A Hybrid Artificial Intelligence System for Automated EEG Background Analysis and Report Generation

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Electroencephalography (EEG) is vital for diagnosing neurological disorders but manual interpretation is prone to errors, especially in under-resourced facilities.
• Lack of advanced EEG analysis systems in small hospitals and clinics leads to misinterpretations.
• Automated interpretation can enhance diagnostic accuracy and accessibility.

Purpose of the Study:

• To develop and validate a hybrid artificial intelligence (AI) system for automatic EEG interpretation and report generation.
• To improve the accuracy of EEG analysis in resource-limited settings.
• To assist neurologists in diagnosing neurological disorders more effectively.

Main Methods:

• A hybrid AI system combining deep learning for posterior dominant rhythm (PDR) prediction and artifact removal with expert algorithms for abnormality detection.
• Ensemble deep learning models were trained on 1530 labeled EEGs for PDR prediction.
• Large language models (LLMs) were utilized for automated report generation.

Main Results:

• The AI system achieved high accuracy in PDR prediction (e.g., 91.8% within 0.6 Hz error).
• The AI system significantly outperformed neurologists in detecting generalized background slowing (F1: 0.93 vs. 0.82).
• AI demonstrated consistent performance on internal and external datasets (F1: 0.884 and 0.835) and LLM-based reports achieved 100% accuracy.

Conclusions:

• The hybrid AI system offers a scalable and accurate solution for EEG interpretation, particularly beneficial for resource-limited settings.
• This AI tool can assist neurologists, enhance diagnostic accuracy, and reduce misdiagnosis rates.
• Automated EEG analysis holds significant potential for improving neurological care globally.