How does artificial intelligence contribute to iEEG research?

Julia Berezutskaya, Anne-Lise Saive, Karim Jerbi, Marcel van GervenarXiv 2022 · DOI · arXiv

Abstract

Artificial intelligence (AI) is a fast-growing field focused on modeling and machine implementation of various cognitive functions with an increasing number of applications in computer vision, text processing, robotics, neurotechnology, bio-inspired computing and others. In this chapter, we describe how AI methods can be applied in the context of intracranial electroencephalography (iEEG) research. IEEG data is unique as it provides extremely high-quality signals recorded directly from brain tissue. Applying advanced AI models to these data carries the potential to further our understanding of many fundamental questions in neuroscience. At the same time, as an invasive technique, iEEG lends itself well to long-term, mobile brain-computer interface applications, particularly for communication in severely paralyzed individuals. We provide a detailed overview of these two research directions in the application of AI techniques to iEEG: (1) the development of computational models that target fundamental questions about the neurobiological nature of cognition (AI-iEEG for neuroscience) and (2) applied research on monitoring and identification of event-driven brain states for the development of clinical brain-computer interface systems (AI-iEEG for neurotechnology).

Introduction

Intracranial electroencephalography (iEEG), encompassing both electrocorticography (ECoG) and stereo-EEG (sEEG), offers unparalleled signal quality and spatio-temporal resolution by recording directly from the cortical surface or deep brain structures. However, the richness of this data presents a "curse of dimensionality." Traditional analysis often relies on human-selected features, such as gamma power in specific regions, which may overlook the complex, non-linear dynamics of neural populations. Artificial intelligence (AI) provides a fundamental framework to move beyond these limitations, transforming iEEG research from simple correlation analysis to predictive modeling of human cognition and intention.

Frameworks for AI in iEEG

Encoding Models

Encoding models map stimulus features—such as pixels in a video or phonemes in speech—to observed brain activity. This research area is currently transitioning from linear models to deep learning architectures.

Comparison of neural encoding models: from linear models on hand-crafted features to end-to-end DNNs.

As illustrated above, traditional linear models rely on explicit feature selection. In contrast, deep neural networks (DNNs) can extract features automatically. The most advanced "end-to-end" models learn to map raw stimuli directly to neural data, often explaining significantly more variance than hand-crafted features.

Decoding Models

Decoding models perform the reverse operation, inferring stimulus or intent from brain activity. These are the backbone of clinical brain-computer interfaces (BCIs). Methods range from "shallow" machine learning (e.g., SVMs, Random Forests) to "deep" architectures like CNNs for spatial patterns and Transformers for temporal sequences. Researchers also use Representational Similarity Analysis (RSA) to compare the "distance" between neural patterns to the activations of computational models, testing if an AI model's internal logic mirrors biological processing.

Key Insights and Results

One of the most profound findings in recent AI-iEEG research is that DNNs trained on objective tasks (like object recognition) often spontaneously develop internal representations that mirror the human hierarchy of visual and auditory processing.

Speech encoding performance and electrode tuning analysis.

DNN-based encoding models consistently outperform linear models in explaining brain responses to complex stimuli. For example, in speech processing, models can achieve high correlation (e.g., r=.51) between predicted and observed neural amplitudes. In the realm of neurotechnology, AI-driven BCIs have achieved remarkable milestones, such as 90% accuracy in binary decoding for ALS patients and the ability to decode full sentences at speeds approaching natural speech. Additionally, AI-guided adaptive stimulation has shown up to a 70% reduction in seizure frequency in epilepsy patients compared to standard treatments.

Challenges and Future Directions

Despite these successes, several challenges remain:

  • Overfitting: iEEG studies often have many features but few subjects. Simpler models or transfer learning are essential mitigations.
  • Generalization: Every patient’s electrode placement is unique, necessitating the search for a "common space" across different subjects.
  • Explainability: As models become "black boxes," there is an urgent need for Explainable AI (XAI) to ensure findings are biologically grounded and clinically safe.
  • Biological Plausibility: Future research aims to move toward "biologically grounded approximations" of learning, as the brain likely does not use standard backpropagation.

Summary updated

Abstract

Artificial intelligence (AI) is a fast-growing field focused on modeling and machine implementation of various cognitive functions with an increasing number of applications in computer vision, text processing, robotics, neurotechnology, bio-inspired computing and others. In this chapter, we describe how AI methods can be applied in the context of intracranial electroencephalography (iEEG) research. IEEG data is unique as it provides extremely high-quality signals recorded directly from brain tissue. Applying advanced AI models to these data carries the potential to further our understanding of many fundamental questions in neuroscience. At the same time, as an invasive technique, iEEG lends itself well to long-term, mobile brain-computer interface applications, particularly for communication in severely paralyzed individuals. We provide a detailed overview of these two research directions in the application of AI techniques to iEEG. That is, (1) the development of computational models that target fundamental questions about the neurobiological nature of cognition (AI-iEEG for neuroscience) and (2) applied research on monitoring and identification of event-driven brain states for the development of clinical brain-computer interface systems (AI-iEEG for neurotechnology). We explain key machine learning concepts, specifics of processing and modeling iEEG data and details of state-of-the-art iEEG-based neurotechnology and brain-computer interfaces.
BibTeX
@misc{berezutskaya2022how,
  author = {Julia Berezutskaya and Anne-Lise Saive and Karim Jerbi and Marcel van Gerven},
  title = {{How does artificial intelligence contribute to iEEG research?}},
  year = {2022},
  eprint = {2207.13190},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2207.13190},
}
Figure