_file)
else:
save2cifti(out_file, coefs, reader.brain_models(), row_names)
def separate_networks():
"""
把ColeNetwork的12个网络分到单独的map里。
每个map的MMP parcel的label保留原来的样子。
需要注意的是multimodal_glasser的MMP_mpmLR32k.dlabel.nii中，
ROI label的编号1~180是右脑，181~360是左脑。0对应的是???
而cortex_parcel_network_assignments.mat中0~359的index是左脑在先
"""
# outputs
out_file = pjoin(work_dir, 'networks.dlabel.nii')
# load
mmp_reader = CiftiReader(mmp_map_file)
mmp_map = mmp_reader.get_data()[0]
lbl_tab_raw = mmp_reader.label_tables()[0]
roi2net = loadmat(cole_net_assignment_file)['netassignments'][:, 0]
roi2net = np.r_[roi2net[180:], roi2net[:180]]
net_labels = np.unique(roi2net)
# prepare
data = np.zeros((len(net_labels), len(mmp_map)), dtype=np.uint16)
map_names = []
label_tables = []
# calculate
for net_idx, net_lbl in enumerate(net_labels):
roi_labels = np.where(roi2net == net_lbl)[0] + 1
lbl_tab = nib.cifti2.cifti2.Cifti2LabelTable()
lbl_tab[0] = lbl_tab_raw[0]
for roi_lbl in roi_labels:
data[net_idx, mmp_map == roi_lbl] = roi_lbl
lbl_tab[roi_lbl] = lbl_tab_raw[roi_lbl]
map_names.append(cole_label2name[net_lbl])
label_tables.append(lbl_tab)
# save
save2cifti(out_file,
data,
mmp_reader.brain_models(),
map_names,
label_tables=label_tables)
def merge_data(dataset_name, meas_name):
"""
把所有被试的数据合并到一个cifti文件里
Args:
dataset_name (str): HCPD | HCPA
meas_name (str): thickness | myelin
"""
# outputs
out_file = pjoin(work_dir, f'{dataset_name}_{meas_name}.dscalar.nii')
# prepare
dataset_dir = dataset_name2dir[dataset_name]
meas2file = {
'myelin':
pjoin(
dataset_dir,
'fmriresults01/{sid}_V1_MR/MNINonLinear/fsaverage_LR32k/'
'{sid}_V1_MR.MyelinMap_BC_MSMAll.32k_fs_LR.dscalar.nii'),
'thickness':
pjoin(
dataset_dir,
'fmriresults01/{sid}_V1_MR/MNINonLinear/fsaverage_LR32k/'
'{sid}_V1_MR.thickness_MSMAll.32k_fs_LR.dscalar.nii')
}
df = pd.read_csv(dataset_name2info[dataset_name])
n_subj = df.shape[0]
data = np.zeros((n_subj, LR_count_32k), np.float64)
# calculate
for subj_idx, subj_id in enumerate(df['subID']):
time1 = time.time()
meas_file = meas2file[meas_name].format(sid=subj_id)
data[subj_idx] = nib.load(meas_file).get_fdata()[0]
print(f'Finished: {subj_idx+1}/{n_subj},'
f'cost: {time.time() - time1} seconds.')
# save
mmp_reader = CiftiReader(mmp_map_file)
save2cifti(out_file, data, mmp_reader.brain_models(), df['subID'])
def mask_maps(data_file, atlas_name, roi_name, out_file):
"""
把data map在指定atlas的ROI以外的部分全赋值为nan
Args:
data_file (str): end with .dscalar.nii
shape=(n_map, LR_count_32k)
atlas_name (str):
roi_name (str):
out_file (str):
"""
# prepare
reader = CiftiReader(data_file)
data = reader.get_data()
atlas = Atlas(atlas_name)
assert atlas.maps.shape == (1, LR_count_32k)
roi_idx_map = atlas.maps[0] == atlas.roi2label[roi_name]
# calculate
data[:, ~roi_idx_map] = np.nan
# save
save2cifti(out_file, data, reader.brain_models(), reader.map_names())