Meeting 21-02-2022

[gjheij]

We discussed some of the results obtained from the Size-Response and line-pRF experiments. Though initial results look sort of promising, there's a few things to consider:

Size-Response:

• Plot the betas for time derivatives > obtain highly detailed characterization of HRF. This entails I'll need to perform the GLM with nilearn, as my glm function doesn't include derivatives
• Perform similar analysis as described here, but instead of using fourier basis sets, we use gamma basis sets
• To gain SNR, we could increase the stimulus duration to 2s. (EDIT: I just checked, we've been scanning with a stimulus duration of 1.5s, so there might not be a lot to gain from that) pRF:
• Fit HRF from pRF: this would give us an idea of what to expect regarding the SR-experiment
• Fit pRF on average across ribbon> use this as starting point for fit of individual voxels? This might serve as a constrain on the wildly off pRFs bordering GM/CSF-boundary

General:

• Check frequency of respiration/cardiac > notch filter on personalized traces on top of low-pass?
• Check fidelity of respiration/trace > power spectrum next to timecourse
• scipy.filfilt > filtering might shift timecourse; messes up onsets

[gjheij]

EDITS as of 25-02-2022

Size-Response:

• Plot the betas for time derivatives > obtain highly detailed characterization of HRF. This entails I'll need to perform the GLM with nilearn, as my glm function doesn't include derivatives Implemented time+dispersion derivatives here and use different flavors of HRF-specifications in the GLM here

• Perform similar analysis as described here, but instead of using fourier basis sets, we use gamma basis sets I now used the canonical_hrf_with_time_derivative as implemented in nideconv. Is that the correct one?

pRF:

• Fit HRF from pRF: this would give us an idea of what to expect regarding the SR-experiment Not done yet

• Fit pRF on average across ribbon> use this as starting point for fit of individual voxels? This might serve as a constrain on the wildly off pRFs bordering GM/CSF-boundary This doesn't seem to help too much.. The timecourse is too wiggly for the model to pick up on the bumps (see cell 131 here. However, this was done on the older data, not using the data from 21-02-2022. That's because we were still playing around with the power spectra for that data

General:

• Check frequency of respiration/cardiac > notch filter on personalized traces on top of low-pass? Notching seems a bit iffy.. I tried to implement retroicor like Gilles did in my data-loading script, but I'm not sure what to think of the result in cell 137. For RETROICOR, I need to z-score, resulting in extremely small fluctuations in the timecourse

• Check fidelity of respiration/trace > power spectrum next to timecourse We seem to have respiration in our signal (see Zulip-chat) or here

[gjheij]

See subsequent issues for updates on denoising