Dear Melissa, I am CCing the senior author Dr. Paz-Alonso so he can also chime in (I wish I could also CC the neuroanatomist that made the annotations, but, sadly, he passed away last year). THOMAS is great but the underlying atlases used Morel as basis for manual segmentation so it is expected that they agree better with the atlas… and, even then, if you look at the Dice scores in the paper, there are plenty of nuclei where FS is higher. I actually find the experiments on AD staging / classification more compelling. Our atlas has the advanrage of working with any MRI contrast “out of the box”, and is pretty good at snapping to the boundaries of the mediodorsal and posterior nuclei (pulvinar), especially if you provide images with good contrast. But if you have images that match the contrast of THOMAS well, then by all means exploit that agreement and use THOMAS! Just my 2 cents 😉 Kind regards, /Eugenio

From: Melissa Thalhammer melissa.thalhammer@tum.de Sent: Friday, October 17, 2025 3:12 AM To: Freesurfer@nmr.mgh.harvard.edu Subject: [Freesurfer] Validation of the thalamic nuclei segmentation

External Email - Use Caution

Dear FreeSurfers,

I read that criticism about the thalamic nuclei segmentation based on T1-images was raised (DOI: 10.1016/j.neuroimage.2022.119340https://secure-web.cisco.com/1V8gZgz38WZjYs0sX-xsLC9vnYMqApMNA_X1Jr2Efj69f_PXS1pgFAGBv1YAwZ6H1YLm6yRH5QnEw5rKyLcgK_qo2zzfLuRvgOyA2_gB05SHLcjfIjUnrDxKwRTisGrKAZkirFlb1KI5M8GVOqg-yil7b68sC42CRSeaKpkZLSsvYxn18xJYJos_1RnNzMow_e7Eeq5I_XeuDi48xWHzg7kXRQndvvH8Jzoq3Cgz1Z-8ASOj3OZ3nXhoHkWlgVpEDrSY_gPZVQPiv-QYkQM1UhODOXkVwdyIZLwFFCjIk1VOyLVdK8mw1Cqg41D1sGMOuhq8BlcTOJp0OY7TyfMGJ7w/https%3A%2F%2Fdoi.org%2F10.1016%2Fj.neuroimage.2022.119340 and DOI: 10.1162/imag_a_00166https://secure-web.cisco.com/1HuYBMxn8grUwviyS1dIybnKr9s2bpABgLeHoaKYUZXVItxs7NIM4E03Fg73ECMHebtdzD4XEbg-oU-FKqrHqSrD3nnhJWKFIgU60pvSLCok8BT0RpLfNsDFBX-LuQlSk5zNkB3ccgbSW0sAh3Krs8nu2MN9r4REaCxCa-K5_C31WW1o51KPZXQaiSwTrldO1-_h_k5og_ukR2KxCe6vmTrnYxwukfBscHrHUsXVfzBDqVKnzNNOMbFBzX46hPRcwx9HgKkfPD9NTTzez0EZ5CMRszJUXJrhXmHx7XrMtsGlbeFD3_Q0cXAdL70GJStBP1-8qBHPEk2vXK0Uef1O2Xg/https%3A%2F%2Fdoi.org%2F10.1162%2Fimag_a_00166) as overlaps with the Morel stereotactic atlas were not always satisfying. However, I was wondering whether the segmentations with the Morel atlas (registered to MNI space) of individual thalami can be considered a proper validation since individual variation of thalamic nuclei sizes are not taken into account and the atlas is not probabilistic.

Can someone elaborate on this criticism?

Thank you, Melissa