Scholarlink

In long-term scalp electroencephalography (sEEG) recordings, physiological artifacts compromise data integrity and hinder its subsequent use and exploration. Presently, extant research predominantly focuses on the targeted mitigation of specific categories of artifacts, thereby failing to adequately account for the inherent diversity exhibited by sEEG artifacts. Consequently, these methods are limited in practical application. To address these limitations, an innovative brain region selection-based artifact detection algorithm is proposed in this paper, which combines deep neural network (DNN) and auto-encoder (AE) model. The DNN is employed to unearth the latent relationships among sEEG while the AE model is designed to complete the recognition of brain region. The proposed method utilizes the concept of anomaly detection to partition areas affected by artifacts, effectively minimizing interference caused by differences in samples. It can effectively identify the diverse artifacts from the sEEG data, consequently augmenting the utility of sEEG data. All sEEG data employed in this paper are collected from the Children’s Hospital of Zhejiang University (CHZU). Through the combination of unsupervised brain region selection and the DNN integrating multi-label classification, the proposed algorithm achieves the accuracy of 90% for artifact detection.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Neural Network Based Unsupervised Artifact Detection Method on Electroencephalography

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Huijun Wang,
  • Dinghan Hu,
  • Zhibo Fang,
  • Tiejia Jiang,
  • Feng Gao,
  • Zuonian Xie,
  • Jiuwen Cao

摘要

In long-term scalp electroencephalography (sEEG) recordings, physiological artifacts compromise data integrity and hinder its subsequent use and exploration. Presently, extant research predominantly focuses on the targeted mitigation of specific categories of artifacts, thereby failing to adequately account for the inherent diversity exhibited by sEEG artifacts. Consequently, these methods are limited in practical application. To address these limitations, an innovative brain region selection-based artifact detection algorithm is proposed in this paper, which combines deep neural network (DNN) and auto-encoder (AE) model. The DNN is employed to unearth the latent relationships among sEEG while the AE model is designed to complete the recognition of brain region. The proposed method utilizes the concept of anomaly detection to partition areas affected by artifacts, effectively minimizing interference caused by differences in samples. It can effectively identify the diverse artifacts from the sEEG data, consequently augmenting the utility of sEEG data. All sEEG data employed in this paper are collected from the Children’s Hospital of Zhejiang University (CHZU). Through the combination of unsupervised brain region selection and the DNN integrating multi-label classification, the proposed algorithm achieves the accuracy of 90% for artifact detection.