SYNcro

SYNcro spatially normalizes clinical medical images to match the shape of the MNI152 template. This allows images from different people to be analyzed in a common space. Spatial normalization methods that work reliably with high-resolution T1-weighted scans from healthy adults can fail in clinical populations. Diverse imaging modalities, 2D sequences with thick slices, nuisance features such as scalp fat, and abnormalities such as lesions can all disrupt automated methods. Previous solutions have included cost function masking (ignoring the injury), lesion healing, using admission modalities that do not yet reveal lesions, and modality-specific templates (for example, CT and FLAIR).

In contrast, SYNcro uses SynthSR to generate a uniform, high-resolution T1-weighted image from almost any brain scan. SynthSR tends to fill in lesions in the synthesized image while preserving the geometry of the input. This provides a good target for estimating how the scan must be deformed to match the template.

SYNcro therefore fills an important niche by providing fast, reliable, and robust alignment of clinical images. It is a minimal wrapper around proven tools developed by other teams:

• SynthSR converts the input into a high-resolution T1-weighted scan.
• MindGrab extracts the brain, preventing scalp fat from influencing registration.
• Greedy performs nonlinear registration, bending the scan to match the template.
• niimath performs supporting image and mask calculations.

SYNcro itself is a minimal Python script that coordinates these standalone executables.

Installation

SYNcro is still experimental. You can try the web version on any operating system. The standalone version is currently available only for Apple Silicon (Arm64) macOS computers.

• Download and install the latest notarized SYNcro macOS package. Once installed, run SYNcro from the command line.

Tutorials

This repository provides sample datasets illustrating several common uses of the tool. Each example includes a brain injury and assumes that a lesion has been drawn on one of the scans. However, because SynthSR tends to fill in lesions in its synthesized image, a lesion map is not required for normalization and could instead be drawn after normalization. You can also run the same workflow for participants without lesions—for example, to measure false alarms when detecting radiological abnormalities. In either case, simply omit --lesion MASK from the SYNcro command.

In the images below, red marks the lesion mask. The mask is stored as a separate image; it is shown as an overlay here to make its location easy to see.

The command grammar is:

SYNcro ANATOMICAL [--lesion MASK ...] [SCAN [--lesion MASK ...]]... [OPTIONS]

The first scan supplies the detailed anatomy used to estimate nonlinear normalization. Every later scan is rigidly aligned to that anatomical scan. Each --lesion (or -l) belongs to the scan immediately before it, so the relationship is explicit even when several scans and masks are supplied.

T1w, T2w, lesion, and 4D fMRI

This session demonstrates the general multi-image workflow. The high-resolution T1w scan supplies the anatomy. The lesion was drawn on the T2w scan, so --lesion follows the T2w filename. The 60-volume BOLD series is a third scan with a much coarser grid. SYNcro rigidly aligns the T2w and the temporal mean of the BOLD series to the T1w, then applies one composed transform to every BOLD volume. Note here we are applying the spatial normalization to the raw BOLD image, in practice you would likely undistort your fMRI data using a fieldmap or an image with reversed phase-encoding polarity. The BOLD data are assumed to have already been motion corrected; SYNcro does not realign individual time points.

Complete session at 2 mm

Input: anatomical T1w	Input: T2w with lesion	Input: mean BOLD

Output: normalized T1w with lesion	Output: normalized T2w with lesion	Output: mean normalized BOLD	Output: normalized synthetic T1w

SYNcro \
  tutorial/sub-M2295_ses-239_acq-tfl3p2_run-3_T1w.nii.gz \
  tutorial/sub-M2295_ses-239_acq-spc3p2_run-4_T2w.nii.gz \
  --lesion tutorial/sub-M2295_ses-239_acq-spc3p2_run-4_T2w_desc-lesion_mask.nii.gz \
  tutorial/sub-M2295_ses-239_task-naming40_acq-epfid2_dir-AP_run-2_bold.nii.gz \
  --target tutorial/MNI152_T1_2mm_brain_mask.nii.gz \
  --directory outputs/m2295

--target chooses the voxel grid for every normalized output. Here it requests a 2 mm MNI grid (91×109×91), which keeps the 4D result much smaller than a 1 mm series. It does not replace the bundled 1 mm MNI brain used to estimate the anatomical deformation, and its voxel values are not used. SYNcro preserves the 60 time points and their 10-second spacing. It writes NIfTI images only; accompanying JSON sidecars are not transformed or copied.

