I am trying to extract lesion volume from my structural MRI/FLAIR data of brain tumor patients using run_samseg program. Here, I am attaching my output colored in red overlapped on ground truth colored in blue.
As you can see it misses identification of some portion of the lesion (non overlapped region) but instead gives some voxels which are not part of the lesion at all. I tried different lesion thresholds 0.95, default 0.30 and 0.1. In all the cases, results are very similar i.e. a chunk of lesion voxels are always missing and there are some unwanted voxels that get identified.
I was wondering (a) how can this issue be fixed; (b) or, maybe run_samseg is not designed for tumor patients and it will work only for MS patients.
In addition, I tried mri_WMHsynthseg pipeline, but it stops after about a minute:
>> mri_WMHsynthseg --i BRAINIX_NIFTI_T1.nii.gz --o WMH_Out.nii
Arguments seem correct; loading Python packages...
Using cpu
Using 1 thread(s)
Preparing model and loading weights
Working on image 1 of 1: BRAINIX_NIFTI_T1.nii.gz
Loading input volume and normalizing to [0,1]
Upscaling to target resolution
Pushing data through the CNN
Killed: 9
I would greatly appreciate any help.
Thanks.