Frontiers in physiology · 2026
Transformer architecture for diagnosing schizophrenia disabilities through EEG analysis.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Alsharif Nizar, Ebrahim Nadhem, Al-Nefaie Abdullah H, Ahmed Zeyad A T, Aldhyani Theyazn H H
Аннотация
INTRODUCTION: Schizophrenia (SZ) presents significant diagnostic challenges in clinical practice.
METHODS: In this study, we explore novel deep learning approaches for the automated detection of this severe psychiatric disorder through electroencephalography (EEG) analysis. Using a publicly available dataset of EEG recordings from 14 SZ patients and 14 healthy controls, we developed a robust processing pipeline that includes bandpass filtering (0.5-45 Hz), artifact removal through Independent Component Analysis, and signal enhancement via wavelet transformation. Our feature extraction approach captured both temporal characteristics, including statistical measures such as mean, standard deviation, skewness, and kurtosis, as well as spectral properties, including power distributions across the delta, theta, alpha, beta, and gamma bands. By combining the F-score method, based on Analysis of Variance (ANOVA), Mutual Information assessment, and random forest techniques, we identified 27 highly discriminative features from the original set of 171 extracted features. We developed and evaluated two novel architectures: an LSTM-based model and a Transformer-based model. Both incorporated attention mechanisms and multi-scale temporal convolutional networks to effectively capture the complex patterns in EEG signals.
RESULTS: While the LSTM model performed strongly with 95.65% accuracy (± 0.16%), 96.28% sensitivity (± 0.59%), and 95.26% specificity (3.84%), our Transformer-based model achieved even more impressive results: 98.20% accuracy (± 0.20%), 98.11% sensitivity (± 0.28%), and 98.12% specificity (± 0.40%) by using k-fold cross-validation. The Transformer-based was achieved 76.95% in Leave-One-Subject-Out (LOSO) cross-validation. These findings highlight the potential of transformer architectures to detect the subtle neurophysiological markers of schizophrenia in EEG recordings.
DISCUSSION: Our approach could eventually provide clinicians with an objective tool to support earlier and more accurate diagnosis of schizophrenia, potentially improving treatment outcomes through timely intervention.
METHODS: In this study, we explore novel deep learning approaches for the automated detection of this severe psychiatric disorder through electroencephalography (EEG) analysis. Using a publicly available dataset of EEG recordings from 14 SZ patients and 14 healthy controls, we developed a robust processing pipeline that includes bandpass filtering (0.5-45 Hz), artifact removal through Independent Component Analysis, and signal enhancement via wavelet transformation. Our feature extraction approach captured both temporal characteristics, including statistical measures such as mean, standard deviation, skewness, and kurtosis, as well as spectral properties, including power distributions across the delta, theta, alpha, beta, and gamma bands. By combining the F-score method, based on Analysis of Variance (ANOVA), Mutual Information assessment, and random forest techniques, we identified 27 highly discriminative features from the original set of 171 extracted features. We developed and evaluated two novel architectures: an LSTM-based model and a Transformer-based model. Both incorporated attention mechanisms and multi-scale temporal convolutional networks to effectively capture the complex patterns in EEG signals.
RESULTS: While the LSTM model performed strongly with 95.65% accuracy (± 0.16%), 96.28% sensitivity (± 0.59%), and 95.26% specificity (3.84%), our Transformer-based model achieved even more impressive results: 98.20% accuracy (± 0.20%), 98.11% sensitivity (± 0.28%), and 98.12% specificity (± 0.40%) by using k-fold cross-validation. The Transformer-based was achieved 76.95% in Leave-One-Subject-Out (LOSO) cross-validation. These findings highlight the potential of transformer architectures to detect the subtle neurophysiological markers of schizophrenia in EEG recordings.
DISCUSSION: Our approach could eventually provide clinicians with an objective tool to support earlier and more accurate diagnosis of schizophrenia, potentially improving treatment outcomes through timely intervention.