Anatomical scans without the BOLD series

Use this shorter command when only the structural images and lesion are needed. The lesion follows the T2w because it was drawn in T2w space. With no --target, outputs use SYNcro's default 1 mm MNI grid.

SYNcro \
  tutorial/sub-M2295_ses-239_acq-tfl3p2_run-3_T1w.nii.gz \
  tutorial/sub-M2295_ses-239_acq-spc3p2_run-4_T2w.nii.gz \
  --lesion tutorial/sub-M2295_ses-239_acq-spc3p2_run-4_T2w_desc-lesion_mask.nii.gz \
  --directory outputs/m2295-structural-1mm

This produces normalized T1w, T2w, and lesion images, as well as the normalized synthetic T1w and brain-extracted original T1w.

T1w and 4D BOLD only

A lesion map is optional. This command rigidly aligns the temporal mean of the BOLD series to the T1w and applies that one alignment to all 60 volumes:

SYNcro \
  tutorial/sub-M2295_ses-239_acq-tfl3p2_run-3_T1w.nii.gz \
  tutorial/sub-M2295_ses-239_task-naming40_acq-epfid2_dir-AP_run-2_bold.nii.gz \
  --target tutorial/MNI152_T1_2mm_brain_mask.nii.gz \
  --directory outputs/m2295-bold-2mm

The normalized series is wsub-M2295_ses-239_task-naming40_acq-epfid2_dir-AP_run-2_bold.nii.gz. It remains a 60-volume 4D image. Do not place the BOLD image first: the first input must be a 3D anatomical scan suitable for SynthSR. If a lesion were drawn on a BOLD-derived image, its 3D mask would follow the BOLD filename and would need to match the BOLD spatial grid exactly.

Choosing output resolution and field of view

Omit --target for the bundled 1 mm MNI grid. This gives the finest standard output, but a 4D series contains about eight times as many spatial voxels as the corresponding 2 mm output:

SYNcro \
  tutorial/sub-M2295_ses-239_acq-tfl3p2_run-3_T1w.nii.gz \
  tutorial/sub-M2295_ses-239_task-naming40_acq-epfid2_dir-AP_run-2_bold.nii.gz \
  --directory outputs/m2295-bold-1mm

Use the supplied target for smaller 2 mm outputs:

SYNcro \
  tutorial/sub-M2295_ses-239_acq-tfl3p2_run-3_T1w.nii.gz \
  tutorial/sub-M2295_ses-239_task-naming40_acq-epfid2_dir-AP_run-2_bold.nii.gz \
  --target tutorial/MNI152_T1_2mm_brain_mask.nii.gz \
  --directory outputs/m2295-bold-2mm

Any 3D NIfTI can serve as the reslicing target, including a cropped MNI-space image with a smaller field of view:

SYNcro \
  tutorial/sub-M2295_ses-239_acq-tfl3p2_run-3_T1w.nii.gz \
  tutorial/sub-M2295_ses-239_acq-spc3p2_run-4_T2w.nii.gz \
  --lesion tutorial/sub-M2295_ses-239_acq-spc3p2_run-4_T2w_desc-lesion_mask.nii.gz \
  --target /path/to/cropped_MNI_grid.nii.gz \
  --directory outputs/m2295-cropped

SYNcro uses the target's dimensions, voxel size, orientation, and origin; its voxel values are ignored. A custom target should already describe the intended MNI coordinate system. The target changes every normalized output from that command, not just the 4D series. Use a new output directory for each target grid so results cannot be confused.

