def extract_roi(self, mask, method='mean'):
""" Extract activity from mask
Args:
mask: nibabel mask can be binary or numbered for different rois
method: type of extraction method (default=mean)
Returns:
out: mean within each ROI across images
"""
if not isinstance(mask, nib.Nifti1Image):
raise ValueError('Make sure mask is a nibabel instance')
if len(np.unique(mask.get_data())) == 2:
all_mask = Brain_Data(mask)
if method is 'mean':
out = np.mean(self.data[:,np.where(all_mask.data)].squeeze(),axis=1)
elif len(np.unique(mask.get_data())) > 2:
all_mask = expand_mask(mask)
out = []
for i in range(all_mask.shape()[0]):
if method is 'mean':
out.append(np.mean(self.data[:,np.where(all_mask[i].data)].squeeze(),axis=1))
out = np.array(out)
return out
file_list = [x for x in file_list if 'betas' not in x]
file_list.sort()
conditions = [
os.path.basename(x).split(f'sub-{sub}_')[1].split('_denoised')[0]
for x in file_list
]
beta = Brain_Data(file_list)
# Next we will compute the pattern similarity across each beta image. We could do this for the whole brain, but it probably makes more sense to look within a single region. Many papers use a searchlight approach in which they examine the pattern similarity within a small sphere centered on each voxel. The advantage of this approach is that it uses the same number of voxels across searchlights and allows one to investigate the spatial topography at a relatively fine-scale. However, this procedure is fairly computationally expensive as it needs to be computed over each voxel and just like univariate analyses, will require stringent correction for multiple tests as we learned about in tutorial 12: Thresholding Group Analyses. Personally, I prefer to use whole-brain parcellations as they provide a nice balance between spatial specificity and computational efficiency. In this tutorial, we will continue to use functional regions of interest from our 50 ROI Neurosynth parcellation. This allows us to cover the entire brain with a relatively course spatial granularity, but requires several orders of magnitude of less computations than using a voxelwise searchlight approach. This means it will run much faster and will require us to use a considerably less stringent statistical threshold to correct for all independent tests. For example, for 50 tests bonferroni correction is p < 0.001 (i.e., .05/50). If we ever wanted better spatial granularity we could use increasingly larger parcellations (e.g., [100 or 200](https://neurovault.org/collections/2099/)).
#
# Let's load our parcellation mask so that we can examine the pattern similarity across these conditions for each ROI.
# In[78]:
mask = Brain_Data(os.path.join('..', 'masks', 'k50_2mm.nii.gz'))
mask_x = expand_mask(mask)
mask.plot()
# Ok, now we will want to calculate the pattern similar within each ROI across the 10 conditions.
#
# We will loop over each ROI and extract the pattern data across all conditions and then compute the correlation distance between each condition. This data will now be an `Adjacency` object that we discussed in the Lab 13: Connectivity. We will temporarily store this in a list.
#
# Notice that for each iteration of the loop we apply the ROI mask to our beta images and then calculate the correlation distance.
# In[79]:
out = []
for m in mask_x:
out.append(beta.apply_mask(m).distance(metric='correlation'))
# Get Cropped & Denoised HDF5 Files
result = ds.get(lsdir(os.path.join(datadir, 'fmriprep', '*', 'func', '*crop*hdf5')))
# Get Annotation File
result = ds.get(os.path.join(datadir, 'onsets', 'Sherlock_Segments_1000_NN_2017.xlsx'))
# Get Preprocessed Video Text
result = ds.get(os.path.join(datadir, 'stimuli', 'video_text.npy'))
## ROI responses while viewing Sherlock
Following the [functional alignment tutorial](http://naturalistic-data.org/features/notebooks/Functional_Alignment.html), we'll select out voxels in early visual cortex from the *Sherlock* dataset. We'll also examine primary auditory cortex and motor cortex responses. Then we'll apply the HyperTools pipeline to the dataset and visualize the responses within each ROI as a 3D image. Note you could also work with the Average ROI csv files as we did with the Dynamic Correlation tutorial. Here, we will load the full dataset and manually extract ROIs.
mask = Brain_Data('https://neurovault.org/media/images/8423/k50_2mm.nii.gz')
vectorized_mask = expand_mask(mask)
mask.plot()
rois = pd.read_csv('https://raw.githubusercontent.com/naturalistic-data-analysis/tutorial_development/master/hypertools/rois.csv', header=None, names=['ID', 'Region'])
rois.head()
roi_names = ['V1', 'A1', 'Precentral gyrus']
roi_ids = [int(rois.query(f'Region == "{r}"')['ID']) for r in roi_names]
my_rois = {k:v for k, v in zip(roi_names, roi_ids)}
# load subject's data and extract roi
data2 = {}
for run in ['Part1', 'Part2']:
file_list = lsdir(os.path.join(datadir, 'fmriprep', '*', 'func', f'*crop*{run}*hdf5'))
all_sub_roi = {}
for f_name in file_list:
