Data-centric strategies enable cross-center generalization of convolutional neural networks for CT brain extraction.
Paper
Loaded from Europe PMC by your browser; OSCR does not store it: doi.org · Europe PMC
The paper is loaded when this pane is shown.
The authors' code
Python · 562 lines · 24 KB · no license
- """
- AugGen
- ======
- This module implements the AugmentationGenerator class which can be used to
- create augmentations of images and their segmentations by performing
- affine transformations.
- """
- import numpy as np
- from scipy import ndimage
- import nibabel.orientations as orx
- class AugmentationGenerator(object):
- """Initializes an augmentation generator object.
- ### Args
- `rotation_z` (int): Max angle of rotation around z-axis in degrees.
- Defaults to 15.
- `rotation_x` (int): Max angle of rotation around x-axis in degrees.
- Defaults to 2.
- `rotation_y` (int): Max angle of rotation around y-axis in degrees.
- Defaults to 2.
- `translation_xy` (int): Max number of pixels to translate in the
- xy plane. Defaults to 15.
- `translation_z` (int): Max number of pixels to translate in the
- z axis. Defaults to 0.
- `scale_xy` (float): Scaling factor is 1 +/- `scale_delta_max`,
- Defaults to 0.1.
- `scale_z` (int): Scaling factor is 1 +/- `scale_delta_max`,
- Defaults to 0.
- `thickness` (array-like): Slice thickness to pick from. Defaults to (1,)
- `flip_h` (bool): Determines whether horizontal flipping can occur.
- Defaults to True.
- `flip_v` (bool): Determines whether vertical flipping can occur.
- Defaults to False.
- `filling_mode` (str): mode passed to ndimage.affine_transform function,
- ('constant', 'nearest', 'reflect', 'mirror', 'wrap). Defaults to
- 'constant
- `padding_value` (float): Determines the value used for filling if
- filling mode is 'constant
- `bone_edges` (bool): Determines whether mask pixels are adjusted
- to account for bone edges. Defaults to False.
- """
- def __init__(self, rotation_z=0, rotation_x=0, rotation_y=0,
- translation_xy=0, translation_z=0, scale_xy=0,
- scale_z=0, thickness=(0,), flip_h=False, flip_v=False,
- filling_mode='constant', padding_value=0.0, bone_edges=False, mask_PV=0):
- self.rotation_z = rotation_z
- self.rotation_x = rotation_x
- self.rotation_y = rotation_y
- self.translation_xy = translation_xy
- self.translation_z = translation_z
- self.scale_xy = scale_xy
- self.scale_z = scale_z
- self.thickness = thickness
- self.flip_h = bool(flip_h)
- self.flip_v = bool(flip_v)
- self.filling_mode = filling_mode
- self.padding_value = padding_value
- self.mask_PV = mask_PV
- self.bone_edges = bone_edges
- def __repr__(self):
- string = 'AugmentationGenerator('
- string += 'rotation_z: ' + str(self.rotation_z) + ', '
- string += 'rotation_x: ' + str(self.rotation_x) + ', '
- string += 'rotation_y: ' + str(self.rotation_y) + ', '
- string += 'translation_xy: ' + str(self.translation_xy) + ', '
- string += 'translation_z: ' + str(self.translation_z) + ', '
- string += 'scale_xy: ' + str(self.scale_xy) + ', '
- string += 'scale_z: ' + str(self.scale_z) + ', '
- string += 'thickness: ' + str(self.thickness) + ', '
- string += 'flip_h: ' + str(self.flip_h) + ', '
- string += 'flip_v: ' + str(self.flip_v) + ')'
- string += 'filling_mode: ' + str(self.filling_mode) + ')'
- string += 'padding_value: ' + str(self.padding_value) + ')'
- string += 'bone_edges: ' + str(self.bone_edges) + ')'
- return string
- def __str__(self):
- string = 'ImageAugmentationGenerator parameters:\n'
- string += ' rotation_z: ' + str(self.rotation_z) + '\n'
- string += ' rotation_x: ' + str(self.rotation_x) + '\n'
- string += ' rotation_y: ' + str(self.rotation_y) + '\n'
- string += ' translation_xy: ' + str(self.translation_xy) + '\n'
- string += ' translation_z: ' + str(self.translation_z) + '\n'
- string += ' scale_xy: ' + str(self.scale_xy) + '\n'
- string += ' scale_z: ' + str(self.scale_z) + '\n'
- string += ' thickness: ' + str(self.thickness) + '\n'
- string += ' flip_h: ' + str(self.flip_h) + '\n'
- string += ' flip_v: ' + str(self.flip_v) + '\n'
- string += ' filling_mode: ' + str(self.filling_mode) + '\n'
- string += ' padding_value: ' + str(self.padding_value) + '\n'
- string += ' bone_edges: ' + str(self.bone_edges) + '\n'
- return string
- def _create_comp_affine(self, affine, offset):
- """Returns a composite matrix from the affine and offset."""
- if affine.shape != (2, 2) and affine.shape != (3, 3):
- raise ValueError('Affine must be a 2x2 or 3x3 matrix.')
- if affine.shape[0] != offset.shape[0]:
- raise ValueError('Affine and offset have incompatible dimensions.')
- affine = np.concatenate((affine, offset.reshape(-1, 1)), axis=1)
- if affine.shape == (3, 4):
- affine = np.concatenate((affine, np.array([[0, 0, 0, 1]])), axis=0)
- else:
- affine = np.concatenate((affine, np.array([[0, 0, 1]])), axis=0)
- return affine
- def generate(self, src, mask, n, affine, return_orig=True, verbose=0):
- """Generate `n` augmentations of the `src` and `mask`.
- ### Args
- `src` (three dimensional ndarray): Image matrix.
- `mask` (three dimensional ndarray): Binary mask matrix.
- `n` (int): Number of augmentations.
- `return_orig` (bool): Determines whether original src and mask are
- returned. Defaults to True.
- `verbose` (int): Determines if augmentation parameters are printed
- to screen. Defaults to 0.
- ### Returns:
- Tuple of two matrices, first matrix contains stacked `src`
- augmentations while the second contains the stacked `mask`
- augmentations.
- """
- # Check if src and mask are compatible
- if src.ndim != mask.ndim:
- raise ValueError('Src and mask dimensions don\'t match.')
- # If the passed src is 2 dimensional, call the generate2d method instead
- if src.ndim == 2:
- return self._generate2d(src, mask, n, return_orig, verbose)
- # (x1,y1,z1) = orx.aff2axcodes(affine)
- # ornt = orx.axcodes2ornt((x1,y1,z1))
- # refOrnt = orx.axcodes2ornt(('R','A','S'))
- # newOrnt1 = orx.ornt_transform(ornt,refOrnt)
- # (x2,y2,z2) = orx.aff2axcodes(affine)
- # ornt = orx.axcodes2ornt((x2,y2,z2))
- # refOrnt = orx.axcodes2ornt(('R','A','S'))
- # newOrnt2 = orx.ornt_transform(ornt,refOrnt)
- # src = orx.apply_orientation(src,newOrnt1)
- # mask = orx.apply_orientation(mask,newOrnt1)
- # src = np.fliplr(np.rot90(src,1))
- # mask = np.fliplr(np.rot90(mask,1))
- # Get parameters from self
- rotation_z = self.rotation_z
- rotation_x = self.rotation_x
- rotation_y = self.rotation_y
- translation_xy = self.translation_xy
- translation_z = self.translation_z
- scale_xy = self.scale_xy
- scale_z = self.scale_z
- thickness=self.thickness
- flip_h = self.flip_h
- flip_v = self.flip_v
- filling_mode = self.filling_mode
- padding_value = self.padding_value
- bone_edges = self.bone_edges
- mask_PV = self.mask_PV
- # Add original images and mask to output stack if warranted
- if return_orig:
- output_img_stack = src.copy()
- output_mask_stack = mask.copy().astype(float)
- if verbose:
- print(self)
- for iteration in range(n):
- # Create base affine
- affine_w_offset = np.identity(4)
- #----------------------------------------------------#
- # Rotate around z-axis (rotation in the axial plane) #
- #----------------------------------------------------#
- if rotation_z:
- # Choose an angle at random and convert it to radians
- rot_z_angle = np.random.randint(-rotation_z, rotation_z + 1)
- rot_z_angle = np.deg2rad(rot_z_angle)
- affine = \
- np.array([[np.cos(rot_z_angle), -np.sin(rot_z_angle), 0],
- [np.sin(rot_z_angle), np.cos(rot_z_angle), 0],
- [0, 0, 1]])
- # Calculate center offset
- center_input = 0.5 * np.array(src.shape)
- center_output = center_input.dot(affine)
- offset = center_input - center_output
- # Calculate new composite affine
- new_affine_w_offset = self._create_comp_affine(affine, offset)
