Using generative models to make probabilistic statements about hippocampal engagement in MEG

Area of Science:

• Neuroscience
• Biophysics
• Medical Imaging

Background:

• Magnetoencephalography (MEG) offers non-invasive, real-time brain activity monitoring.
• Detecting signals from deep brain structures like the hippocampus with MEG is challenging due to depth, complexity, and proximity to the neocortex.

Purpose of the Study:

• To develop and validate a probabilistic method for quantifying hippocampal engagement using MEG.
• To overcome limitations in detecting signals from deep brain sources with advanced source modeling.

Main Methods:

• Constructed two generative models: one for cortical surface activity, another for both cortical and hippocampal surface activity.
• Employed Bayesian model comparison to determine the most likely generative model for given data.
• Conducted control experiments, including simulating medial temporal lobe sources and testing anatomical specificity and robustness against co-registration error and sensor noise.

Main Results:

• The model comparison framework successfully identified simulated hippocampal activity.
• The method is sensitive to hippocampal signals when co-registration error is below 3 mm and sensor-level signal-to-noise ratio (SNR) is above -20 dB.
• These technical requirements are achievable with current subject-specific head-cast technology.

Conclusions:

• This study presents a robust method for probabilistically quantifying hippocampal engagement using MEG.
• The findings significantly advance the capability of MEG to investigate deep brain structures, particularly the hippocampus.
• This technique holds promise for future research into neurological and psychiatric conditions involving the hippocampus.