Related Experiment Video
Updated: Jan 18, 2026
10:14
3D Scanning Technology Bridging Microcircuits and Macroscale Brain Images in 3D Novel Embedding Overlapping Protocol
Published on: May 12, 2019
7.6K
Neural Mosaics: Detecting Aberrant Brain Interactions using Algebraic Topology and Generative Artificial
Katrina Prantzalos1, Dipak Upadhyaya1, Pedram Golnari1
1Department of Population and Quantitative Health Sciences, Case Western Reserve University School of Medicine, Cleveland, OH, USA.
Summary
This study explored using algebraic topology and AI for epilepsy seizure detection. While direct AI analysis of brain topology data showed limitations, the approach promises scalable future analyses without extensive training data.
Area of Science:
- Computational Neuroscience
- Artificial Intelligence in Medicine
- Epilepsy Research
Background:
- Epilepsy impacts millions globally, with surgical outcomes often limited.
- Current electroencephalography (EEG) seizure detection is resource-intensive and misses complex brain interactions.
- Algebraic topology, specifically persistent homology, offers advanced methods for analyzing brain network dynamics.
Purpose of the Study:
- To investigate the efficacy of using persistent homology and a large language model (LLM) for seizure detection in refractory epilepsy.
- To develop a novel prompting template for classifying topological structures derived from intracranial EEG (iEEG) data using an LLM.
Main Methods:
- Computed topological structures from iEEG recordings of epilepsy patients using persistent homology.
- Utilized a novel prompting template to input persistence diagrams into the Google Gemini Pro Vision 1.0 LLM for classification.
- This represents the first application of persistence diagrams as input for an LLM in analyzing brain interaction dynamics.
Main Results:
- Directly prompting LLMs with persistence diagrams proved insufficient for accurate seizure detection.
- The approach demonstrated potential for scalable analysis, unlike traditional machine learning methods.
- It bypasses the need for extensive training datasets and complex hyperparameter tuning.
Conclusions:
- While direct LLM classification of topological data is currently limited for seizure detection, the methodology shows promise.
- This novel approach highlights future possibilities for more efficient and scalable epilepsy analysis.
- Further research is warranted to refine LLM integration with topological data for improved clinical applications.
