[WIP] Cortical projection

Source reconstruction approaches aims at finding the localization of focal generators of some activity observed in the channel. They assume that the meaningful activity has already been extracted from channel space measurements. For instance, one computes window-averaging time-course for all channels and then the reconstruction process tries to find the source of this time-course over the cortex.

Reconstruction approaches are suited if one wants to do activity extraction first (win avg, GLM, FIR...) and then project on the cortex. However, the reconstruction of statistics cannot properly be done. Indeed t-stat values in the channel-space will loose their distribution properties when reconstructed. The control for multiple comparison is also difficult. The obtain proper cortical statistics, we should run the activity extraction in the cortical space. However, reconstruction approaches are not suited for the projection of raw measures from the channel-space to the cortical mesh. Indeed, no meaningful activity has been extracted so there is no focal activity to search for. Consequently, reconstruction will over-regularized signals and especially disrupt the signal temporal structure.

Instead, projection approaches interpolate the full raw time-series onto the cortex, without assumption of structured activity. From there, one can apply activity extraction methods at each cortical vertex and obtain well-defined statistics.

To help differentiate the two processes, here are the operation pipelines:

• Reconstruction approach: raw signal -> activity extraction -> cortical reconstruction
• Projection approach: raw signal -> cortical projection -> activity extraction

Here, we propose a projection method that simply performs a weighted interpolation based on the sensitivity computations.

Input:

• Cortical mesh
• Gain matrix (sensitivities)
• preprocessed NIRS data (channel-space)

Ouput:

• Cortical signals (one time-series per vertex)