def meta_confound(label_eg,
label_cg,
csv_dir='./data/meta_csv/{}_{}/confound/',
output_dir='./results/meta/{}_{}/confound/'):
models = {}
csv_dir = csv_dir.format(label_eg, label_cg)
output_dir = output_dir.format(label_eg, label_cg)
if not os.path.exists(output_dir):
os.mkdir(output_dir)
for confound in confounds:
csv_path = csv_dir + confound + '.csv'
output_path = output_dir + confound + '.png'
model = csv_meta_analysis(csv_path, model_type='random')
print(label_eg, label_cg, confound, model.p)
model.plot_forest(title=confound, save_path=output_path, show=False)
models[confound] = model
return models
def meta_gmv(label_eg,
label_cg,
mask,
save_nii=True,
csv_prefix='roi_gmv_removed',
csv_dir='./data/meta_csv',
out_dir='./results/meta'):
models = {}
prefix = '{}_{}/{}'.format(label_eg, label_cg, csv_prefix)
csv_dir = os.path.join(csv_dir, prefix)
out_dir = os.path.join(out_dir, prefix)
if not os.path.isdir(out_dir):
os.mkdir(out_dir)
csvs = os.listdir(csv_dir)
for f in csvs:
csv_path = os.path.join(csv_dir, f)
model = csv_meta_analysis(csv_path, model_type='random')
models[f[:-4]] = model
if save_nii:
nii_array = mask.data.astype(np.float32)
p_array = nii_array
ll = mask.labels.tolist()
for k, v in models.items():
_id = int(k)
nii_array[nii_array == _id] = v.total_effect_size
p_array[p_array == _id] = v.p
ll.remove(_id)
for i in ll:
nii_array[nii_array == i] = 0
path = os.path.join(out_dir, 'es.nii')
p_path = os.path.join(out_dir, 'p.nii')
utils.gen_nii(nii_array, mask.nii, path)
utils.gen_nii(p_array, mask.nii, p_path)
return models
def test_csv():
result_model = main.csv_meta_analysis('../test_data/test.csv', )
return result_model.get_results()
def meta_ct(label_eg,
label_cg,
p_thres=0.001,
topn=0.3,
save_gii=True,
save_nii=False,
mask=None,
csv_prefix='roi_ct_removed',
csv_dir_prefix='./data/meta_csv',
out_dir_prefix='./results/meta'):
models = {}
surfix = '{}_{}/{}'.format(label_eg, label_cg, csv_prefix)
out_dir = os.path.join(out_dir_prefix, surfix)
if not os.path.isdir(out_dir):
os.mkdir(out_dir)
csv_dir = os.path.join(csv_dir_prefix, surfix)
csvs = os.listdir(csv_dir)
for f in csvs:
csv_path = os.path.join(csv_dir, f)
model = csv_meta_analysis(csv_path, model_type='random')
models[f[:-4]] = model
if save_gii:
for annot, surf, lr in zip(annots, surfs, l_r):
a = surface.load_surf_data(annot_dir.format(annot))
a = a.astype(np.float32)
b = surf_dir.format(surf)
tmp_gii = nib.load(b)
cor_model, _ = bon_cor(models, thres=p_thres)
ll = np.unique(a).tolist()
_, sorted_models = sort_models(cor_model, descend=False)
top_es = sorted_models[int(len(sorted_models) *
topn)].total_effect_size
for k, v in cor_model.items():
_id = np.float32(k)
if v.total_effect_size <= top_es:
a[a == _id] = v.total_effect_size
if _id in ll:
ll.remove(_id)
for i in ll:
a[a == i] = 0
gdarray = GiftiDataArray.from_array(a, intent=0)
tmp_gii.remove_gifti_data_array_by_intent(0)
tmp_gii.add_gifti_data_array(gdarray)
path = os.path.join(
out_dir,
'es_{}_bon{}_top{}.gii'.format(lr,
str(p_thres)[2:],
str(topn)[1:]))
nib.save(tmp_gii, path)
if save_nii:
nii_array = mask.data.astype(np.float32)
p_array = nii_array
ll = mask.labels.tolist()
for k, v in models.items():
_id = int(k)
nii_array[nii_array == _id] = v.total_effect_size
p_array[p_array == _id] = v.p
ll.remove(_id)
for i in ll:
nii_array[nii_array == i] = 0
path = os.path.join(out_dir, 'es.nii')
p_path = os.path.join(out_dir, 'p.nii')
utils.gen_nii(nii_array, mask.nii, path)
utils.gen_nii(p_array, mask.nii, p_path)
return models
def test_plot():
result_model = main.csv_meta_analysis('../test_data/test.csv', )
result_model.plot_forest()