- affine_w_offset = np.dot(affine_w_offset, new_affine_w_offset)
- #-------------------------------------------------------#
- # Rotate around x-axis (rotation in the sagittal plane) #
- #-------------------------------------------------------#
- if rotation_x:
- # Choose an angle at random and convert it to radians
- rot_x_angle = np.random.randint(-rotation_x, rotation_x + 1)
- rot_x_angle = np.deg2rad(rot_x_angle)
- affine = \
- np.array([[1, 0, 0],
- [0, np.cos(rot_x_angle), -np.sin(rot_x_angle)],
- [0, np.sin(rot_x_angle), np.cos(rot_x_angle)]])
- # Calculate center offset
- center_input = 0.5 * np.array(src.shape)
- center_output = center_input.dot(affine)
- offset = center_input - center_output
- # Calculate new composite affine
- new_affine_w_offset = self._create_comp_affine(affine, offset)
- affine_w_offset = np.dot(affine_w_offset, new_affine_w_offset)
- #------------------------------------------------------#
- # Rotate around y-axis (rotation in the coronal plane) #
- #------------------------------------------------------#
- if rotation_y:
- # Choose an angle at random and convert it to radians
- rot_y_angle = np.random.randint(-rotation_y, rotation_y + 1)
- rot_y_angle = np.deg2rad(rot_y_angle)
- affine = \
- np.array([[np.cos(rot_y_angle), 0, np.sin(rot_y_angle)],
- [0, 1, 0],
- [-np.sin(rot_y_angle), 0, np.cos(rot_y_angle)]])
- # Calculate center offset
- center_input = 0.5 * np.array(src.shape)
- center_output = center_input.dot(affine)
- offset = center_input - center_output
- # Calculate new composite affine
- new_affine_w_offset = self._create_comp_affine(affine, offset)
- affine_w_offset = np.dot(affine_w_offset, new_affine_w_offset)
- #----------------------------------------#
- # Scale along x, y axes, possibly z-axis #
- #----------------------------------------#
- if scale_xy:
- scale_factor = 1 + np.random.uniform(-scale_xy, scale_xy)
- affine = np.identity(3) * scale_factor
- # Eliminate scaling in the z-axis (may be preferable with
- # thick slices)
- if not scale_z:
- affine[2, 2] = 1
- else:
- affine[2, 2] = scale_z
- # Calculate center offset
- center_input = 0.5 * np.array(src.shape)
- center_output = center_input.dot(affine)
- offset = center_input - center_output
- # Calculate new composite affine
- new_affine_w_offset = self._create_comp_affine(affine, offset)
- affine_w_offset = np.dot(affine_w_offset, new_affine_w_offset)
- #------------------------------------------------#
- # Translate in the x and y axes, possibly z-axis #
- #------------------------------------------------#
- if translation_xy:
- x_offset = np.random.randint(-translation_xy,
- translation_xy + 1)
- y_offset = np.random.randint(-translation_xy,
- translation_xy + 1)
- else:
- x_offset, y_offset = 0, 0
- if translation_z:
- z_offset = np.random.randint(-translation_z, translation_z + 1)
- else:
- z_offset = 0
- offset = np.array([x_offset, y_offset, z_offset])
- # Add offset to existing affine offset
- affine_w_offset[:3, 3] = affine_w_offset[:3, 3] + offset
- #----------------------------------------#
- # Apply composite affine to src and mask #
- #----------------------------------------#
- output_img = ndimage.affine_transform(
- input=src,
- matrix=affine_w_offset[:3, :3].T,
- offset=affine_w_offset[:3, 3].ravel(),
- mode=filling_mode,
- cval=padding_value)
- output_mask = ndimage.affine_transform(
- input=mask,
