Data Labeling

[Mahmoud]

Hello Everyone,

I hope this message finds you well. We are currently working on a project using TVB to generate synthetic EEG data for training a classifier model. The goal is to create two distinct classes of EEG signals: one representing normal brain activity and the other simulating epileptic seizures.

While I've found a helpful example here: Exploring Epileptor 2D, I'm unsure about the best approach to label these generated signals accurately. Specifically, I'm looking to understand how to distinguish between 'normal' and 'epileptic' states within the data. Is there a method to identify the exact moments when the simulated seizures occur, allowing for precise labeling?

I'm reaching out for any advice, insights, or guidance you might have on this topic.

Thank you very much for your time and support.

[meysam....@gmail.com]

Hi,

Embedding the Epiletor model  in the VEP framework is mainly designed to fit Stereotactic‐electroencephalography (SEEG) data, which entails invasive recordings. Typically, we fit the envelope of SEEG data using a 2D Epileptor model in Bayesian framework (eg., see here). For simulating (pre)interictal spikes, a 6D model is often employed. Nevertheless, one can tune the parameters such as the time scale separation to get EEG resemblance data.

One of the main reason to use the VEP, is to make model-based inference, for the causal estimation and the interpretability, rather than the classifying based on the seizure onset by signal processing, due to the complexity of seizure generation and propagation (network connectivity, noise etc) .  The main objective then become to estimate a bifurcation parameter (x_0  or sometimes called eta) in the 2D Epileptor model. If x_0 is less than -2.05, the region is considered healthy, exhibiting damped oscillations. Conversely, if x_0 is greater than -2.05, the region is deemed the Epileptogenic Zone (EZ), as the model displays sustained oscillations. Near the threshold value of -2.05, the region is known as the Propagation Zone (PZ), where, in the absence of coupling or noise, it exhibits damped oscillations but may be recruited to the seizure activity with coupling or noise (see Fig S2 here). Some of our works have focused on identifying the EZ by fitting the model to the envelope of SEEG data. Neverthess, by the fitting model, we the get access to the hidden states and latent dynamics, which then one can make the classification based on the onset time (e.g., see here). Nevertheless, for the sake of simplicity, one may just calculate the the envelope of seizures, and just classify the 'normal' and 'epileptic'  based on the onset time (see Fig S3, in Wang et al).

Best,

Meysam

[meysam....@gmail.com]

BTW, it is important to pint out that Epileptor is a complete taxonomy of system bifurcations to realistically reproduce the dynamics of onset, progression, and offset of seizure-like events (Saggio et al., 2020), hence is able to mimic the source data and applies for all imaging data. To map for specific modality (sEEG or EEG, ..), we need the corresponding lead-field (gain matrix).

For an example see Fig S1 in this paper, or this paper using deep learning.

Best,

Meysam