Hi Dilek,

0.2 Dice seems extremely low. Also, in my experience, a threshold of 0.05 on T1w-FLAIR scans will show up many FP lesions.

We usually reported a Dice in the range of 0.50-0.65 on various datasets (MS and WMH, default threshold, T1w-FLAIR), so I'm surprised by a such low value.

Can you share with us your run_samseg command?

Stefano

Date: Sun, 4 Sep 2022 15:53:17 +0300

From: Dilek ?zkan <dozkan521@gmail.com>

Subject:    

To: Freesurfer support list <freesurfer@nmr.mgh.harvard.edu>

Message-ID:

        <CAPQSyvLqkvjk7JFxFoAG8RcX7ff40TS70vzvF6uWWHwC6CgqEg@mail.gmail.com>
Content-Type: text/plain; charset="utf-8"

        External Email - Use Caution       

Hello all,

I use Freesurfer-Samseg for MS lesion segmentation(cross-sectional).

I use the T1 and Flair for this. (just these are inputs) I registered *flair
and t1 to DIR *for every patient. (My manual lesion masks are on DIR
imaging.) I controlled the registrations, these are true.

I created a binary lesion mask using *threshold 0.05. *(mri_binarize --i
/data/projects/MS_MRI/posteriors/Lesions.mgz --o
/data/projects/MS_MR/binarize/lesion_binary.mgz --min 0.05)
I checked the dice score with ground truth values with python.

*threshold *0.05= *dice* 0,2 of the dataset

Do you have any recommendations to increase the dice?

# cvs_version 7.2.0
# sysname Linux
# machine x86_64

Best regards