Skip to content

README.md

Latest commit

 

History

History
30 lines (24 loc) · 1.13 KB

README.md

File metadata and controls

30 lines (24 loc) · 1.13 KB

Deep Source

This is the Github repository for the Deep Source project.

To-Do:

    • Define head model (e.g. single sphere, overlapping spheres or single shell)
    • Calculate lead field matrix after sensor locations are specified. Fortunately this step is done for us by the MEG scanner, as long as we are in the same co-ordinate system, i.e. we have done co-registration.
    • Given our newly calculated lead field matrix, we can “make up” some ground truth source space data.
    • We then multiply this ground truth data by the lead field, to get synthetic sensor space data. We can then add some noise to this to get a more feasible/realistic sensor space data set.
    • We then run the inversion algorithm to estimate the sources.
    • We compare the accuracies

Possibilities:

Simulation

  • Deep sources
  • Correlated sources
  • Close sources
  • Different SNR
  • Co-registration error

Source reconstruction method

  • Minimum norm estimation (MNE)
  • Beamformer
  • HMM Beamformer (Maximum Likelihood)

Accuracy measures

  • Crosstalk-to-Signal Ratio (CSR)
  • Neural Activity Index (NAI)
  • Point-Spread Functions