Table 4.
Spatial EEG features and machine learning applications in the diagnosis of depression.
| Researchers | Aim | Material and methods | Results |
|---|---|---|---|
| Flor-Henry et al. (2004) | Investigate the source of traditional EEG signals in men with depression | EEG tomography (LORETA) in drug-free patients | Hemispheric asymmetry has been observed, which may be characteristic of depression |
| Bachmann et al. (2017) | The objective of this study was to identify a simple and effective method for detecting depression based on the analysis of short, single-channel EEG signals. | The study involved 34 participants, including 17 diagnosed with depression and 17 healthy controls. EEG was recorded using 18 channels with a common Cz reference. | SASI values were significantly higher in the depression group compared to controls |
| Yang et al. (2023) | Evaluation of the influence of selected brain areas and combinations of regions on the effectiveness of MDD detection based on EEG | EEG analysis at rest (eyes closed/open) | Higher LZC and lower PSD in MDD; temporal region achieves 87.4% accuracy, frontal+temporal+central combination |
| Jaworska et al. (2017) | To assess whether the variability of the traditional EEG signal (MSE) at different time scales before antidepressant treatment can predict its effectiveness in people with MDD. | 36 patients with MDD (untreated) and 36 healthy individuals. Resting EEG (eyes open/closed) was recorded before treatment. | Responders had lower MSE on small scales and higher MSE on large scales (especially frontocentral). These patterns did not occur in non-responders or the control group. |
| Korb et al. (2008) | Evaluate differences in cortical activity between people with MDD and healthy individuals. | Traditional EEG (36 channels) from 74 patients with MDD and a control group | MDD patients had a higher current density in the delta-beta2 bands in the anterior ACC and prefrontal cortex. |
| Li et al. (2023) | Investigate whether different levels of depressive states in healthy individuals are associated with different neuronal activity during the perception of emotional stimuli. | Healthy participants were divided into groups with low, medium and high levels of depression. ERPs and ERSPs were recorded during a visual perception task of emotional stimulation. | Individuals with high levels of depression showed a reduced P300 amplitude and differences in fast/slow neural responses in the frontal and parietal lobes. |
| Mathersul et al. (2008) | Investigation of the relationship between depression/anxiety and lateralisation of EEG activity in the frontal and parietotemporal regions. | Study on 428 people with varying levels of negative mood; EEG measurement (alpha waves) and lateralisation analysis were used. | In people with anxiety, right-sided frontal lateralisation was found, in depressive people - symmetrical frontal activity and increased right parietal-temporal activity. |
| Ahmadlou et al. (2012) | Investigation of the complexity of frontal EEG signals in MDD patients using non-linear methods (HFD, KFD) | Traditional EEG divided into 5 sub-bands of brainwave frequencies; KFD and HFD were calculated, statistically compared (ANOVA), and then used in the EPNN classifier | HFD revealed greater complexity in the frontal regions of the brain of MDD patients, especially in the beta and gamma bands. HFD beta differentiated MDD from healthy subjects particularly well. |
| Cai et al. (2016) | To enhance the accuracy of detecting mild depression using EEG by applying differential evolution for feature optimization and k-nearest neighbors for classification. | EEG data from 10 individuals with mild depression and 10 healthy controls were analyzed. Differential evolution was used to optimize the extracted EEG features | Combining differential evolution with k-NN classification enhances the detection of mild depression from EEG data. |
| Li et al. (2019) | To develop an accurate and portable diagnostic method for depression using a three-electrode EEG setup and compare the performance of various classification algorithms. | EEG data were collected from 178 participants using three scalp electrodes placed at Fp1, Fp2, and Fpz—regions closely related to emotion and unobstructed by hair. The algorithms used for classification included k-NN, SVM, ANN, and DBN | The Deep Belief Network (DBN) achieved the highest accuracy (78.24%) when combined with absolute beta wave power. |
| Leuchter et al. (2009a) | Assessment of the usefulness of the ATR index from QEEG in predicting response to various antidepressants in patients with MDD | 375 patients with MDD; QEEG before and after a week of escitalopram (10 mg), then randomized to: escitalopram, bupropion or a combination of the two | High ATR predicted the effectiveness of escitalopram (68% vs. 28%); low ATR – greater effectiveness of bupropion after changing treatment (53% vs. 28%) |
| Bruder et al. (2008) | To investigate whether the resting power and asymmetry of EEG alpha waves differ between depressed patients who respond and do not respond to SSRI treatment and whether it changes after treatment. | 18 patients with depression and 18 matched healthy individuals | Responders had greater alpha power, especially in the occipital regions. Alpha asymmetry (greater power on the right) was observed in responders, unlike in non-responders |