- matrix=affine_w_offset[:3, :3].T,
- offset=affine_w_offset[:3, 3].ravel(),
- mode=filling_mode, cval=mask_PV )
- #---------------------#
- # Make thicker slices #
- #---------------------#
- # if len(thickness) > 1 or thickness[0] != 1:
- # th = np.random.choice(thickness)
- #
- # temp_output_img = np.zeros(output_img.shape[:2] + (output_img.shape[2] // th,))
- # temp_output_mask = np.zeros(output_img.shape[:2] + (output_img.shape[2] // th,))
- #
- # for i, j in enumerate(range(0, output_img.shape[2] // th * th, th)):
- # temp_output_img[:, :, i] = np.mean(output_img[:, :, j:j+th], axis=2)
- # temp_output_mask[:, :, i] = np.mean(output_mask[:, :, j:j+th], axis=2)
- #
- # output_img = temp_output_img
- # output_mask = temp_output_mask
- # else:
- # th = 1
- #---------------------------------------#
- # Flip image horizontally or vertically #
- #---------------------------------------#
- if flip_v:
- flip_v_bool = np.random.choice((True, False))
- if flip_v_bool:
- output_img = np.fliplr(output_img)
- output_mask = np.fliplr(output_mask)
- if flip_h:
- flip_h_bool = np.random.choice((True, False))
- if flip_h_bool:
- output_img = np.flipud(output_img)
- output_mask = np.flipud(output_mask)
- # Print augmentation parameters to screen
- # if verbose:
- # print(f'Aug {iteration+1}:')
- # if rotation_z:
- # print(' Rotation about z axis:',
- # f'{np.rad2deg(rot_z_angle):3} degrees')
- # if rotation_x:
- # print(' Rotation about x-axis:',
- # f'{np.rad2deg(rot_x_angle):3} degrees')
- # if rotation_y:
- # print(' Rotation about y-axis:',
- # f'{np.rad2deg(rot_y_angle):3} degrees')
- # if scale_xy:
- # print(f' Scaling in the x and y axes: {scale_factor:3}')
- # if scale_z:
- # print(f' Scaling in the z-axis: {scale_factor:3}')
- # if translation_xy:
- # print(f' X offset: {x_offset:3}')
- # if translation_z:
- # print(f' Y offset: {y_offset:3}')
- # if th != 1:
- # print(f' Slice thickness: {th}')
- # if flip_h:
- # print(f' Horizontal flip: {flip_h_bool}')
- # if flip_v:
- # print(f' Vertical flip: {flip_v_bool}')
- # print('')
- # Adjust for mask pixels that overlie bone
- if bone_edges:
- output_mask[np.logical_and(output_img > 100, output_mask < 0.85)] = 0
- # Make output_mask a binary mask again
- output_mask = (output_mask > 0.65).astype(float)
- # Concatenate augmentation with remainder of stack
- try:
- output_img_stack = np.concatenate(
- [output_img_stack, output_img], axis=2)
- output_mask_stack = np.concatenate(
- [output_mask_stack, output_mask], axis=2)
- except NameError:
- output_img_stack = output_img
- output_mask_stack = output_mask
- # if verbose:
- # print(f'Generated {n} augmentations(s) with a resulting' +
- # f' stack size of {output_img_stack.shape}.')
- return output_img_stack, output_mask_stack
- def _generate2d(self, src, mask, n, return_orig=True, verbose=0):
- """Generate `n` augmentations of a 2d `src` and `mask`.
- ### Args
- `src` (two dimensional ndarray): Image matrix.
- `mask` (two dimensional ndarray): Binary mask matrix.
- `n` (int): Number of augmentations.
- `return_orig` (bool): Determines whether original src and mask are
- returned. Defaults to True.
- `verbose` (int): Determines if augmentation parameters are printed
- to screen. Defaults to 0.
- ### Returns:
- Tuple of two matrices, first matrix contains stacked `src`
- augmentations while the second contains the stacked `mask`
- augmentations.
