fix: use sample_mask of nilearn.signal.clean for scrubbing in fMRIPre…

…pConfoundRemover

junifer/preprocess/confounds/fmriprep_confound_remover.py

@@ -366,14 +366,6 @@ def _process_fmriprep_spec(
                     "Check if this file is really an fmriprep confounds file. "
                     "You can also deactivate spike (set spike = None)."
                 )
-        # Add std_dvars
-        if self.std_dvars_threshold is not None:
-            if "std_dvars" not in available_vars:
-                raise_error(
-                    "Invalid confounds file. Missing std_dvars "
-                    "(standardized DVARS) confound. "
-                    "Check if this file is really an fMRIPrep confounds file. "
-                )
         out = to_select, squares_to_compute, derivatives_to_compute, spike_name
         return out
 
@@ -423,14 +415,6 @@ def _pick_confounds(self, input: dict[str, Any]) -> pd.DataFrame:
             out_df["spike"] = fd
             to_select.append("spike")
 
-        # add binary std_dvars regressor if needed at given threshold
-        if self.std_dvars_threshold is not None:
-            std_dvars = out_df["std_dvars"].copy()
-            std_dvars.loc[std_dvars > self.std_dvars_threshold] = 1
-            std_dvars.loc[std_dvars != 1] = 0
-            out_df["std_dvars"] = std_dvars
-            to_select.append("std_dvars")
-
         # Now pick all the relevant confounds
         out_df = out_df[to_select]
 
@@ -442,6 +426,41 @@ def _pick_confounds(self, input: dict[str, Any]) -> pd.DataFrame:
 
         return out_df
 
+    def _get_scrub_mask(self, input: dict[str, Any]) -> np.ndarray:
+        """Get boolean mask for scrubbing.
+
+        Parameters
+        ----------
+        input : dict
+            Dictionary containing the ``BOLD.confounds`` value from the
+            Junifer Data object.
+
+        Returns
+        -------
+        numpy.ndarray
+            Index of volumes to be kept.
+
+        Raises
+        ------
+        RuntimeError
+            If ``std_dvars`` is not found in the confounds file.
+
+        """
+        confounds_df = input["data"]
+        # Check confounds file
+        if "std_dvars" not in confounds_df.columns:
+            raise_error(
+                "Invalid confounds file. Missing std_dvars "
+                "(standardized DVARS) confound. "
+                "Check if this file is really an fMRIPrep confounds file. "
+            )
+        # Make first row 0 and then threshold
+        return np.flatnonzero(
+            (
+                confounds_df["std_dvars"].fillna(0) > self.std_dvars_threshold
+            ).to_numpy()
+        )
+
     def _validate_data(
         self,
         input: dict[str, Any],
@@ -605,6 +624,13 @@ def preprocess(
                     }
                 }
             )
+
+        # Set up scrubbing mask if needed
+        if self.std_dvars_threshold is not None:
+            sample_mask = self._get_scrub_mask(input["confounds"])
+        else:
+            sample_mask = None
+
         # Clean image
         logger.info("Cleaning image using nilearn")
         logger.debug(f"\tdetrend: {self.detrend}")
@@ -622,6 +648,7 @@ def preprocess(
             high_pass=self.high_pass,
             t_r=t_r,
             mask_img=mask_img,
+            clean__sample_mask=sample_mask,
         )
         # Fix t_r as nilearn messes it up
         cleaned_img.header["pixdim"][4] = t_r