Automatic delineation of MS lesions. The model, a nnU-Net, was trained on 549 clinical patients.
Please cite for use:
Hindsholm AM, Andersen FL, Cramer SP, Simonsen HJ, Askløf MG, Magyari M, Madsen PN, Hansen AE, Sellebjerg F, Larsson HBW, Langkilde AR, Frederiksen JL, Højgaard L, Ladefoged CN and Lindberg U (2023) Scanner agnostic large-scale evaluation of MS lesion delineation tool for clinical MRI. Front. Neurosci. 17:1177540. doi: 10.3389/fnins.2023.1177540
The models require FLAIR, T2, and T1 input files. You can call the function with:
AIMS -flair <FLAIR.nii.gz> -t2 <T2.nii.gz> -t1 <T1.nii.gz> -o <AIMS_mask.nii.gz>
You can also run the function on a folder with preprocessed and correctly formatted files:
AIMS_folder -input <INPUT_FOLDER> -o <OUTPUT_FOLDER>
The files in the folder must be named <PatientID>_0000.nii.gz
(the flair), <PatientID>_0001.nii.gz
(the t2), <PatientID>_0002.nii.gz
(the t1). The output dir will contain the file <PatientID>.nii.gz
(the output mask), for each PatientID. AIMS_folder is much faster than AIMS when you need to process many files. AIMS_folder require already preprocessed files (see below).
Before you run the above command, you first need to perform the following preprocessing steps:
- Standard orientation with
reorient2std
, - Resample to same spacing (e.g. FLAIR) with
flirt
- Skull strip, e.g. with
hd-bet
If you wish to perform these steps as part of the algorithm, call the function with the --preprocess
flag:
AIMS -flair <FLAIR.nii.gz> -t2 <T2.nii.gz> -t1 <T1.nii.gz> -o <AIMS_mask.nii.gz> --preprocess