Source level analysis II: Analysing source time-series

PracticalMEEG | 2025 – Guiomar Niso Source level analysis II: Analysing source time-series Guiomar Niso Aix-Marseille Université France | 27-31 October 2025 PracticalMEEG 1 guiomar[email protected] @GuiomarNiso PracticalMEEG | 2025 – Guiomar Niso The human brain 2 (S. Ramón y Cajal) The human brain ~86.000 millon of interconnected neurons Each with ~1000 synaptic connections PracticalMEEG | 2025 – Guiomar Niso 3 Human brain Electroencephalography EEG EEG Magnetoencephalography MEG MEG Measuring human brain activity non invasively? PracticalMEEG | 2025 – Guiomar Niso Human brain Electro/Magneto encephalography (EEG/MEG) Electric/Magnetic fields generated by cell assemblies 4 PracticalMEEG | 2025 – Guiomar Niso Source reconstruction 5 PracticalMEEG | 2025 – Guiomar Niso Source reconstruction 6 Forward Inverse PracticalMEEG | 2025 – Guiomar Niso Sources How do we model the source activity mathematically? Forward model Head How do currents flow from the source towards our sensors through the tissues? Sensors What sensors do we have and their position with respect to the brain? 7 PracticalMEEG | 2025 – Guiomar Niso Planar gradiometer δB/δx Axial gradiometer δB/δy Magnetometer Bz 00 8 Forward model: Sensors (Adapted from Fieldtrip) EEG MEG 0 PracticalMEEG | 2025 – Guiomar Niso Alignment of MEG data: channel positions, headpoints Corregistration: MRI + M/EEG space → one coordinate system (fiducials) 9 HEAD POINTS Alignment CHANNELS ANATOMY MRI T1W dicom/nii SEGMENTED Freesurfer FUNCTIONAL MEG MEG DATA Raw data Forward model: Sensors CORREG MRI-MEG FIDUCIALS Mark fiducials Sensors What sensors do we have and their position with respect to the brain? PracticalMEEG | 2025 – Guiomar Niso ll posed problem: more source points than sensors (thousands vs hundreds) → infinite number of solutions 16 SENSORS SOURCES ~300 ~15.000 Inverse model Constraints to make it solvable Inverse model Estimate source brain activity from measured sensor data PracticalMEEG | 2025 – Guiomar Niso 17 Ŝ = W mŜ: estimated source activity W: inverse model (imaging kernel) m: measured sensor data Spatial filters Ŝ: independent sources unit gain and minimize variance ●Beamformers Distributed sources Ŝ: distributed sources minimize residuals and noise ●Minimum norm estimation Dipole fitting Ŝ: one or very few sources constraints: limit sources ●Single dipole Inverse model (Westner et al. 2022) PracticalMEEG | 2025 – Guiomar Niso Constrained: normal to cortex Unconstrained: 3 orthogonal dipoles Source activity 19 1 dipole 3 dipoles (norm) Absolute valuesAbsolute values PracticalMEEG | 2025 – Guiomar Niso SOURCES 20 ALL ROIs ATLAS PracticalMEEG | 2025 – Guiomar Niso SOURCES 21 ALL ROIs Scout function Mean: Average all the signals. Mean(norm): Average absolute values of all the signals. PCA: First mode of the Principal Component Analysis. Max: For each time point, get the maximum across all the vertices. Power: Average the square of all the signals. RMS: Square root of average the square of all the signals. All: Returns all the signals. ATLAS PracticalMEEG | 2025 – Guiomar Niso We can’t average individual source maps → project to a common template Project to default template 22 PracticalMEEG | 2025 – Guiomar Niso SOURCES ~15000 23 Brain signal analysis ? SIGNAL t SENSORS ~300 CONNECTIVITY PREPROCESSING t EVOKED POTENTIALS t SPECTRAL ANALYSIS f TIME– FREQUENCY t f WAVEFORM SHAPE t CROSS-FREQ COUPLING fp fa NETWORKS ACQUISITION PracticalMEEG | 2025 – Guiomar Niso Connectivity 24 Structural physical connection 21 Functional relation between signals 12 PracticalMEEG | 2025 – Guiomar Niso Functional Connectivity 25 1 2 3 ? For a comprehensive review on functional and effective connectivity metrics: (Niso et al. 2013) PracticalMEEG | 2025 – Guiomar Niso Pearson’s correlation (COR) Linear correlation in time domain between x(t) and y(t) at zero lag COR = 1 COR = -1 COR = 0 COR = 0 Coherence (COH) Linear correlation between x(t) and y(t) as a function of the frequency 26 Classical Methods PracticalMEEG | 2025 – Guiomar Niso Connectivity 49 Non Directed Functional connectivity 12 Directed Effective connectivity 12 Symmetrical Non symmetrical PracticalMEEG | 2025 – Guiomar Niso 50 Strength: sum of weights of links connected to it Degree: number of links connected to it Clustering: Likelihood of neighbours also connected Modularity: Max with-in, Min between modules Walk Shortest path Path Betweenness: Number of all shortest paths in the network that contain it Ring Lattice Small World Random Complete Complex Networks (Rubinov & Sporns, 2010) PracticalMEEG | 2025 – Guiomar Niso SOURCES ~15000 51 Brain signal analysis ? SIGNAL t SENSORS ~300 CONNECTIVITY PREPROCESSING t EVOKED POTENTIALS t SPECTRAL ANALYSIS f TIME– FREQUENCY t f WAVEFORM SHAPE t CROSS-FREQ COUPLING fp fa NETWORKS ACQUISITION PracticalMEEG | 2025 – Guiomar Niso Good scientific practice (Niso et al. 2022a,b) 54 Online resource: https://oreoni.github.io PracticalMEEG | 2025 – Guiomar Niso ¡Muchas gracias! 52 ¡Muchas gracias! @GuiomarNiso guiomar[email protected] Grupo de Neuroimagen PracticalMEEG | 2025 – Guiomar Niso Source level analysis II: Analysing source time-series Guiomar Niso Aix-Marseille Université France | 27-31 October 2025 PracticalMEEG 53 guiomar[email protected] @GuiomarNiso