- """
- # Get parameters from self
- rotation_z = self.rotation_z
- rotation_x = self.rotation_x
- rotation_y = self.rotation_y
- translation_xy = self.translation_xy
- translation_z = self.translation_z
- scale_xy = self.scale_xy
- scale_z = self.scale_z
- flip_h = self.flip_h
- flip_v = self.flip_v
- filling_mode = self.filling_mode
- padding_value = self.padding_value
- mask_PV = self.mask_PV
- # Add original images and mask to output stack if warranted
- if return_orig:
- output_img_stack = src.reshape(src.shape + (1,)).copy()
- output_mask_stack = mask.reshape(mask.shape + (1,)).copy().astype(float)
- if verbose:
- print(self)
- for i in range(n):
- # Create base affine
- affine_w_offset = np.identity(3)
- #----------------------------------------------------#
- # Rotate around z-axis (rotation in the axial plane) #
- #----------------------------------------------------#
- if rotation_z:
- # Choose an angle at random and convert it to radians
- rot_z_angle = np.random.randint(-rotation_z, rotation_z + 1)
- rot_z_angle = np.deg2rad(rot_z_angle)
- affine = \
- np.array([[np.cos(rot_z_angle), -np.sin(rot_z_angle)],
- [np.sin(rot_z_angle), np.cos(rot_z_angle)]])
- # Calculate center offset
- center_input = 0.5 * np.array(src.shape)
- center_output = center_input.dot(affine)
- offset = center_input - center_output
- # Calculate new composite affine
- new_affine_w_offset = self._create_comp_affine(affine, offset)
- affine_w_offset = np.dot(affine_w_offset, new_affine_w_offset)
- #-----------------------#
- # Scale along x, y axes #
- #-----------------------#
- if scale_xy:
- scale_factor = 1 + np.random.uniform(-scale_xy, scale_xy)
- affine = np.identity(2) * scale_factor
- # Calculate center offset
- center_input = 0.5 * np.array(src.shape)
- center_output = center_input.dot(affine)
- offset = center_input - center_output
- # Calculate new composite affine
- new_affine_w_offset = self._create_comp_affine(affine, offset)
- affine_w_offset = np.dot(affine_w_offset, new_affine_w_offset)
- #------------------------------------------------#
- # Translate in the x and y axes, possibly z-axis #
- #------------------------------------------------#
- if translation_xy:
- x_offset = np.random.randint(-translation_xy,
- translation_xy + 1)
- y_offset = np.random.randint(-translation_xy,
- translation_xy + 1)
- else:
- x_offset, y_offset = 0, 0
- offset = np.array([x_offset, y_offset])
- # Add offset to existing affine offset
- affine_w_offset[:2, 2] = affine_w_offset[:2, 2] + offset
- #----------------------------------------#
- # Apply composite affine to src and mask #
- #----------------------------------------#
- output_img = ndimage.affine_transform(
- input=src,
- matrix=affine_w_offset[:2, :2].T,
- offset=affine_w_offset[:2, 2].ravel(),
- mode=filling_mode,
- cval=padding_value)
- output_mask = ndimage.affine_transform(
- input=mask,
- matrix=affine_w_offset[:2, :2].T,
- offset=affine_w_offset[:2, 2].ravel(),
- mode=filling_mode)
- #---------------------------------------#
- # Flip image horizontally or vertically #
- #---------------------------------------#
- if flip_h:
- flip_h_bool = np.random.choice((True, False))
- if flip_h_bool:
- output_img = np.fliplr(output_img)
- output_mask = np.fliplr(output_mask)
- if flip_v:
- flip_v_bool = np.random.choice((True, False))
- if flip_v_bool:
- output_img = np.flipud(output_img)
- output_mask = np.flipud(output_mask)
- # Print augmentation parameters to screen
- # if verbose:
- # print(f'Aug {i+1}:')
- # if rotation_z:
- # print(' Rotation about z axis:',
- # f'{np.rad2deg(rot_z_angle):3} degrees')
- # if rotation_x:
- # print(' Rotation about x-axis: 2d data, not performed')
- # if rotation_y:
- # print(' Rotation about y-axis: 2d data, not performed')
- # if scale_xy:
- # print(f' Scaling in the x and y axes: {scale_factor:3}')
- # if scale_z:
- # print(' Scaling in the z-axis: 2d data, not performed.')
