ENH: Add --rescale-intensities and name_source to N4BiasFieldCorrection

Author: oesteban

nipype/interfaces/ants/segmentation.py

@@ -7,7 +7,7 @@
 
 import os
 from ...external.due import BibTeX
-from ...utils.filemanip import split_filename, copyfile, which
+from ...utils.filemanip import split_filename, copyfile, which, fname_presuffix
 from ..base import TraitedSpec, File, traits, InputMultiPath, OutputMultiPath, isdefined
 from .base import ANTSCommand, ANTSCommandInputSpec
 
@@ -274,18 +274,20 @@ class N4BiasFieldCorrectionInputSpec(ANTSCommandInputSpec):
         argstr='--input-image %s',
         mandatory=True,
         desc=('input for bias correction. Negative values or values close to '
-            'zero should be processed prior to correction'))
+              'zero should be processed prior to correction'))
     mask_image = File(
         argstr='--mask-image %s',
         desc=('image to specify region to perform final bias correction in'))
     weight_image = File(
         argstr='--weight-image %s',
         desc=('image for relative weighting (e.g. probability map of the white '
-            'matter) of voxels during the B-spline fitting. '))
+              'matter) of voxels during the B-spline fitting. '))
     output_image = traits.Str(
         argstr='--output %s',
         desc='output file name',
-        genfile=True,
+        name_source=['input_image'],
+        name_template='%s_corrected',
+        keep_extension=True,
         hash_files=False)
     bspline_fitting_distance = traits.Float(argstr="--bspline-fitting %s")
     bspline_order = traits.Int(requires=['bspline_fitting_distance'])
@@ -306,6 +308,14 @@ class N4BiasFieldCorrectionInputSpec(ANTSCommandInputSpec):
         usedefault=True,
         desc='copy headers of the original image into the '
         'output (corrected) file')
+    rescale_intensities = traits.Bool(
+        False, usedefault=True, argstr='-r',
+        desc="""\
+At each iteration, a new intensity mapping is calculated and applied but there
+is nothing which constrains the new intensity range to be within certain values.
+The result is that the range can "drift" from the original at each iteration.
+This option rescales to the [min,max] range of the original image intensities
+within the user-specified mask.""")
 
 
 class N4BiasFieldCorrectionOutputSpec(TraitedSpec):
@@ -314,7 +324,10 @@ class N4BiasFieldCorrectionOutputSpec(TraitedSpec):
 
 
 class N4BiasFieldCorrection(ANTSCommand):
-    """N4 is a variant of the popular N3 (nonparameteric nonuniform normalization)
+    """
+    Bias field correction.
+
+    N4 is a variant of the popular N3 (nonparameteric nonuniform normalization)
     retrospective bias correction algorithm. Based on the assumption that the
     corruption of the low frequency bias field can be modeled as a convolution of
     the intensity histogram by a Gaussian, the basic algorithmic protocol is to
@@ -373,28 +386,9 @@ class N4BiasFieldCorrection(ANTSCommand):
     input_spec = N4BiasFieldCorrectionInputSpec
     output_spec = N4BiasFieldCorrectionOutputSpec
 
-    def _gen_filename(self, name):
-        if name == 'output_image':
-            output = self.inputs.output_image
-            if not isdefined(output):
-                _, name, ext = split_filename(self.inputs.input_image)
-                output = name + '_corrected' + ext
-            return output
-
-        if name == 'bias_image':
-            output = self.inputs.bias_image
-            if not isdefined(output):
-                _, name, ext = split_filename(self.inputs.input_image)
-                output = name + '_bias' + ext
-            return output
-        return None
-
     def _format_arg(self, name, trait_spec, value):
-        if ((name == 'output_image') and
-            (self.inputs.save_bias or isdefined(self.inputs.bias_image))):
-            bias_image = self._gen_filename('bias_image')
-            output = self._gen_filename('output_image')
-            newval = '[ %s, %s ]' % (output, bias_image)
+        if name == 'output_image' and getattr(self, '_out_bias_file', None):
+            newval = '[ %s, %s ]' % (value, getattr(self, '_out_bias_file'))
             return trait_spec.argstr % newval
 
         if name == 'bspline_fitting_distance':
@@ -418,34 +412,40 @@ def _format_arg(self, name, trait_spec, value):
             name, trait_spec, value)
 
     def _parse_inputs(self, skip=None):
-        if skip is None:
-            skip = []
-        skip += ['save_bias', 'bias_image']
+        skip = (skip or []) + ['save_bias', 'bias_image']
+        if self.inputs.save_bias or isdefined(self.inputs.bias_image):
+            bias_image = self.inputs.bias_image
+            if not isdefined(bias_image):
+                bias_image = fname_presuffix(os.path.basename(self.inputs.input_image),
+                                             suffix='_bias')
+            setattr(self, '_out_bias_file', bias_image)
+        else:
+            try:
+                delattr(self, '_out_bias_file')
+            except AttributeError:
+                pass
         return super(N4BiasFieldCorrection, self)._parse_inputs(skip=skip)
 
     def _list_outputs(self):
-        outputs = self._outputs().get()
-        outputs['output_image'] = os.path.abspath(
-            self._gen_filename('output_image'))
-
-        if self.inputs.save_bias or isdefined(self.inputs.bias_image):
-            outputs['bias_image'] = os.path.abspath(
-                self._gen_filename('bias_image'))
+        outputs = super(N4BiasFieldCorrection, self)._list_outputs()
+        bias_image = getattr(self, '_out_bias_file', None)
+        if bias_image:
+            outputs['bias_image'] = bias_image
         return outputs
 
     def _run_interface(self, runtime, correct_return_codes=(0, )):
         runtime = super(N4BiasFieldCorrection, self)._run_interface(
             runtime, correct_return_codes)
 
         if self.inputs.copy_header and runtime.returncode in correct_return_codes:
-            self._copy_header(self._gen_filename('output_image'))
-            if self.inputs.save_bias or isdefined(self.inputs.bias_image):
-                self._copy_header(self._gen_filename('bias_image'))
+            self._copy_header(self.inputs.output_image)
+            if getattr(self, '_out_bias_file', None):
+                self._copy_header(getattr(self, '_out_bias_file'))
 
         return runtime
 
     def _copy_header(self, fname):
-        """Copy header from input image to an output image"""
+        """Copy header from input image to an output image."""
         import nibabel as nb
         in_img = nb.load(self.inputs.input_image)
         out_img = nb.load(fname, mmap=False)

nipype/interfaces/ants/tests/test_auto_N4BiasFieldCorrection.py

@@ -41,8 +41,14 @@ def test_N4BiasFieldCorrection_inputs():
         ),
         output_image=dict(
             argstr='--output %s',
-            genfile=True,
             hash_files=False,
+            keep_extension=True,
+            name_source=['input_image'],
+            name_template='%s_corrected',
+        ),
+        rescale_intensities=dict(
+            argstr='-r',
+            usedefault=True,
         ),
         save_bias=dict(
             mandatory=True,