Merge pull request #116 from Parietal-INRIA/feat/one-parcellation-tem…

…plate

Refactor template alignments to keep a fixed parcellation

examples/plot_template_alignment.py

@@ -1,13 +1,14 @@
 # -*- coding: utf-8 -*-
-
 """
 Template-based prediction.
 ==========================
 
-In this tutorial, we show how to better predict new contrasts for a target
-subject using many source subjects corresponding contrasts. For this purpose,
-we create a template to which we align the target subject, using shared information.
-We then predict new images for the target and compare them to a baseline.
+In this tutorial, we show how to improve inter-subject similarity using a template
+computed across multiple source subjects. For this purpose, we create a template
+using Procrustes alignment (hyperalignment) to which we align the target subject,
+using shared information. We then compare the voxelwise similarity between the
+target subject and the template to the similarity between the target subject and
+the anatomical Euclidean average of the source subjects.
 
 We mostly rely on Python common packages and on nilearn to handle
 functional data in a clean fashion.
@@ -36,7 +37,7 @@
 )
 
 ###############################################################################
-# Definine a masker
+# Define a masker
 # -----------------
 # We define a nilearn masker that will be used to handle relevant data.
 #   For more information, visit :
@@ -64,22 +65,17 @@
 template_train = []
 for i in range(5):
     template_train.append(concat_imgs(imgs[i]))
-target_train = df[df.subject == "sub-07"][df.acquisition == "ap"].path.values
 
-# For subject sub-07, we split it in two folds:
-#   - target train: sub-07 AP contrasts, used to learn alignment to template
-#   - target test: sub-07 PA contrasts, used as a ground truth to score predictions
-# We make a single 4D Niimg from our list of 3D filenames
+# sub-07 (that is 5th in the list) will be our left-out subject.
+# We make a single 4D Niimg from our list of 3D filenames.
 
-target_train = concat_imgs(target_train)
-target_test = df[df.subject == "sub-07"][df.acquisition == "pa"].path.values
+left_out_subject = concat_imgs(imgs[5])
 
 ###############################################################################
 # Compute a baseline (average of subjects)
 # ----------------------------------------
 # We create an image with as many contrasts as any subject representing for
 # each contrast the average of all train subjects maps.
-#
 
 import numpy as np
 
@@ -92,70 +88,53 @@
 # ---------------------------------------------
 # We define an estimator using the class TemplateAlignment:
 #   * We align the whole brain through 'multiple' local alignments.
-#   * These alignments are calculated on a parcellation of the brain in 150 pieces,
+#   * These alignments are calculated on a parcellation of the brain in 50 pieces,
 #     this parcellation creates group of functionnally similar voxels.
 #   * The template is created iteratively, aligning all subjects data into a
 #     common space, from which the template is inferred and aligning again to this
 #     new template space.
 #
 
-from nilearn.image import index_img
-
 from fmralign.template_alignment import TemplateAlignment
 
+# We use Procrustes/scaled orthogonal alignment method
 template_estim = TemplateAlignment(
-    n_pieces=150, alignment_method="ridge_cv", mask=masker
+    n_pieces=50,
+    alignment_method="scaled_orthogonal",
+    mask=masker,
 )
 template_estim.fit(template_train)
+procrustes_template = template_estim.template
 
 ###############################################################################
 # Predict new data for left-out subject
 # -------------------------------------
-# We use target_train data to fit the transform, indicating it corresponds to
-# the contrasts indexed by train_index and predict from this learnt alignment
-# contrasts corresponding to template test_index numbers.
-# For each train subject and for the template, the AP contrasts are sorted from
-# 0, to 53, and then the PA contrasts from 53 to 106.
-#
-
-train_index = range(53)
-test_index = range(53, 106)
-
-# We input the mapping image target_train in a list, we could have input more
-# than one subject for which we'd want to predict : [train_1, train_2 ...]
+# We predict the contrasts of the left-out subject using the template we just
+# created. We use the transform method of the estimator. This method takes the
+# left-out subject as input, computes a pairwise alignment with the template
+# and returns the aligned data.
 