TRACE and T1w

This acute scan shows a lesion on the low-resolution diffusion TRACE image before it is visible on the high-resolution T1-weighted image. The lesion was therefore drawn on the TRACE image. SYNcro first aligns the TRACE image and its lesion map to the T1-weighted image, then normalizes all three images to the MNI152 template.

Input: anatomical T1w	Input: TRACE with lesion	Output: normalized T1w with lesion	Output: normalized TRACE with lesion	Output: normalized synthetic T1w

SYNcro \
  tutorial/sub-101_T1w.nii.gz \
  tutorial/sub-101_rec-TRACE_dwi.nii.gz \
  --lesion tutorial/sub-101_space-TRACE_desc-lesion_mask.nii.gz \
  --directory outputs/traceAndT1

TRACE only

This example provides only the low-resolution diffusion TRACE scan and its lesion map. SYNcro creates a synthetic T1-weighted image from the TRACE scan and uses it to normalize both inputs to the MNI152 template.

Input: TRACE with lesion	Output: normalized TRACE with lesion	Output: normalized synthetic T1w

SYNcro \
  tutorial/sub-101_rec-TRACE_dwi.nii.gz \
  --lesion tutorial/sub-101_space-TRACE_desc-lesion_mask.nii.gz \
  --directory outputs/traceOnly

T1w only

This example provides a T1-weighted scan and a lesion map drawn in the same image space. SYNcro creates a more uniform synthetic T1-weighted image, uses it to estimate the normalization, and applies that normalization to both inputs.

Input: T1w with lesion	Output: normalized T1w with lesion	Output: normalized synthetic T1w

SYNcro \
  tutorial/T1w.nii.gz \
  --lesion tutorial/T1w-lesion.nii.gz \
  --directory outputs/t1Only

CT only

This example provides a CT scan and a lesion map drawn in the same image space. The --ct option tells SynthSR that the input intensities are Hounsfield units, allowing it to process the CT appropriately.

Input: CT with lesion	Output: normalized CT with lesion	Output: normalized synthetic T1w

SYNcro \
  tutorial/CT.nii.gz \
  --lesion tutorial/CT-lesion.nii.gz \
  --directory outputs/ctOnly \
  --ct --keep-synth

This example also demonstrates the --keep-synth option. It retains the T1-weighted scan created by SynthSR before brain extraction and spatial normalization. This image includes scalp signal and can be inspected or analyzed with tools designed for whole-head T1-weighted scans, such as brain-age methods. The command therefore generates five outputs:

• wCT.nii.gz: normalized input CT image
• wbCT.nii.gz: normalized, brain-extracted input CT image
• wCT-lesion.nii.gz: normalized lesion image
• t1CT.nii.gz: synthesized T1-weighted scan before spatial normalization
• wbt1CT.nii.gz: normalized, brain-extracted synthetic T1-weighted scan

Input and reslicing rules

• The anatomical input and lesion masks must be 3D. Other scans may be 3D or 4D.
• A lesion must have exactly the same spatial dimensions and geometry as the scan that precedes its --lesion option. Multiple lesions may be paired with one scan by repeating the option.
• Additional scans need not match the anatomical grid. SYNcro estimates a separate rigid-body alignment for each one before composing it with the anatomical-to-MNI deformation.
• Continuous scans use linear interpolation. Lesions use nearest-neighbor interpolation and are written as binary uint8 images.
• --ct tells SynthSR that the anatomical input is CT and applies CT air-background handling to every scalar scan in the command, but never to lesion masks.
• Without --target, normalized outputs use the bundled 1 mm MNI grid. A 3D NIfTI supplied with --target changes the output dimensions, resolution, orientation, and field of view for every result.

References

• SynthSR
• MindGrab
• Greedy registration
• niimath
• The sub-101 images come from the Stroke Outcome Optimization Project (SOOP).
• The T1 and T2 scans come from the clinical toolbox.
• The CT image is from the acute ischemic stroke dataset.
• The MNI152 template is from the McConnell Brain Imaging Centre.