- # if translation_xy:
- # print(f' X offset: {x_offset:3}')
- # if translation_z:
- # print(' Y offset: 2d data, not performed.')
- # if flip_h:
- # print(f' Horizontal flip: {flip_h_bool}')
- # if flip_v:
- # print(f' Vertical flip: {flip_v_bool}')
- # print('')
- # Make output_mask a binary mask again
- output_mask = (output_mask > 0.5).astype(int)
- # Convert output_img and output_mask to 3d matrices for concatenation
- output_img = output_img.reshape(output_img.shape + (1,))
- output_mask = output_mask.reshape(output_mask.shape + (1,))
- # Concatenate augmentation with remainder of stack
- try:
- output_img_stack = np.concatenate(
- [output_img_stack, output_img], axis=2)
- output_mask_stack = np.concatenate(
- [output_mask_stack, output_mask], axis=2)
- except NameError:
- output_img_stack = output_img
- output_mask_stack = (output_mask > 0.5).astype(int)
- # if verbose:
- # print(f'Generated {n} 2d augmentations(s) with a resulting' +
- # f' stack size of {output_img_stack.shape}.')
- return output_img_stack, output_mask_stack
auggen.py at commit d687482, no license · at the source
Overview
- School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai, China
- Institute of Diagnostic and Interventional Radiology, Shanghai Sixth People’s Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China
- Department of Radiology, Liyang Hospital of Chinese Medicine, Changzhou, Jiangsu, China
Abstract
Background: To determine if a small number of annotated scans can enable clinically meaningful adaptation of a pre-trained convolutional neural network across different centers. Using non-contrast head CT brain extraction as the test case, we also sought to identify the factors contributing to performance drops when applying models to new sites.
Materials and methods: Non-contrast head CT scans were used in this study, from two separate institutions (Center A: 595 scans; Center B: 486 scans). A public pre-trained U-Net model for brain tissue extraction was first evaluated on Center A dataset as a baseline. The model was then fine-tuned incrementally using stratified subsets of annotated scans (5, 15, 25, and 75). The pre-trained model was subsequently transferred to Center B dataset and adapted through the same incremental fine-tuning strategy. The primary metric of interest was the three-dimensional Dice similarity coefficient (DSC), with a pre-defined minimal important difference (MID) set at 0.01. To identify predictors of baseline DSC and fine-tuning gain driving cross-center performance degradation, variance inflation factor (VIF) screening and logit Gaussian linear models (GLMs) were applied, with SHapley Additive exPlanations (SHAP) analysis providing qualitative feature importance rankings.
Results: At center A, the pre-trained model achieved a DSC of 0.9634 ± 0.0151. Fine-tuning with five annotated scans raised DSC to 0.9784 ± 0.0070 (p < 0.0001), exceeding the MID by 1.5-fold and reducing clinically unacceptable segmentations from 8.7 to 0.8%. A comparable gain was observed at Center B (p < 0.0001). Further increases to 15, 25, and 75 samples offered only marginal improvement below the MID threshold. Regression and SHAP analyses identified slice thickness as the dominant driver of both baseline performance degradation and fine-tuning improvement, accounting for 80–90% of predictive weight at both centers.
Conclusion: For NCCT brain extraction assessed across two centers, fine-tuning a publicly pre-trained CNN using five local annotated scans can generate clinically valuable cross-center segmentation gains; slice thickness heterogeneity appears to be the dominant source of inter-center domain shift.
Reproduced under the paper's license (CC BY), from the paper cited above.
Repository
Its files are read in the Code ↔ Paper reader above.
aqqush/CT_BET
d6874828abf1e76fa398cbb86a36137dde120f50, 7 October 2020Availability: 1 check, the latest on 26 September 2026: the link answers
- 26 September 2026: the link answers
6 files
- auggen.py, Python, 562 lines
- deepModels.py, Python, 189 lines
- load3Ddata.py, Python, 58 lines
- model_CT_SS.py, Python, 645 lines
- unet_CT_SS.py, Python, 125 lines
- README.md, Text, 33 lines
Tracing map
Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.
