Dear FreeSurfer developers,

Recently, I attempted to run FreeSurfer on my new work computer (MacBook Pro, Apple M3 Max, macOS 14.6). I found that v8.0.0 beta is the only compatible version, and the installation appeared to be successful.

However, when running recon-all on my 7T anatomical data using the following command:

recon-all -i <path_to_anatomy_nii_file> -s <subject_initials> -all

I encountered an error during subcortical segmentation (see attached "7T_recon-all.log" and screenshots). It looks like the deep learning model used by mri_sclimbic_seg was reporting an input dimension mismatch. 

To diagnose the issue, I tested the following possibilities:

1. Is the error specific to this data?

I ran recon-all on other 7T anatomical data, but the same error occurred.

2. Is the error specific to 7T data?

I tested recon-all with 3T data, but in this case, the process was killed even earlier, and no useful error message was provided (see "3T_recon-all.log"). I’m unsure whether these two issues are related.

3. Is the issue related to the FreeSurfer version or my work environment?

I ran recon-all on an older MacBook Pro (Apple M1, macOS 12.6) using FreeSurfer 7.2.0, and the process completed successfully. This suggests that the issue is specific to FreeSurfer v8.0.0 beta or my work environment rather than the dataset itself.

Do you have any suggestions on this issue?

Additionally, I noticed that FreeSurfer v8.0.0 beta uses a deep learning model for segmentation, which differs from previous versions. Is there a way to switch back to the traditional segmentation method?

I would greatly appreciate any insights or troubleshooting suggestions.

Best regards,

Lee Lin