Deep Source

This is the Github repository for the Deep Source project.

To-Do:

1. • Define head model (e.g. single sphere, overlapping spheres or single shell)
2. • 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.
3. • Given our newly calculated lead field matrix, we can “make up” some ground truth source space data.
4. • 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.
5. • We then run the inversion algorithm to estimate the sources.
6. • 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