What the map holds:
- 1 repository of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
- 5 scripts, each with its path and the digest of its content;
- no match between paragraphs and code yet;
- neither the text of the paper nor the code itself.
Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.
Data
No dataset and no data link were found in the paper.
Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
Reproduced under the paper's license (CC BY), from the paper cited above.
Versions
The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.
Version 1, 27 September 2026: the first record
Recorded: type, language, journal, volume, pages, dates, 4 authors, 5 keywords, 4 MeSH terms, 23 references.
Cite
This paper
Sun, Z., Ye, L., Xu, Y., & Li, Y. (2026). Data-centric strategies enable cross-center generalization of convolutional neural networks for CT brain extraction. Frontiers in neurology, 17, 1895626. https://
BibTeX
@article{sun2026data,
author = {Sun, Zheng and Ye, Lifang and Xu, Yao and Li, Yuehua},
title = {{Data-centric strategies enable cross-center generalization of convolutional neural networks for CT brain extraction}},
journal = {Frontiers in neurology},
year = {2026},
month = aug,
volume = {17},
pages = {1895626},
publisher = {Frontiers Media SA},
issn = {1664-2295},
doi = {10.3389/
url = {https://
pmid = {42609258},
pmcid = {PMC13477930}
}
RIS
TY - JOUR
AU - Sun, Zheng
AU - Ye, Lifang
AU - Xu, Yao
AU - Li, Yuehua
TI - Data-centric strategies enable cross-center generalization of convolutional neural networks for CT brain extraction
T2 - Frontiers in neurology
J2 - Front Neurol
PY - 2026
DA - 2026/
VL - 17
SP - 1895626
SN - 1664-2295
PB - Frontiers Media SA
DO - 10.3389/
UR - https://
LA - en
ER -
CSL-JSON
{
"id": "10.3389/
"type": "article-journal",
"title": "Data-centric strategies enable cross-center generalization of convolutional neural networks for CT brain extraction",
"container-title": "Frontiers in neurology",
"author": [
{
"family": "Sun",
"given": "Zheng"
},
{
"family": "Ye",
"given": "Lifang"
},
{
"family": "Xu",
"given": "Yao"
},
{
"family": "Li",
"given": "Yuehua"
}
],
"container-title-short":
"volume": "17",
"page": "1895626",
"DOI": "10.3389/
"PMID": "42609258",
"PMCID": "PMC13477930",
"ISSN": "1664-2295",
"publisher": "Frontiers Media SA",
"URL": "https://
"language": "en",
"issued": {
"date-parts": [
[
2026,
8,
3
]
]
}
}
The tracing map gets a citation of its own once an author has validated it and it has a DOI.
Similar papers
The papers with a page that share the most with this one: the tools found in their code, their categories, datasets, cited references and authors, the rarest counting most.
- [1] doi:10.1016/j.media.2023.102789 [code]
- SynthSeg: Segmentation of brain MRI scans of any contrast and resolution without retraining.Journal: Medical image analysisIn common: SimpleITK, Keras, NiBabel, 3 other tools, 1 reference
- [2] doi:10.1016/j.media.2023.103058 [code]
- Acquisition-invariant brain MRI segmentation with informative uncertainties.Journal: Medical image analysisIn common: SimpleITK, Keras, NiBabel, 2 other tools, 1 reference
- [3] doi:10.3390/diagnostics13243604 [code]
- Deep Learning-Based Extraction of Biomarkers for the Prediction of the Functional Outcome of Ischemic Stroke Patients.Journal: Diagnostics (Basel, Switzerland)In common: Keras, NiBabel, scikit-learn, 2 other tools, other, 1 reference
- [4] doi:10.1002/hbm.70303 [code]
- A Contrast-Agnostic Method for Ultra-High Resolution Claustrum Segmentation.Journal: Human brain mappingIn common: SimpleITK, Keras, NiBabel, 2 other tools, 1 reference
- [5] doi:10.1002/hbm.26674 [code]
- Comprehensive analysis of synthetic learning applied to neonatal brain MRI segmentation.Journal: Human brain mappingIn common: SimpleITK, NiBabel, scikit-learn, 2 other tools, 1 reference
- [6] doi:10.3389/fmed.2026.1875760 [code]
- Adaptive multi-stage domain unlearning for white-matter lesion segmentation.Journal: Frontiers in medicineIn common: SimpleITK, NiBabel, scikit-learn, 2 other tools, 1 reference
