Brain Intensity AbNormality Classification Algorithm (BIANCA)¶
Library: FSL | Docker Image: brainlife/fsl
Function¶
Automated white matter hyperintensity (WMH) segmentation using supervised machine learning (k-nearest neighbor) trained on manually labeled data.
Modality: T1-weighted and FLAIR images (3D NIfTI), plus training data with manual WMH masks.
Typical Use: Automated white matter lesion segmentation in aging, small vessel disease, or MS studies.
Key Parameters¶
--singlefile (input file list), --labelfeaturenum (which feature is the manual label), --brainmaskfeaturenum (brain mask feature), --querysubjectnum (subject to segment), --trainingnums (training subjects)
Key Points¶
Requires training data with manual WMH labels. Uses spatial and intensity features. Performance depends on training data quality and similarity to test data.
Inputs¶
| Name | Type | Required | Label | Flag |
|---|---|---|---|---|
singlefile |
File |
Yes | Master file listing subjects, images, masks, and transformations | --singlefile= |
training_data |
Directory |
Yes | Directory containing all subject data files referenced in master file | |
querysubjectnum |
int |
Yes | Row number in master file for the subject to segment | --querysubjectnum= |
brainmaskfeaturenum |
int |
Yes | Column number in master file containing brain mask | --brainmaskfeaturenum= |
labelfeaturenum |
int |
Yes | Column number in master file containing manual lesion mask | --labelfeaturenum= |
trainingnums |
string |
Yes | Training subject row numbers (comma-separated) or "all" | --trainingnums= |
output_name |
string |
Yes | Output file basename | -o |
featuresubset |
string |
No | Comma-separated column numbers for intensity features | --featuresubset= |
matfeaturenum |
int |
No | Column number containing MNI transformation matrices | --matfeaturenum= |
spatialweight |
double |
No | Weighting for MNI spatial coordinates (default 1) | --spatialweight= |
patchsizes |
string |
No | Patch sizes in voxels (comma-separated) | --patchsizes= |
patch3D |
boolean |
No | Enable 3D patch processing | --patch3D |
selectpts |
enum |
No | Non-lesion point selection strategy | --selectpts= |
trainingpts |
string |
No | Max lesion training points per subject (number or "equalpoints") | --trainingpts= |
nonlespts |
int |
No | Max non-lesion points per subject | --nonlespts= |
saveclassifierdata |
string |
No | Save training data to specified file | --saveclassifierdata= |
loadclassifierdata |
string |
No | Load pre-saved classifier data from file | --loadclassifierdata= |
verbose |
boolean |
No | Verbose output | -v |
Accepted Input Extensions¶
- singlefile:
.txt
Outputs¶
| Name | Type | Glob Pattern |
|---|---|---|
wmh_map |
File |
$(inputs.output_name).nii.gz, $(inputs.output_name) |
log |
File |
bianca.log |
err_log |
File |
bianca.err.log |
Output Extensions¶
- wmh_map:
.nii.gz
Docker Tags¶
Available versions: latest, 6.0.4-patched2, 6.0.4-patched, 6.0.4, 6.0.4-xenial, 5.0.11, 6.0.0, 6.0.1, 5.0.9
Categories¶
- Structural MRI > FSL > Lesion Segmentation