Diagnosis of Alzheimer’s Disease by Time-Dependent Power Spectrum Descriptors and Convolutional Neural Network Using EEG Signal

Table 1

Summary of machine learning method for brain disease diagnosis with EEG signal.

Author	Year	Disease	Feature extraction	Classification	Results
Xin et al. [80]	2021	Epilepsy	Dimensionality reduction principal component analysis (PCA)	Convolution SVM	The method’s accuracy, sensitivity, and specificity reach up to 99.56%, 99.72%, and 99.52%, respectively
Aliyu and Lim [81]	2021	Epilepsy	Discrete wavelet transforms (DWT)	LSTM network	Reduction of the number of LSTM trainable parameters needed to achieve extreme accuracy
Tuncer [82]	2021	Epileptic seizure	Nonlinear textural feature extraction (Hamsi hash)	-nearest neighborhood	This model has an accuracy in the EEG dataset of 99.20% for five classes and has 100.0% accuracy in other conditions
Cicalese et al. [83]	2020	AD	Pearson correlation coefficient-based feature selection (PCCFS)	LDA	The EEG-fNIRS feature set combination was expected to obtain greater precision (79.31%) by combining its supplementary properties as compared with the EEG (65.52%) or fNIRS alone (58.62%). Moreover, AD development is associated with the right and left parietal lobe
Ferri et al. [84]	2020	AD		Low-resolution brain electromagnetic tomography (LORETA)	Classification accuracy of 80%, 85%, and 89% using rsEEG, sMRI, and features, respectively, discriminates against them
Trambaiolli et al. [85]	2017	AD	Feature selection (FS)	SVM classifier	Since eliminating % of the initial elements, the filtered subset evaluator technique obtained the highest efficiency gain, both on a per-patient basis (91.18% accuracy) and on a per-epoch basis (%)
Nobukawa et al. [86]	2020	AD	Functional connectivity	SVM	A novel interpretation of neural network functions in healthy brains and unhealthy disorders can be provided by applying a mixture of both machine learning approaches to neurophysiological evidence
Kulkarni and Bairagi [87]	2017	AD	Extracting salient features that are spectral-, wavelet-, and complexity-based	SVM	The increased performance in AD diagnosis
Vecchio et al. [88]	2020	AD	—	SVM	A low-cost and noninvasive process uses readily available tools that, when integrated, achieve high sensitivity/specificity and optimum individual classification accuracy (0.97 of AUC)