It is critical for clinical diagnosis and research to extract and utilize the medical information from electroencephalogram (EEG) reports of epilepsy. With the widespread application of deep learning techniques in health care, particularly in natural language processing tasks, named entity recognition (NER) has emerged as an essential tool for information extraction from medical text. This study proposed an epilepsy NER model for EEG reports to improve the efficiency of epilepsy text processing. 17,606 paragraphs from real Chinese epilepsy EEG reports as data samples, were meticulously annotated with 17 entity types by clinical experts. The model used the BERT and the Global Pointer (GP) algorithm to identify nested and non-nested entities, achieved outstanding performance in the Precision, Recall and F1 score. The experimental results demonstrate that our method significantly enhances the effectiveness of epilepsy entity recognition, providing robust support for the automated extraction and analysis of medical information.