- [7] doi:10.1038/s41531-025-00982-5 [code]
- Baseline [&
lt;sup& gt;18& lt;/ sup& gt;F]FP-CIT PET-based deep learning prediction of levodopa-induced dyskinesia in Parkinson's disease. Journal: NPJ Parkinson's diseaseIn common: SimpleITK, Keras, scikit-learn, 2 other tools, 1 reference - [8] doi:10.3389/frai.2026.1771088 [code]
- Few-shot deployment of pretrained MRI transformers in brain imaging tasks.Journal: Frontiers in artificial intelligenceIn common: SimpleITK, NiBabel, scikit-learn, 2 other tools, 1 reference
- [9] doi:10.1038/s41597-025-04926-9 [code]
- A dataset of synthetic, maturation-informed magnetic resonance images of the human fetal brain.Journal: Scientific dataIn common: SimpleITK, NiBabel, scikit-learn, 2 other tools, 1 reference
- [10] doi:10.1162/imag.a.960 [code]
- GOUHFI: A novel contrast- and resolution-agnostic segmentation tool for ultra-high-field MRI.Journal: Imaging neuroscience (Cambridge, Mass.)In common: SimpleITK, NiBabel, scikit-learn, 2 other tools, 1 reference
Contribute
The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.
Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.
Claim this paper
Correct its record
Say what each link of this record is, remove the ones that are not the paper's, add the ones that are missing. The correction becomes a new version of the record, in its Versions section.
Validate its tracing map
You validate the map as this page shows it: 1 repository of the authors' code, each at its verified commit and with its license, 5 scripts, and 0 matches between paragraphs and code (see the Code and Map sections). It then receives a DOI on Zenodo, with you (your ORCID iD) and OSCR as its creators; the code itself is not deposited.
The map's fingerprint: sha256:6e226e153431a1a4…
Add the badge to the code's README
If you are an author or a maintainer of this paper's code, you can add this badge. It says what OSCR knows of the paper (its code, its validated tracing map) and links the code to this page. Paste one of these snippets under the README's title. Keep the two marks around it, so that the badge can be replaced in place and is never added twice. OSCR changes nothing in your repository.
Markdown
<!-- oscr-badge:start -->
[](https://openscicode.org/badge/doi/10.3389/fneur.2026.1895626)
<!-- oscr-badge:end -->reStructuredText
.. oscr-badge:start
.. image:: https://openscicode.org/badge/doi/10.3389/fneur.2026.1895626.svg
:target: https://openscicode.org/badge/doi/10.3389/fneur.2026.1895626
:alt: OSCR
.. oscr-badge:endHTML
<!-- oscr-badge:start -->
<a href="https://openscicode.org/badge/doi/10.3389/fneur.2026.1895626"><img src="https://openscicode.org/badge/doi/10.3389/fneur.2026.1895626.svg" alt="OSCR"></a>
<!-- oscr-badge:end -->On GitHub, open the README, choose Edit (the pencil), paste the snippet under the title, then "Commit changes".
If you choose to, you can have the badge kept up to date for you, with oscr badge --write from the command line or with the GitHub Action in your repository's workflows. Every other form and option (a card, AsciiDoc, LaTeX, shields.io) is on the badge page.
Request its removal
To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).
Discussion, reproductions, activity
Discussion: code errors and mismatches between the code and the paper about this paper and its code, from readers and its authors, beside the paper and its tracing map.
Report a mismatch between the code and the paper · Report a code error (signed in)
Reproductions: reports from readers who ran the authors' code and did not get the paper's result, what they ran, with which environment, commit and data.
Report a reproduction failure (signed in)
Versions of its code: the releases of its code tied to a version of this paper (the preprint, the accepted manuscript, the version of record…), each with the tracing map versioned with it.
Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.