-prediction_from_template = template_estim.transform(
-    [target_train], train_index, test_index
-)
-
-# As a baseline prediction, let's just take the average of activations across subjects.
-
-prediction_from_average = index_img(average_subject, test_index)
+predictions_from_template = template_estim.transform(left_out_subject)
 
 ###############################################################################
 # Score the baseline and the prediction
 # -------------------------------------
 # We use a utility scoring function to measure the voxelwise correlation
-# between the prediction and the ground truth. That is, for each voxel, we
-# measure the correlation between its profile of activation without and with
-# alignment, to see if alignment was able to predict a signal more alike the ground truth.
-#
+# between the images. That is, for each voxel, we measure the correlation between
+# its profile of activation without and with alignment, to see if template-based
+# alignment was able to improve inter-subject similarity.
 
 from fmralign.metrics import score_voxelwise
 
-# Now we use this scoring function to compare the correlation of predictions
-# made from group average and from template with the real PA contrasts of sub-07
-
 average_score = masker.inverse_transform(
-    score_voxelwise(target_test, prediction_from_average, masker, loss="corr")
+    score_voxelwise(left_out_subject, average_subject, masker, loss="corr")
 )
 template_score = masker.inverse_transform(
     score_voxelwise(
-        target_test, prediction_from_template[0], masker, loss="corr"
+        predictions_from_template, procrustes_template, masker, loss="corr"
     )
 )
 
-
 ###############################################################################
 # Plotting the measures
 # ---------------------
@@ -167,13 +146,12 @@
 baseline_display = plotting.plot_stat_map(
     average_score, display_mode="z", vmax=1, cut_coords=[-15, -5]
 )
-baseline_display.title("Group average correlation wt ground truth")
+baseline_display.title("Left-out subject correlation with group average")
 display = plotting.plot_stat_map(
     template_score, display_mode="z", cut_coords=[-15, -5], vmax=1
 )
-display.title("Template-based prediction correlation wt ground truth")
+display.title("Aligned subject correlation with Procrustes template")
 
 ###############################################################################
 # We observe that creating a template and aligning a new subject to it yields
-# a prediction that is better correlated with the ground truth than just using
-# the average activations of subjects.
+# better inter-subject similarity than regular euclidean averaging.

fmralign/pairwise_alignment.py

@@ -237,12 +237,12 @@ def fit(self, X, Y):
 
         return self
 
-    def transform(self, X):
+    def transform(self, img):
         """Predict data from X.
 
         Parameters
         ----------
-        X: Niimg-like object
+        img: Niimg-like object
             Source data
 
         Returns
@@ -255,7 +255,7 @@ def transform(self, X):
                 "This instance has not been fitted yet. "
                 "Please call 'fit' before 'transform'."
             )
-        parceled_data_list = self.parcel_masker.transform(X)
+        parceled_data_list = self.parcel_masker.transform(img)
         transformed_img = Parallel(
             self.n_jobs, prefer="threads", verbose=self.verbose
         )(
@@ -266,7 +266,6 @@ def transform(self, X):
             return transformed_img[0]
         else:
             return transformed_img
-        return transformed_img
 
     # Make inherited function harmless
     def fit_transform(self):
@@ -291,7 +290,7 @@ def get_parcellation(self):
         if hasattr(self, "parcel_masker"):
             check_is_fitted(self)
             labels = self.parcel_masker.get_labels()
-            parcellation_img = self.parcel_masker.get_parcellation()
+            parcellation_img = self.parcel_masker.get_parcellation_img()
             return labels, parcellation_img
         else:
             raise AttributeError(

fmralign/preprocessing.py

@@ -194,7 +194,7 @@ def get_labels(self):
             )
         return self.labels
 
-    def get_parcellation(self):
+    def get_parcellation_img(self):
         """Return the parcellation image.
 
         Returns