Questions for HMM applied on task fmri for rule inference

[Chuanyong-Xu]

I am a student from Shenzhen University, and very interested in your papers about HMM. And I am trying your HMM model to analysis my task-fMRI data for rule inference task. However, I have being in trouble for continuing this project. If you are convenient, I want to consult you some questions.
1. This task including two parts: (1) instructed task, where subjects do not need to inference the task rule (the rule has been cued), with 150 trials and collected 3 run of fMRI data (each have 1148 volumes). (2) inference task, the only difference subjects need to inference the task rule (150 trials, 3 runs of fMRI data, each 1148 volumes). The ROIs used for this analysis are 12 regions related to rule inference (instructed task: (11483)volumes30 subs * 12 rois; inference task: (11483)volumes30 subs * 12 rois), these two tasks time series were contacted for HMM analysis.
2. The following is the task procedure and my options for HMM, and I did not find any difference:

4. Can I just focus on the feedback durations to do the HMM, it means that to extract the onset time (1 volume) and following 10 volumes of each trial (instructed: 11*150, inference: 11*150,) for each subjects? And then to compare FO and DW time between these two task. Because we hypothesized that the subjects only need to do inference after feedback presents.

[vidaurre]

Hi Chuanyong,
You can try the HMM on just that subset of the data. This is not a lot of data but it may be worth a try.
You can also try to apply the statistical tests that are built-in , based on permutation testing, to see if they give you a bit of better statistical efficiency: https://github.com/vidaurre/glhmm/blob/main/docs/notebooks/Testing_across_trials_within_session.ipynb
Another thing is that there are various types of HMM (e.g. Gaussian with state-specific covariance, with shared covariance, etc), perhaps some are more sensitive than others.
Diego