Table 6.
Application of the deep learning method in EEG research.
| Ref. | Domain | Proposed Method | Conclusion |
|---|---|---|---|
| Morabito [150] | Alzheimer’s disease | A method was proposed to generate a suitable feature set using convolution and then use full connectivity to make predictions | The method achieved 80% classification accuracy in Alzheimer’s disease |
| Morabito [151] | Alzheimer’s disease | A deep learning processing system to reduce the dimensionality of the feature space | The system achieved nearly 90% classification accuracy in diagnosing Alzheimer’s disease |
| Kim [152] | Alzheimer’s disease | A novel end-to-end model designed for the purpose of low-cost and noninvasive diagnosis of brain disorders | Their method achieved a high ROC-AUC score of 0.9 |
| Kunekar [153] | Epilepsy | A deep learning and multimodal fusion approach was proposed for the diagnosis of epilepsy | The method allowed for improved diagnostic accuracy and earlier prediction of seizures due to the continuous performance of the data |
| Sagga [154] | Epilepsy | Proposed a simple CNN model to identify epileptic seizures | The CNN model achieved 98% accuracy in seizure detection |
| Qing [155] | Epilepsy | Using neural network model to process one-dimensional time series and two-dimensional EEG image EEG data types to detect seizures | The classification accuracy of EfficientNetV2 model for epileptic EEG was 98.69% |
| Ouyu [156] | Ischemic stroke | A deep learning-based stroke evaluation model for stroke diagnosis | CNN was 22.86% more accurate than logistic regression |
| Kumar and Sengupta [157] | Ischemic stroke | Stroke detection using VGG-16 and Resnet-50 models | The accuracy of the model in predicting stroke reached 90% |
| Seal [158] | Depression | A CNN DeprNet was proposed for depression diagnosis | The accuracy of the results obtained in recording split and subjective split experiments was 99.37% and 91.4%, respectively |
| Rafiei [159] | Depression | Automatic detection of MDD Using EEG data and deep neural network architecture | The accuracy reached 91.67% when all 19 channels were used and 87.5% after the channels were reduced |
| Sudhakar [160] | Sleep | Alexnet and GoogleNet used EEG signals to detect sleep disorders | AlexNet was better at detecting sleep disorders with an accuracy of 93.33% |
| Leino [161] | Sleep | Combined CNN and RNN to determine the sleep stage of the EEG channel measured by AES | When considering all datasets, the highest automatic scoring accuracy was 79.7% |
| Kang and Hong [162] | Sleep | The optimized GoogleNet model was used to construct CNN automatic sleep stage classification in single channel EEG | The accuracy of the sleep state of the EEG F4 channel was the highest at 77.6% |
| Almogbel [163] | Cognitive | An end-to-end deep neural network could accommodate the original EEG signals from 4 channels within a month as input | This model could successfully promote EEG signals and classify drivers’ cognitive workload with high accuracy |
| Bhardwaj [164] | Cognitive | A highly accurate, EEG based driver fatigue classification system to reduce fatigue related road accidents | Based on different indicators, the accuracy of the deep learning automatic encoder was as high as 99.7% |
CNN: Convolutional neural network; DL: Deep learning; MF: Multimodal fusion; MDD: Major depression disorder; RNN: Recursive neural network; AES: Dynamic electrode set.