A mild cognitive impairment diagnostic model based on IAAFT and BiLSTM. (February 2023)

• Record Type: Journal Article
• Authors: Li, Xin; Zhou, Hao; Su, Rui; Kang, Jiannan; Sun, Yu; Yuan, Yi; Han, Ying; Chen, Xiaoling; Xie, Ping; Wang, Yulin; Liu, Qinshuang
• Is Part Of: Biomedical signal processing and control. Volume 80:Part 2(2023)
• Publication Date: 2023-02
• Subjects: Iterative amplitude adjusted Fourier transform (IAAFT) -- Bidirectional long short-term memory (BiLSTM) -- Electroencephalography (EEG) -- Sample entropy (SampEn) -- Mild cognitive impairment (MCI); Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104349
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Surrogate data generation method IAAFT solves the issue of EEG data shortage. BiLSTM obtains better classification result than other traditional methods. The SampEn extraction method retaining the high temporal resolution of EEG. Abstract: The early diagnosis of mild cognitive impairment (MCI) is a essential prevention of further development of MCI into Alzheimer's disease (AD). Electroencephalogram (EEG) has many advantages compared to other methods in the analysis of AD in an early stage, but there are some limitations of EEG such as small size of datasets caused by difficulty in clinical data collection and too many other interfering signals are contained. Recent years, deep learning (DL) have overcome these limitations relatively. In this study, a novel model which aims to classify MCI and healthy control (HC) was constructed based on iterative amplitude adjusted Fourier transform (IAAFT) and bidirectional long short-term memory (BiLSTM). IAAFT is used to overcome the problems caused by small datasets; sample entropy (SampEn) feature extraction is used to further reduce computational time and obtain better classification results; BiLSTM for better capture of EEG temporal connections. The performance of the model was evaluated on a clinical dataset containing 10 MCI and 10 HC. Compared with the traditional EEG classification method, the result shows that BiLSTM is more suitable for the EEG classification task, and the classification accuracy is significantly improved by data augmentation. After performing 10-fold cross-validation and 10-fold data augmentation, the model achieved a maximum classification accuracy of 97.20 ± 1.74 %. The results indicate that the model can be used to diagnose MCI patients with the EEG small datasets. Meanwhile, The data augmentation used in this study has a high reference value for other resting-state EEG classification tasks.