Scholarlink

<p>Scientists have made substantial progress in diagnosing Schizophrenia (SZ) using electroencephalogram (EEG) data through machine learning and deep learning models. This study aims to enhance these models by improving their feature extraction capability. We conducted experiments primarily on the publicly available Kaggle SZ EEG dataset (81 subjects, 64 EEG channels), and further validated the proposed method on two additional datasets: the IBIB PAN adult EEG dataset (28 subjects, 19 EEG channels) and the NNCI pediatric EEG dataset (84 subjects, 16 EEG channels), in order to assess cross-dataset generalization across different age groups, channel configurations, and acquisition protocols. Traditional EEG analysis commonly applies the Short-Time Fourier Transform (STFT) or Continuous Wavelet Transform (CWT) to obtain time–frequency representations. In contrast, we embed these transforms directly within the convolutional layers of a deep neural network, allowing the model to learn adaptive time–frequency features automatically. Furthermore, a Long Short-Term Memory Projection (LSTM-P) layer is integrated to capture long-range temporal dependencies in EEG sequences. The resulting hybrid CWT/STFT–CNN–LSTM-P architecture achieved a peak accuracy of 96.85%, with strong precision, recall, and F1-score, significantly surpassing conventional approaches. This work highlights the potential of integrating adaptive time–frequency transformations within deep neural frameworks for EEG-based mental disorder diagnosis.</p>

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

Design hybrid neural network for classifying the electroencephalography of schizophrenia patients

  • 期刊论文

  • 发表日期：

    2026年6月1日

查看全文
  • Viet Quoc Huynh,
  • Ly Thi Trinh,
  • Tuan Van Huynh

摘要

Scientists have made substantial progress in diagnosing Schizophrenia (SZ) using electroencephalogram (EEG) data through machine learning and deep learning models. This study aims to enhance these models by improving their feature extraction capability. We conducted experiments primarily on the publicly available Kaggle SZ EEG dataset (81 subjects, 64 EEG channels), and further validated the proposed method on two additional datasets: the IBIB PAN adult EEG dataset (28 subjects, 19 EEG channels) and the NNCI pediatric EEG dataset (84 subjects, 16 EEG channels), in order to assess cross-dataset generalization across different age groups, channel configurations, and acquisition protocols. Traditional EEG analysis commonly applies the Short-Time Fourier Transform (STFT) or Continuous Wavelet Transform (CWT) to obtain time–frequency representations. In contrast, we embed these transforms directly within the convolutional layers of a deep neural network, allowing the model to learn adaptive time–frequency features automatically. Furthermore, a Long Short-Term Memory Projection (LSTM-P) layer is integrated to capture long-range temporal dependencies in EEG sequences. The resulting hybrid CWT/STFT–CNN–LSTM-P architecture achieved a peak accuracy of 96.85%, with strong precision, recall, and F1-score, significantly surpassing conventional approaches. This work highlights the potential of integrating adaptive time–frequency transformations within deep neural frameworks for EEG-based mental disorder diagnosis.